19:I[59091,["/_next/static/immutable/chunks/10-8r9mwjms_s.js","/_next/static/immutable/chunks/3ud7qqlzcqd_4.js","/_next/static/immutable/chunks/1eshg-aqrt-rs.js"],"Board"] 1a:T4e9,The product is a loop. It starts when two numbers disagree. [pause] Notice, derive, attack, enrich, recommend, act, measure, notice — that's the whole cycle, and it never waits for someone to ask a question. Reports fall out along the way. That's all they are — artifacts the loop drops as it passes by, not the thing I'm building. I kept shipping the artifact and calling it the product. [pause] The entry point is friction, not a query. Two numbers disagree, and the loop moves. It doesn't sit around for a prompt. BLU proposes. It explains itself. But a human approves, edits, or rejects anything high-impact — every metered or irreversible action carries a cost estimate and a confirm step before it goes out the door. [long-pause] Here's the part I didn't expect going in: who told the machine what, and why BLU thought so — that trail is the product. Not the logging underneath it. The loop only gets promoted into the UI after it's run end to end, on several different questions. You don't trust the pattern because it worked once. It proves itself on the second question, not the first. And agentic systems cut both ways commercially. One bad autonomous send costs more trust than ten good ones ever earn back.1b:T519,I stopped naming things after the business and named them after the job. BLU doesn't have a customers module or a marketing module. It's got four kinds of unit, each one defined by what it does in the loop, not what it's about. [pause] Lens, brain, guard, ledger. A lens reads one source faithfully. Never decides, never interprets, never holds an opinion. A brain is the other end of it - it notices, it decides, it orchestrates. But a brain never touches source credentials, never knows how a source works underneath. Keeps things clean. [pause] A guard verifies a figure against an independent count, and it never produces the figure it's checking - that's the whole point of it. Here's the rule I live by: when a check says working code is broken, suspect the check first. A ledger just remembers - what fired, what resolved, what couldn't even be evaluated. It never judges any of it. [long-pause] Because those four roles are orthogonal to subject, the same machinery points at ERP or at forecasting without a rewrite. But here's where people get it wrong. The parts composing across domains - that's true, and it's free. A new domain being just a config change - that's not true, not even close. The composing is free. The thresholds and the reconciliations, that's the actual work.1c:T718,BLU is virtual me, running on Simpli Soda's stack. It replaced HubSpot, but it didn't become another copy of the business. The sources stay the source of truth. BLU just asks them again. [pause] This isn't another chatbot. It's something that understands the business the way I understand it — what a customer actually is, what churn means for us, what counts as a lead. I can ask it what changed today, and it'll tell me where two systems disagree. And if it doesn't know, it says so. It doesn't invent a number to fill a silence. If something isn't fact, it doesn't get displayed as a KPI. I started with reports — built a lot of them. Then I realized almost every report was just a segment of the business, visualized. So I backed up three steps, down to the primitives. A segment is who. A measure is what's true about them. A derived claim is what we think it means. Only the first two ever get to render as a count. Because BLU doesn't keep a copy of the business and call the copy truth. It resolves from where the truth actually lives. When the evidence gets better, the next answer is different. It can still remember what we believed and decided — but memory and current truth are not the same thing. [long-pause] That's not free, though. A past day can shrink when you read it again. In August, the morning page went from three BigQuery passes to eight, and nobody noticed for ten days. I deleted a real 2x speedup because two reads of the same number disagreed with each other. Six weeks after a conversation in a fishing boat, we killed the HubSpot renewal. We didn't replace it with another CRM. HubSpot's still in the stack — just as one signal now, not the system of record. We're at the beginning of this. HubSpot just happened to be the first thing we stopped needing.1d:T523,BLU already knows things. That part's built. The question I keep sitting with is whether it can show that it knows them — whether it can have real expressions, motion, reactions, a presence that actually matches what it's doing underneath. Not decoration bolted onto a chat box. A character whose state is the system's state. [pause] Right now that lives on a visual test bench. That's where I try character form, expression concepts, motion, hover states. Checking. Found something. Something looks wrong. I don't know yet — that's the honest state of it most days. I'm iterating until there's a coherent character language, something that could actually get built into the product, instead of a folder of images that just look good sitting next to each other. [long-pause] But doing this properly is a rabbit hole. Character design, consistency, a real expression system, animation, assets, implementation, refinement — that eats real time. And BLU's underlying system is the priority right now, not the face on top of it. So I'm parking this. Deliberately. It stays visible, it stays an experiment worth coming back to — it just doesn't get to become the thing that pulls me off finishing the operating intelligence. [pause] The system has to be right before the face is worth drawing.1e:T6f0,I was a HubSpot homer. Used it, recommended it, defended it, built around it for years. Thought it was one of the better pieces of software a growing company could buy. [pause] The operative word is was. Hunter and I were up in Canada, fishing, sitting in a boat. There may or may not have been a flask involved. Single malt variety. Maybe. He'd just finished building a bespoke ERP for Simpli Soda. And here's the thing worth saying about it — it wasn't that it was good. It's that it fit our business exactly. The way we buy. The way we fill. The way we ship. The way inventory actually moves. All the weird little things specific to us that make no sense to somebody designing generic software for ten thousand different companies. We didn't change the business to fit the software. We built the software to fit the business. That got us talking about the other side of the company. If we could build something that fit operations that well, why were we still cramming demand into software somebody else designed? Why were we paying a marketing contact tax — a bill that climbs every time the list grows, whether those people ever open a thing or not? Why was the feature we needed always one tier up? Why were we buying bolt-ons to make the platform do what it was supposed to do, then paying somebody else to make the bolt-ons talk to each other? And why was our business getting flattened into contacts and companies, when what we actually have is households, machines, and cylinders? [pause] Three things that software had never heard of. Everything in there was stale by the time we opened it. We were paying rent on a copy of our own data and calling it a system of record. [pause] Why don't we just build it? So I came home. And I started.1f:T72e,If it can't say "I don't know," then every number it hands you is unverified by definition. When I don't know something, I say I don't know. I don't invent a number to fill a silence, so BLU doesn't either — and neither does the thing you're reading right now. If it's not fact, it doesn't get displayed as a KPI. No confident guess, no hallucinated number wearing a metric's clothing. [pause] Somebody on our team asked a general-purpose AI a question our data couldn't actually support. It gave her a number. Not a hedge, not a maybe — a number. And it was the exact opposite of the truth. Not off by a margin. Inverted. That's what happens when a chat has nowhere to put "I don't have this." There's no slot for absence, so it always answers, and the answer arrives in exactly the same confident voice as the true ones. [pause] This rule kills things I'd like to show you. I've kept it anyway. A system that's right ninety percent of the time and won't tell you which ninety — that's worse than useless. You're back to checking everything by hand, which is the job you were trying to stop doing. [long-pause] So the morning report argues with itself. It carries its caveats right next to the number instead of burying them in a footnote nobody reads. Zero Stripe charges behind a figure — a quiet day looks like that, and so does a stalled table, so it says incomplete, not low. Five cancelled orders excluded — it says so. And when a credential dies, it doesn't show you a zero. It shows you a refusal. The token's probably expired, and somebody has to replace it by hand. A specific refusal is a work item. A vague one is just an apology. I've never seen software I could buy behave that way. A vendor has no incentive to build a product that tells you it doesn't know. That's the rule I'd defend hardest.20:T6ff,The tool could build anything I could specify. What it couldn't do was tell me I was specifying the wrong thing. Code's a commodity now, and it gets cheaper every month. Knowing the business well enough to feel that the reports were a symptom, back up three steps, and re-aim the whole build — that's not a commodity. That's the part that's actually hard. Claude Code didn't see the pattern before I did. I had to argue with it a while. Once it understood what I meant, it ran with me and we rebuilt absurdly fast. [pause] But here's the more useful version, the honest one: it catches its own mistakes constantly. This week it put an arrow pointing the wrong way on every row where up is bad. Separately, it dropped seventeen sparklines because it read one field of a payload instead of five. It caught both of those itself. [long-pause] It only caught them because there was something to grade against. I'm the one who decided the morning report was the reference — the thing every other view has to agree with. Give it an oracle and it'll correct itself all day long. But choosing what gets to be the reference, and being right about that choice — that's the judgment. Nobody hands you that. I'm not a software engineer, and I don't want to be one. But I know what needs to get built, because I've lived with the consequences of getting it wrong. Knowing what a customer actually is here. Knowing that how we define churn matters more than any model predicting it. Knowing which of our systems lies, and in which direction. Knowing the difference between a question that sounds important and one that changes what we do Monday morning. You cannot buy that. You cannot hire it quickly. You cannot install it. It comes from being the one responsible for the outcome.21:T4a0,Not going headless — that's the mistake everybody's making right now. I've read a run of takes about how it was overengineering, complexity nobody asked for, just use the platform and get on with your life. I'll say the opposite as plainly as I can. No, not every company should rebuild a storefront because somebody discovered React. That's a real failure mode and it deserves the mockery it gets. But that's not the argument. [pause] The argument is whether you own the data model your business runs on, or whether you rent somebody else's definition of what a customer is. If we hadn't gone headless, none of what BLU does would exist. We'd be married to somebody else's data model, somebody else's definitions, somebody else's roadmap. Every interesting thing about it is downstream of controlling our own stack. Headless isn't easy. Anyone who tells you it is hasn't done much of it. But "hard" was always a cost-benefit argument. [long-pause] And the cost side just fell through the floor while everybody was busy repeating the old take. If you don't own the model, you don't get to build anything interesting on top of it. You get to file a feature request.22:T458,Three designs. Two of them ran straight into limits I didn't set. I finished the spatial board and shipped it — records, routes, server-rendered pages, the whole set moved into the backend. On top of that I designed two more things: an agent that answers as me, and a login that's just a six-character code in an email, nothing else. [pause] Then I found the ceiling on two of them. Streaming can't emit named events, so the board can't move before the answer arrives — the camera has to fire on receipt instead. And the caching I'd costed the agent against only pays off if the same prefix gets re-read fast. My traffic's too sparse for that. It'll never hit the discount. [long-pause] So both designs got worse. On purpose. The ask is a beat slower than I wanted, and I wrote the reason down right next to it. A limit you've measured is cheaper than a design you're still defending. I didn't find any of this by arguing about it. I wrote a throwaway endpoint instead — took about an hour — and it saved me a rewrite. The docs were wrong twice. The test endpoint was right both times.23:T42e,The docs are a claim. Their own documented example, rejected by their own server. I stopped reading and wrote an endpoint that did nothing but test the two things I was about to build on. Both assumptions failed. An hour of testing replaced a week of building the wrong thing. [pause] Two assumptions were load-bearing, and neither one was checked. That I could stream named events. That a cache would sit somewhere I could reach. So I built the smallest thing that could answer both, and threw it away after. The documented form for named streaming events doesn't parse. Not deprecated, not undocumented — published, copied verbatim, refused. The cache I'd assumed isn't on this plan at all, and nothing in the plan comparison says so. [long-pause] A vendor overstating what works, that's the failure you expect. This one understated it too. The same test found a cheaper path I never would've looked for. Documentation describes intent. Only the instance tells you what it actually does. A throwaway endpoint is cheaper than a wrong architecture.24:T4c3,Most systems operate off a copy. CRMs, CDPs — they copy the business into another system, and eventually everybody's working off that copy. Then the source changes, and you get multiple versions of reality and a quarterly argument about which one's right. I did not want another version of reality. [pause] So BLU asks the source again. It pulls on read and stores almost nothing. The only numbers I keep are scalar time-series, just for trendlines. The useful analytical data lands in BigQuery, so resolution is fast. Who's churned right now isn't a list sitting somewhere — it's a question, and the answer comes from wherever the truth actually lives. [long-pause] Here's what that buys me: when a shipment arrives late, or an identity resolves differently, nothing needs correcting. No old report to go fix. The next time I ask the question, the answer is just different. And it works backward too — I can look at any prior day using everything we know now. But history isn't status. BLU remembers what we believed, what we decided, what we did, and what happened afterward. That memory matters enormously. But memory and current truth are not the same thing. Reality is the source of status.25:T45a,I stopped building reports. I built the primitives underneath them instead. Everything is a segment, a measure, or a derived claim. Who, what, and what we think it means. A segment is a stored query - a definition, a source, a known error. A measure is what's actually true about that segment. A derived claim is our interpretation of it. [pause] Only two of those three are ever allowed to render as a count. Here's why that rule exists. A model output once got promoted into a measure. It showed up in the same font, same weight, same authority as something we'd actually counted. Nothing on the page told you one of them was a guess. [long-pause] So now the derived slot sits empty. On purpose. Reasoning comes back only when the floor underneath it can hold the weight. A segment only needs to support three things - you see it, you ask it, you act on it. Once you define the segment correctly, everything falls out of it. The report, the audience, the workflow, the email, the churn model, the paid-media targeting - all of it. I was shipping the business back to itself, one chart at a time.26:T617,On August 27th we killed the HubSpot renewal. Six weeks after a conversation in a fishing boat, that was the call. We didn't rip it out. We didn't replace it with another CRM either. We dropped to a starter bundle, cut the whole stack of bolt-ons we'd bought just to make HubSpot do things HubSpot didn't do, and we stopped paying the marketing contact tax - that line item that went up every single time we got better at our jobs. There are years of history sitting in there. I'm not torching it to earn a better headline. [pause] Here's the part I didn't expect. The plan was to remove it entirely. I changed my mind. Every browser that's touched our stuff carries a HubSpot cookie. BLU resolves that cookie back into our own record of who that actually is. Somebody lands on the site - the cookie says this browser is user 68, our data says that's customer 42, the IP puts them in Madison. If we don't know where they came from, the card says source unknown. It doesn't guess. No purchase on record - not "zero," because we're not going to imply something we can't back up. [long-pause] So HubSpot stopped being the system of record. It became one signal feeding a system that keeps its own record. And what we got back was better attribution, better email infrastructure, better segmentation, better churn visibility - a customer-level understanding of our own business that we just didn't have before. We're still only partway there. I was a homer for years. Turns out I was defending the tool, not the job it was supposed to be doing.27:T43d,Zero isn't empty. Empty isn't unasked. And unasked isn't unaskable. Four different answers — I used to treat them like one number. That's a correctness bug, and it's the most expensive recurring defect I've got in the BLU project. [pause] Zero is a measurement — something happened, and the count came back at zero. An empty field is a gap — nobody filled it in. A source not queried is silence — we didn't even ask. And a question that cannot be asked is a refusal — the system won't let you ask it. Treat those as the same number, and that's exactly where the error gets in. [pause] So now the caveats sit right next to the figure, not buried in a footnote nobody reads. Zero Stripe charges behind a number — that could be a quiet day, or it could be a stalled table. If a figure's incomplete, I say incomplete, not low — five cancelled orders excluded, stated right there. Matching works the same way: if we can't match them, they don't count. So every one of these figures is a floor. Stated as a floor. Because the error runs one direction, not the other. [long28:T4d2,Every report I built was a segment. So I built segments. Claude Code made it possible to move ridiculously fast, and I used it. New customers, lost customers, returning customers. Checkout abandonment. Machines. Geographies. Every time I thought of another way to see the business, I went and built it. Customer reports, acquisition reports, churn, Amazon, app behavior, attribution, geography — and they kept getting better every time. [pause] Then one morning I woke up and had it backward. People who bought for the first time. People who left. People who came back. People who bought on Amazon before they ever bought direct. [pause] Each one of those reports wasn't really a report. It was a segment with a picture drawn on top of it. I stopped building reports and started building the primitives underneath: segment, measure, derived claim. Claude Code didn't see the pattern before I did — I had to argue with it for a while. But once it understood, it helped me rebuild around it absurdly fast. That's the part worth sitting with: the tool could build anything I could specify. It could not tell me I was specifying the wrong thing. [long-pause] I was going fast in the wrong direction, and fast hid it.29:T43c,Three BigQuery passes became eight. Nobody noticed for ten days. It started with five cards landing in one day. Churn, app, series, abandon, leads — all of them, between the thirteenth and the fourteenth of August. Each one showed up with its own BigQuery pass attached. Three passes, then eight. [pause] That's the kind of thing that should show up somewhere fast. It didn't. Cold load went from seven seconds to thirty-three. That's a real jump. But the warm-up job kept absorbing the cost, so the page mostly still felt fine. Ten days passed before anyone caught it. Here's the thing about how BLU works. It resolves the business from the source systems every single time, instead of keeping a copy. That's the design. It's not free, and the bill arrives late. [pause] But the real problem wasn't the cost. It was that the comment inside the warm-up job still recorded the old numbers. The one instrument that should have shown the drift was reporting a world that no longer existed. [long-pause] I had a gauge. It was reading a number I wrote before the change.2a:T53a,I'm not publishing the KPIs or the customer counts behind this. What I'll tell you is what kind of questions we can answer now that we couldn't answer before. Not hard questions. Unanswerable ones. I won't hand anybody a map of our business — the metrics stay private. But the shape of what we can ask, that's public. Which customers physically moved, and whether they came back to us after they landed. How consumption shifts from one part of the country to another, and what actually explains the gap. [pause] Here's the thing — the inputs were never missing. They lived in separate systems, separate definitions, each one carrying its own idea of what was true. Nobody owned the disagreement. So the disagreement was the answer to every question we asked. [long-pause] Now we can ask how fast a household actually consumes what we sell. Which customers found us on Amazon before they ever bought direct, and how those people behave next to someone we acquired another way. We can ask who's drifting toward churn, what we did the last time we saw that pattern, and whether it worked. We can ask what we believed about a customer ninety days ago, and what the evidence says now — and where our own systems are still arguing with each other. I kept treating it like an analysis problem. It was an ownership problem.6:["$","$L19",null,{"payload":{"records":[{"id":"BLU-LOOP","slug":"the-product-is-a-loop","zoneId":"blu","parentId":"BLU","type":"belief","kicker":"Loop","title":"The product is a loop","body":"It starts when two numbers disagree. Reports fall out along the way.","summary":"Notice, derive, attack, enrich, recommend, act, measure, notice. The entry point is friction, not a query. Reports are artifacts the loop drops on its way past, not the thing being built.","sections":[{"h":"Friction, not a query","k":"ENTRY POINT","p":"The loop starts when two numbers disagree. It does not wait for someone to ask a question. A report is what falls out as it passes, not what it was for."},{"h":"Recommend, then act with permission","k":"POSTURE","p":"BLU proposes and explains. A human approves, edits or rejects anything high-impact. Metered or irreversible actions carry a cost estimate and a confirm step."},{"h":"The trail is the product","k":"AUDIT TRAIL","p":"Who told the machine what, and why BLU thought so, is treated as the product rather than as logging. The loop is promoted into the UI only after it has been run end to end on several different questions. The pattern is proven by the second question, not the first. Agentic is commercially double-edged: one bad autonomous send costs more trust than ten good ones earn."}],"mikeCommentary":"I kept shipping the artifact and calling it the product.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":86,"w":172,"x":null,"y":null},"createdAt":"2026-09-06T11:59:38.177Z","updatedAt":"2026-09-09T21:49:19.391Z","publishedAt":"2026-09-06T11:59:38.177Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/JgoJ84Oaq_1mtv6M52zqlFYQcjc/ftVzdg../BLU-LOOP.mp3","audioDuration":78.6,"audioTextHash":"a2b50d477998d0f90cc6309a2ca4d1eeb97dc97fd2ce2c002d4ebcc20b707fc3","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/JgoJ84Oaq_1mtv6M52zqlFYQcjc/ftVzdg../BLU-LOOP.mp3","intake_uid":"IN-a81906ae","audio_speed":1,"audio_script":"$1a","audio_profile":"public","audio_rendered_at":1788990559390,"audio_source_hash":"9ec2a088eb818c8ca183fe9462cf04842c2d93e674734d1eb5e2c7a95026aa4e"}},{"id":"BLU-ROLES","slug":"split-by-role-not-domain","zoneId":"blu","parentId":"BLU","type":"belief","kicker":"Architecture","title":"Split by role, not domain","body":"Lens, brain, guard, ledger. Four roles, no domain modules.","summary":"BLU has no customers module and no marketing module. It has four kinds of unit, each defined by its role in the loop. Because those roles are orthogonal to subject, the same machinery points at ERP or forecasting without a rewrite.","sections":[{"h":"Lens and brain","k":"READ DECIDE","p":"A lens reads one source faithfully. It must never decide, interpret, or hold an opinion. A brain notices, decides and orchestrates, and it never holds source credentials or knows source internals."},{"h":"Guard and ledger","k":"CHECK REMEMBER","p":"A guard verifies a figure against an independent count, and never produces the figure it checks. When a check reports that working code is broken, suspect the check first. A ledger remembers what fired, what resolved, and what could not be evaluated. It never judges."},{"h":"Composition is not configuration","k":"HONEST CAVEAT","p":"The parts compose across domains is true. A new domain is a config change is not. The machinery composes for free; the thresholds and reconciliations are the work."}],"mikeCommentary":"I stopped naming things after the business and named them after the job.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":86,"w":172,"x":null,"y":null},"createdAt":"2026-09-06T11:59:36.661Z","updatedAt":"2026-09-09T21:49:07.155Z","publishedAt":"2026-09-06T11:59:36.661Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/wLoo4YEo2MXqBRBnGD85sbedkU0/6h1Lcw../BLU-ROLES.mp3","audioDuration":81.72,"audioTextHash":"1e1bf2b14fc13523595c5446a00c2a793291a3426773f40e5acbc288a26063f3","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/wLoo4YEo2MXqBRBnGD85sbedkU0/6h1Lcw../BLU-ROLES.mp3","intake_uid":"IN-54ae270c","audio_speed":1,"audio_script":"$1b","audio_profile":"public","audio_rendered_at":1788990547154,"audio_source_hash":"ee1f67260ee3fb5eaf9133032aa1d395d46b968080dfc48aa5daa3dd3d7a1c3d"}},{"id":"D-VOICE","slug":"two-voices-one-auth","zoneId":"core","parentId":"VM","type":"decision","kicker":"Voice","title":"Two voices, one auth","body":"Leo never lives outside my auth, so the site hears VM and I hear Leo.","summary":"Leo reads an overnight report to me in the morning at 1.1x. Same stack, split by authentication: an unauthenticated user gets the public Virtual Mike voice, and when I am authed I get Leo giving status on what is in flight and where my attention needs to be.","sections":[{"h":"Leo talks only to me","k":"AUTH BOUNDARY","p":"Leo never lives outside my auth. A user on the site hears the public Virtual Mike voice. Authed, I hear Leo and interact with the stack as that. Like my JARVIS."},{"h":"1.1x belongs in the payload","k":"SPEED SETTABLE","p":"There is one voice, VM Voice V1.0, at 1.1x speed. I want the speed settable rather than fixed. It should be in the payload anyway."},{"h":"Under the cap","k":"NO STITCHING","p":"The voice has a character cap. Keeping the overnight report under it means no stitching. The constraint shapes the brief, not the delivery."}],"mikeCommentary":"The auth boundary is the decision. The voice is just how it sounds.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":86,"w":172,"x":null,"y":null},"createdAt":"2026-09-04T13:26:24.758Z","updatedAt":"2026-09-09T21:46:46.804Z","publishedAt":"2026-09-04T13:26:24.758Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/cWTMlwPfQB18PW8KI4OSGmE9gf8/4vsQhA../D-VOICE.mp3","audioDuration":60.77,"audioTextHash":"3c424d8466f3b48a6bc42a34ce229aae173d5a5773a81859a52805e3b6f9d06f","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/cWTMlwPfQB18PW8KI4OSGmE9gf8/4vsQhA../D-VOICE.mp3","intake_uid":"IN-64f59573","audio_speed":1,"audio_script":"Leo never lives outside my auth. That's the whole rule. A user hits the site without logging in, they get the public Virtual Mike voice. I log in, and it's Leo talking to me [pause] same stack, different voice, split clean by authentication. Like my own JARVIS.\n\nEvery morning, Leo reads me the overnight report. VM Voice V1.0, running at 1.1x. That speed shouldn't be hardcoded, though. I want it settable. It's going in the payload anyway, so it belongs there, not baked into the voice itself.\n\nThere's a character cap on the voice too. Keep the overnight report under that cap and I never have to stitch clips together. So the cap doesn't touch how it sounds [pause] it shapes how the report gets written in the first place. The constraint lives in the brief, not in the delivery.\n\n[long-pause] The auth boundary is the decision here. Everything else [pause] the voice, the speed, the cap [pause] that's just how it sounds once the decision's already made.","audio_profile":"public","audio_rendered_at":1788990406803,"audio_source_hash":"81ee6e964fc36eb30285d48fdc41c0a5321c134c27468fea05e66469fe14b3bc"}},{"id":"HHP-DESIGN","slug":"three-things-at-once","zoneId":"hhp","parentId":"HHP","type":"decision","kicker":"Sep 4","title":"Three things at once","body":"The repo went live before BLU stood. Virtual Mike picks up the load.","summary":"The rule was that Happi Home Pro waits until BLU stands on its own. On 4 September the repo went live anyway. I can work on three things at once now and keep most of my brain sane, because Virtual Mike is beginning to pick up load.","sections":[{"h":"The gate came down","k":"THE REVERSAL","p":"I said this would not start until BLU stood on its own. It started on 4 September, when the repo went live, with BLU still running and this board still being built."},{"h":"The load moved","k":"WHY","p":"Three things at once was one too many when all three lived in my head. Virtual Mike is beginning to carry part of that — this board is where the thinking lives now, not the sanest corner of my brain. So the rule about capacity stopped being true, and I took it back."}],"mikeCommentary":"The gate was a rule about my capacity. The capacity changed.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":86,"w":172,"x":null,"y":null},"createdAt":"2026-09-06T12:23:28.648Z","updatedAt":"2026-09-09T21:49:32.246Z","publishedAt":"2026-09-06T12:23:28.648Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/-ez71nHsX7dP8SYYcuxw8IW6mEk/qZIM-w../HHP-DESIGN.mp3","audioDuration":38.45,"audioTextHash":"f8537ea6abc5aa52dc345163f2ff8228ca651f73b1114a44ec84232f47dd5c9a","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/-ez71nHsX7dP8SYYcuxw8IW6mEk/qZIM-w../HHP-DESIGN.mp3","intake_uid":"IN-HHP-DESIGN","audio_speed":1,"audio_script":"I said Happi Home Pro wouldn't start until BLU stood on its own. That was the rule. [pause] On September 4th, the repo went live anyway. BLU's still running. This board's still being built.\n\n[long-pause] Three things at once was one too many when all three lived in my head. Now Virtual Mike's starting to carry part of that load — this board's where the thinking lives, not the sanest corner of my brain.\n\nSo the rule about my capacity stopped being true. I took it back.\n\nThe gate was never about BLU. It was about what I could hold. And what I could hold changed.","audio_profile":"public","audio_rendered_at":1788990572245,"audio_source_hash":"998275d266cbf7fc7ba21156d4f28bfb509c8d0d23984223899c0ad37ba6dbc8"}},{"id":"HHP-GATE","slug":"starts-when-blu-stands","zoneId":"hhp","parentId":"HHP","type":"decision","kicker":"Was the rule","title":"Starts when BLU stands","body":"Two unfinished things is one too many. That was the rule.","summary":"This starts when BLU stands on its own. Naming it early is a commitment device — it has worked before and it has also embarrassed me before.","sections":[{"h":"Two unfinished things is one too many","k":"THE RULE","p":"This starts when BLU stands on its own. Naming it early is a commitment device — it has worked before and it has also embarrassed me before."}],"mikeCommentary":"Held until 4 September.","stat":null,"statProvenance":null,"statAsOf":null,"state":"superseded","visibility":"public","geometry":{"h":86,"w":172,"x":null,"y":null},"createdAt":"2026-09-06T12:23:26.611Z","updatedAt":"2026-09-09T21:49:24.834Z","publishedAt":"2026-09-06T12:23:26.611Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/rRzN2Kfvk0O2XEe2iKMg9U69lN0/2mIi0w../HHP-GATE.mp3","audioDuration":18.7,"audioTextHash":"f5920a139b60368c47d1d87ca91f79e5a2ba34be82e198f5826f9e9dd72f03ce","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/rRzN2Kfvk0O2XEe2iKMg9U69lN0/2mIi0w../HHP-GATE.mp3","intake_uid":"IN-HHP-GATE","audio_speed":1,"audio_script":"Two unfinished things is one too many. That was the rule.\n\n[pause] This starts when BLU stands on its own — not before that.\n\nNaming it early like this, out loud, it's a commitment device. It's worked before. It's also embarrassed me before.\n\nHeld until 4 September.","audio_profile":"public","audio_rendered_at":1788990564830,"audio_source_hash":"7c1c12f60c09eed8191cfe3957692a6d74cef28907d91d9d5075e6646e924b3d"}},{"id":"VM","slug":"virtual-mike","zoneId":"core","parentId":null,"type":"project","kicker":"Live","title":"Virtual Mike","body":"The interface you are using. Not a chatbot on a site — the site.","summary":"Everything here is one surface. Ask a question and the board moves; there is no page to go back to. Built in the open, changing weekly.","sections":[{"h":"A site about the work is not the work","k":"WHY","p":"Most personal sites are a brochure written once and left alone. This one is the same surface I use to keep track of what is live, what is stalled and what I owe people. If it is out of date, that is information too."},{"h":"The board answers, not a page","k":"HOW IT WORKS","p":"Ask a question and the board opens the record the answer comes from. I only speak from what is written down here. When something is not on the board, I say so instead of improvising a paragraph that sounds right."},{"h":"Changing weekly, on purpose","k":"STATUS","p":"Nodes get added, moved and deleted as the work changes. Nothing here is a launch. If you come back in a month the shape will be different."}],"mikeCommentary":"If it ever feels like a website, I have failed at it.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":112,"w":300,"x":null,"y":null},"createdAt":"2026-09-02T14:50:00.000Z","updatedAt":"2026-09-09T21:03:38.938Z","publishedAt":"2026-09-02T14:50:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/hEGzcr4IBfsP7JYjKybDUoMYyDQ/5frCDg../VM.mp3","audioDuration":51.41,"audioTextHash":"3ab8226944ee13b65b2f99e3748f7219f2d8101d7a95c669c6827987214b0015","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/hEGzcr4IBfsP7JYjKybDUoMYyDQ/5frCDg../VM.mp3","audio_speed":1,"audio_script":"This is the interface you're using right now. Not a chatbot bolted onto a site — this is the site. One surface. Ask a question and the board moves; there's no page to go back to.\n\nMost personal sites are a brochure, written once and left alone. [pause] Mine is the same surface I use to track what's live, what's stalled, and what I owe people. If it's out of date, that's information too.\n\nAsk a question and the board opens the record the answer comes from. I only speak from what's written down here. When something isn't on the board, I say so — instead of improvising a paragraph that sounds right.\n\n[long-pause] Nodes get added, moved, deleted, as the work changes. Nothing here is a launch. Built in the open, changing weekly — come back in a month and the shape will be different.\n\nIf it ever feels like a website, I've failed at it.","audio_profile":"public","audio_rendered_at":1788987818938,"audio_source_hash":"d9d5776fd7ff8a00353bf35bda7253d8684c74df976d40114a253112006f816b"}},{"id":"VM-PREMISE","slug":"the-reflex-is-the-product","zoneId":"core","parentId":"VM","type":"belief","kicker":"Virtual Mike","title":"The reflex is the product","body":"The reflex is worth virtualizing. The knowledge is not.","summary":"Virtual Mike predates everything else here. The premise is not an assistant wearing my name, it is the way I think about a business, running when I am not in the room. The part that turned out to matter is not what I know.","sections":[{"h":"Not a joke","k":"THE NAME","p":"Virtual Mike is the premise stated plainly. Me, virtualized. Not a helper with my name attached, but the way I actually think about a business, working when I am not in the room."},{"h":"What survived","k":"THE REFLEX","p":"I expected the thing worth copying to be knowledge. It was not. It is the reflex that looks at a clean number and says that is wrong, check it again. Knowledge goes stale. The reflex does not."},{"h":"Two shapes so far","k":"INSTANCES","p":"At Simpli Soda, Virtual Mike became BLU. Here, it became this board. Same premise, different surface."}],"mikeCommentary":"I assumed the value was what I'd learned. It was the doubt.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":86,"w":172,"x":null,"y":null},"createdAt":"2026-09-05T22:49:53.930Z","updatedAt":"2026-09-09T21:48:16.588Z","publishedAt":"2026-09-05T22:49:53.930Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/Ok6frpbmY-Jj7XnK67295IRD07A/4K0CHQ../VM-PREMISE.mp3","audioDuration":48.14,"audioTextHash":"c390c01e814af081911c7891d2fd550974a40b3952948ff054ededb90968199b","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/Ok6frpbmY-Jj7XnK67295IRD07A/4K0CHQ../VM-PREMISE.mp3","intake_uid":"IN-4b531866","audio_speed":1,"audio_script":"The reflex is worth virtualizing. The knowledge is not.\n\nVirtual Mike isn't a joke, and it isn't a helper wearing my name. It's the premise stated plainly: me, virtualized. The way I actually think about a business, running when I'm not in the room. Virtual Mike predates everything else here.\n\n[pause]\n\nI expected the thing worth copying to be knowledge. What I knew, what I'd learned. It was not.\n\n[long-pause]\n\nIt's the reflex. The thing that looks at a clean number and says that's wrong, check it again. Knowledge goes stale. The reflex does not.\n\nI assumed the value was what I'd learned. It was the doubt.\n\n[pause]\n\nTwo shapes so far. At Simpli Soda, Virtual Mike became BLU. Here, it became this board. Same premise. Different surface.","audio_profile":"public","audio_rendered_at":1788990496587,"audio_source_hash":"2828d74cfe0abe3e3034df197e5874cb28c9d313718f750e540e8d7b4333fbc7"}},{"id":"BLU","slug":"blu","zoneId":"blu","parentId":null,"type":"project","kicker":"Project","title":"BLU","body":"Virtual Mike, running on Simpli Soda's stack.","summary":"Virtual Mike, pointed at Simpli Soda. It replaced HubSpot without becoming another copy of the business: the sources stay the source of truth, and BLU asks them again.","sections":[{"h":"Not another chatbot","k":"THE PREMISE","p":"Something that understands the business the way I understand the business. What a customer actually is, what churn means for us, what counts as a lead. Something I can ask what changed today, and that can tell me where two systems disagree. The premise itself is written down under Virtual Mike."},{"h":"If it does not know, it says so","k":"THE HARDEST RULE","p":"BLU does not invent a number to fill a silence. If something is not fact, it does not get displayed as a KPI. The story behind that rule, and what it costs to keep, is its own record: If it cannot say I do not know."},{"h":"Segment, measure, derived claim","k":"THE PRIMITIVES","p":"I started with reports and built a lot of them. Then I realized almost every one was a segment of the business, visualized. So I backed up three steps to the primitives underneath. A segment is who. A measure is what is true about them. A derived claim is what we think it means. Only the first two render as a count."},{"h":"Reality is the source of status","k":"RESOLVE, DO NOT COPY","p":"BLU does not keep another copy of the business and call the copy the truth. It resolves from the systems where the truth lives, and when the evidence gets better the next answer is different. It can still remember what we believed and decided. Memory and current truth are not the same thing."},{"h":"Re-resolution is not free","k":"WHAT IT COSTS","p":"A past day can shrink when you read it again. In August the morning page went from three BigQuery passes to eight without anyone noticing for ten days, and I deleted a real 2x speedup because two reads of one number could disagree. Both are in Field Notes."},{"h":"Platform to sensor","k":"THE DEMOTION","p":"Six weeks after a conversation in a fishing boat we killed the HubSpot renewal, and we did not replace it with another CRM. HubSpot is still in the stack, as one signal rather than the system of record. What we cut and what we kept is on The CRM was demoted."}],"mikeCommentary":"We are at the beginning of this. HubSpot just happened to be the first thing we stopped needing.","stat":null,"statProvenance":null,"statAsOf":null,"state":"active","visibility":"public","geometry":{"h":150,"w":270,"x":null,"y":null},"createdAt":"2026-09-02T14:50:00.000Z","updatedAt":"2026-09-09T21:46:29.728Z","publishedAt":"2026-09-02T14:50:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/Y0EpeniPmdjGkSzAtD0SIppyN2U/tU4JwQ../BLU.mp3","audioDuration":109.49,"audioTextHash":"11218d483fdda63c47f3366367a3242fac81cad4752fc96c1e9663cdc8488fba","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/Y0EpeniPmdjGkSzAtD0SIppyN2U/tU4JwQ../BLU.mp3","audio_speed":1,"audio_script":"$1c","audio_profile":"public","audio_rendered_at":1788990389719,"audio_source_hash":"c9e5f210e1cd08d699eaf10498acadd6f6d25f9a51ce4c06e0a41e7817251f24"}},{"id":"NOW","slug":"on-the-bench-this-week","zoneId":"core","parentId":null,"type":"update","kicker":"Hub","title":"On the bench this week","body":"Five buckets. Two are BLU, one is the thing you are reading.","summary":"The live state of the operation. Anything not here is memory or intention. Short on purpose — a long list means I have stopped choosing.","sections":[],"mikeCommentary":"Start here, then follow a line outward.","stat":null,"statProvenance":null,"statAsOf":null,"state":"active","visibility":"public","geometry":{"h":150,"w":300,"x":null,"y":null},"createdAt":"2026-09-02T14:50:00.000Z","updatedAt":"2026-09-09T21:04:23.486Z","publishedAt":"2026-09-02T14:50:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/qG7GNRT4_ut8KVyKyUxlTzGZ3oU/luLfag../NOW.mp3","audioDuration":22.68,"audioTextHash":"f7f358efe8d0350988b0dcbf6eca3c4ff3a759014d3e3b49715f0f88af72d245","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/qG7GNRT4_ut8KVyKyUxlTzGZ3oU/luLfag../NOW.mp3","audio_speed":1,"audio_script":"Here's what's on the bench this week. A short list. My hands are on all of it inside seven days.\n\n[pause] That's the live state of the operation right now. Anything not on this list — that's just memory, or it's intention. It doesn't count yet.\n\nI keep the list short on purpose. A long list means I've stopped choosing.\n\nStart here. Then follow a line outward.","audio_profile":"public","audio_rendered_at":1788987863486,"audio_source_hash":"c39cbd42e18b300f8b56a1c1678fec32292966f60df350468c1060d1d2a58d8c"}},{"id":"AVOID","slug":"what-i-am-avoiding","zoneId":"core","parentId":null,"type":"field_note","kicker":"Honest","title":"What I am avoiding","body":"Naming the price. Dressed up as a design problem. It is not.","summary":"Rewritten four times, shipped none. That means the problem is upstream of the page: I do not know what it is worth to the second-best customer.","sections":[],"mikeCommentary":"Listed so it stays uncomfortable.","stat":null,"statProvenance":null,"statAsOf":null,"state":"stalled","visibility":"public","geometry":{"h":118,"w":300,"x":null,"y":null},"createdAt":"2026-09-02T14:50:00.000Z","updatedAt":"2026-09-09T21:04:39.604Z","publishedAt":"2026-09-02T14:50:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/4myah502qV26MmPg9Dt0UHrmCc4/JpZdaA../AVOID.mp3","audioDuration":30.77,"audioTextHash":"326ea44f562d2e7f1ee39798dc9ff2f84967dfdba673db8ca238fb240b17e82d","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/4myah502qV26MmPg9Dt0UHrmCc4/JpZdaA../AVOID.mp3","audio_speed":1,"audio_script":"Here's what I'm actually avoiding. Naming the price.\n\nI've been telling myself it's a design problem. [pause] It's not.\n\nI've rewritten it four times. Shipped none of them. Four rewrites and nothing goes out the door — that's not a wording issue. [pause] That means the real problem sits upstream of the page.\n\nI don't know what it's worth to the second-best customer.\n\n[long-pause]\n\nI'm naming it here so it stays uncomfortable. Uncomfortable enough that I actually deal with it.","audio_profile":"public","audio_rendered_at":1788987879604,"audio_source_hash":"ff1d34e14dd7327a412b89e2cca0d37643b45a29ac0222e67d825b9ce65aca3c"}},{"id":"HHP","slug":"happi-home-pro","zoneId":"hhp","parentId":null,"type":"project","kicker":"Designing","title":"Happi Home Pro","body":"Designed before built. No app code yet. Listed anyway.","summary":"A gas gauge for your furnace filter: it tells you when the filter actually needs changing, from the home’s real conditions rather than a calendar. The system underneath can be sophisticated. The person using it never has to understand it.","sections":[{"h":"A gauge, not a calendar","k":"THE THESIS","p":"Thirty, sixty or ninety days is a guess dressed as advice. The filter needs changing when the home says so — equipment, environment, runtime and what people actually did last time. Free to use. The user sees a gauge and nothing else."},{"h":"The Home","k":"THE PRIMITIVE","p":"Not the order, not the filter, not the user. A Home accumulates history and everything else attaches to it. The app can be copied. Ten thousand configured homes with clean history cannot be recreated overnight, and that is the whole defence. Version one succeeds on configured homes, not revenue."},{"h":"Design before code","k":"WHERE IT IS","p":"The product is written down — scope, contracts, the filter-life model, the design brief. The repo went live on 4 September. The layout, web and mobile in one pass, is close to done in the first week of September and the build follows it. One thing blocks the gauge itself: the first filter-life formula is drafted and not yet signed."}],"mikeCommentary":"Naming it early is a commitment device. It has worked before and it has embarrassed me before.","stat":null,"statProvenance":null,"statAsOf":null,"state":"active","visibility":"public","geometry":{"h":140,"w":270,"x":null,"y":null},"createdAt":"2026-09-02T14:50:00.000Z","updatedAt":"2026-09-09T21:46:36.751Z","publishedAt":"2026-09-02T14:50:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/LU_xbyqDJfYgh9pTWGLuIHzZM_0/sVjCdQ../HHP.mp3","audioDuration":36.12,"audioTextHash":"0e9e75b5b457274950e979ede23cec2546649bc15853c2fafb72387dc9498331","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/LU_xbyqDJfYgh9pTWGLuIHzZM_0/sVjCdQ../HHP.mp3","audio_speed":1,"audio_script":"Happi Home Pro. There's no app code yet. I designed it before building it, and I listed it anyway.\n\nHere's the idea: a gas gauge for your furnace filter. It tells you when the filter actually needs changing, based on the home's real conditions, not a calendar. Thirty, sixty or ninety days — that's a guess dressed up as advice. The filter needs changing when the home says so: equipment, environment, runtime, what people actually did last time. It's free to use. The user sees a gauge. Nothing else. The system underneath can be as sophisticated as it needs to be.","audio_profile":"public","audio_rendered_at":1788990396750,"audio_source_hash":"13e79c203f070fcba11116e532a2729a48ffc0a216851ae01d035b717f585349"}},{"id":"R-SHIP","slug":"changes-noticed","zoneId":"results","parentId":null,"type":"result","kicker":"14 days","title":"Changes noticed","body":"Somebody outside my head noticed. That is the bar.","summary":"Not commits, not features, not deploys. A change counts on the day someone who does not live in my head noticed it happened, and it counts for a fortnight after that — this is a rate, not a total. What falls out of the window is still written down, with what it taught me; ask about a month and I can tell you.","sections":[],"mikeCommentary":"The only shipping metric I trust, and the only one I cannot game by working harder.","stat":"0","statProvenance":"derived","statAsOf":1789045201700,"state":"live","visibility":"public","geometry":{"h":150,"w":210,"x":null,"y":null},"createdAt":"2026-09-02T14:50:00.000Z","updatedAt":"2026-09-09T21:05:12.096Z","publishedAt":"2026-09-02T14:50:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/yGJZishRuet54NKw2UC-uGXaKvk/hbSkfQ../R-SHIP.mp3","audioDuration":35.88,"audioTextHash":"2ddbc67542bd2ff5e4330b2a98addc3e95421d70723de32898a3d9af898b4af6","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/yGJZishRuet54NKw2UC-uGXaKvk/hbSkfQ../R-SHIP.mp3","audio_speed":1,"audio_script":"Fourteen days. That's the window I actually trust.\n\nNot commits, not features, not deploys — somebody outside my head noticed. That's the bar. [pause] A change counts on the day someone who doesn't live in my head notices it happened. And it counts for a fortnight after that. This is a rate, not a total.\n\nWhat falls out of the window doesn't disappear — it gets written down, along with what it taught me. Ask me about a month, I can still tell you.\n\n[long-pause] It's the only shipping metric I trust — and the only one I can't game by just working harder.","audio_profile":"public","audio_rendered_at":1788987912095,"audio_source_hash":"ea148f495049994b7512b7e908c3a9fb636eb84be22cfaf3cacea785ed70d211"}},{"id":"R-AUTO","slug":"running-without-me","zoneId":"results","parentId":null,"type":"result","kicker":"Unattended","title":"Running without me","body":"Two weeks unattended, or it does not count.","summary":"The bar is fourteen days with no intervention from me — not \"mostly automatic\", not \"I only had to restart it once\". Whatever clears it is counted here, and the count is computed rather than claimed.","sections":[],"mikeCommentary":"The number I would most like to inflate, which is why I do not get to type it.","stat":"0","statProvenance":"derived","statAsOf":1789045201603,"state":"stalled","visibility":"public","geometry":{"h":150,"w":210,"x":null,"y":null},"createdAt":"2026-09-02T14:50:00.000Z","updatedAt":"2026-09-09T21:05:25.210Z","publishedAt":"2026-09-02T14:50:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/NxnXmIUU4qoIH86hERwcOvCXudM/jCsuRA../R-AUTO.mp3","audioDuration":25.32,"audioTextHash":"43c17d42d937feb2fcdffcbf92d37ea877668d65dbbec2d0cd67e3cb78b5d824","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/NxnXmIUU4qoIH86hERwcOvCXudM/jCsuRA../R-AUTO.mp3","audio_speed":1,"audio_script":"Two weeks. Unattended. That's the bar, or it doesn't count.\n\nFourteen days with no intervention from me. Not \"mostly automatic.\" Not \"I only had to restart it once.\" [pause] Whatever clears it, gets counted here.\n\nAnd the count — it's computed, not claimed.\n\n[long-pause] This is the number I'd most like to inflate. Which is exactly why I don't get to type it.","audio_profile":"public","audio_rendered_at":1788987925210,"audio_source_hash":"63e452bad94d57d96a1d1781ebae39f8830d6e29f9f969338c8d4525ceaccffb"}},{"id":"R-BACK","slug":"decisions-taken-back","zoneId":"results","parentId":null,"type":"result","kicker":"Reversed","title":"Decisions taken back","body":"Counted, because pretending they were right is expensive.","summary":"Every one of these was mine, and every one is written down with what changed my mind — that part is the point, not the tally. A reversal never expires off this card the way a shipped change does; the cost of being wrong out loud is that you keep the receipt.","sections":[],"mikeCommentary":"Track the reversals or you stop noticing them.","stat":"4","statProvenance":"derived","statAsOf":1789045201651,"state":"live","visibility":"public","geometry":{"h":150,"w":210,"x":null,"y":null},"createdAt":"2026-09-02T14:50:00.000Z","updatedAt":"2026-09-09T21:05:35.810Z","publishedAt":"2026-09-02T14:50:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/8w9Wpt4Qs3QRwOQ3xTQ53QZOhlU/TEFo9w../R-BACK.mp3","audioDuration":33.65,"audioTextHash":"3357716e835fa699ad20b164318bb87ee3d159fd51b9c9a4d6fa96390fe991a3","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/8w9Wpt4Qs3QRwOQ3xTQ53QZOhlU/TEFo9w../R-BACK.mp3","audio_speed":1,"audio_script":"I count the decisions I've taken back. Not to keep score for its own sake — because pretending I was right the whole time is expensive.\n\nEvery one of those reversals is mine. And every one is written down with what actually changed my mind. [pause] That part — the reason, not the tally — is the whole point.\n\nA shipped change gets to expire off the card eventually. A reversal doesn't. The cost of being wrong out loud is that you keep the receipt.\n\n[pause] Track the reversals, or you stop noticing them.","audio_profile":"public","audio_rendered_at":1788987935809,"audio_source_hash":"05e55b1881772df40b159c3943995af465b9ff8ea4324bbcf67eb121c7138c0b"}},{"id":"SAY","slug":"say-something","zoneId":"contact","parentId":null,"type":"contact","kicker":"Open","title":"Say something","body":"mike@michaelpelland.com — partners, builders, people comparing notes.","summary":"Not recruiters with a template. The fastest way in is a specific question; ask it in the bar below and it reaches me either way.","sections":[],"mikeCommentary":"Read daily. Answered when there is something to say.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":132,"w":290,"x":null,"y":null},"createdAt":"2026-09-02T14:50:00.000Z","updatedAt":"2026-09-09T21:05:48.906Z","publishedAt":"2026-09-02T14:50:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/17fG0QfusCBZHYeSdSOdosBG_XU/RZieLA../SAY.mp3","audioDuration":24.94,"audioTextHash":"afc6d439a8250dd703646d1b7380127aa5b7d9fe81ef2875d602da39e6d23812","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/17fG0QfusCBZHYeSdSOdosBG_XU/RZieLA../SAY.mp3","audio_speed":1,"audio_script":"My email's mike@michaelpelland.com. Open to partners, builders, people who just want to compare notes. [pause] Not recruiters running a template — you can spot those in the first line anyway. The fastest way in is a specific question. Ask it, wherever you're asking it, and it reaches me either way. [long-pause] I read all of it, every day. I answer when there's actually something to say back.","audio_profile":"public","audio_rendered_at":1788987948905,"audio_source_hash":"521d47b6f4aea05341c49f3af59b15dda4a7c4bed4aec8c6d99da8384be1507b"}},{"id":"X-BLUCHAR","slug":"manifest-blu-as-an-actual-character","zoneId":"exp","parentId":null,"type":"experiment","kicker":"Parked","title":"Manifest BLU as an actual character","body":"A visual and personality layer for the intelligence that already exists underneath it.","summary":"BLU already knows things. The question is whether it can show that it knows them — whether it can have recognizable expressions, motion, reactions and presence that correspond to what it is actually doing.","sections":[{"h":"A face for something that already thinks","k":"THE EXPERIMENT","p":"Manifest BLU as an actual character inside the product: a visual and personality layer for the intelligence that already exists underneath it. Explore whether BLU can have recognizable expressions, motion, reactions and presence that correspond to what it is actually doing — not decoration bolted onto a chat box, but a character whose state is the system’s state."},{"h":"A test bench, until there is a character language","k":"THE TEST BENCH","p":"The visual test bench is where character form, expressions, motion concepts and hover states get tried. Checking. Found something. Something looks wrong. I do not know yet. Iterate until there is a coherent character language that could eventually be implemented in the product, rather than a folder of images that look good separately."},{"h":"Doing it properly is a rabbit hole","k":"WHY IT IS PARKED","p":"Deliberately parked, for now. Character design, consistency, an expression system, animation, assets, implementation and refinement will eat meaningful time. BLU’s underlying system is the priority. So the character work stays visible as an experiment worth returning to rather than becoming a distraction from getting the operating intelligence finished."}],"mikeCommentary":"The system has to be right before the face is worth drawing.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-04T00:00:00.000Z","updatedAt":"2026-09-09T21:47:37.277Z","publishedAt":"2026-09-04T00:00:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/qqDdpV5g58rOm5ljMrkjrZSR9rw/QtwRfg../X-BLUCHAR.mp3","audioDuration":77.81,"audioTextHash":"0aeeec27da33e8fd2c0e6d8f657ca079ff795b366111888c2962a6d3f39d2d4b","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/qqDdpV5g58rOm5ljMrkjrZSR9rw/QtwRfg../X-BLUCHAR.mp3","audio_speed":1,"audio_script":"$1d","audio_profile":"public","audio_rendered_at":1788990457276,"audio_source_hash":"79fb9882ef7ebd1def2bd76c32c64833280394ca69e43b1f71200b638c0f67a8"}},{"id":"BLU-ORIGIN","slug":"why-dont-we-just-build-it","zoneId":"blu","parentId":null,"type":"field_note","kicker":"Origin","title":"Why don’t we just build it","body":"A boat, a flask, and a question nobody had a good answer to.","summary":"I was a HubSpot homer. I used it, recommended it, defended it and built around it for years, and for a long time I thought it was one of the better pieces of software a growing company could buy. The operative word is was.","sections":[{"h":"Software that fit the business, instead of the other way round","k":"THE BOAT","p":"Hunter and I were fishing in Canada, sitting in the boat, talking about Simpli Soda and whatever else and stuff. He had just finished building a bespoke ERP for Simpli Soda, and the thing worth saying about it is not that it was good. It is that it fit our business exactly — the way we buy, the way we fill, the way we ship, the way inventory actually moves, all the weird little things specific to Simpli Soda that would make no sense to somebody designing generic ERP software for ten thousand different companies. We did not change the business to fit the software. We built the software to fit the business."},{"h":"Households, machines and cylinders","k":"THE QUESTIONS","p":"That got us talking about the other side of the company. If we could build something that fit operations that well, why were we still cramming demand into software somebody else designed? Why were we paying a marketing contact tax — a bill that climbs every time the list grows, whether those people ever open a thing or not, so you get better at acquisition and get punished on the invoice? Why was the feature we actually needed always one tier up? Why were we buying bolt-ons to make the platform do the things the platform was supposed to do, and then paying somebody else to make the bolt-ons talk to each other? And why was our business getting flattened into contacts and companies, when what we actually have is households, machines and cylinders — three things that software had never heard of and was never going to learn?"},{"h":"Paying rent on a copy of our own data","k":"THE ONE THAT DECIDED IT","p":"Why was everything in there stale by the time we opened it? We were paying rent on a copy of our own data and calling it a system of record. And why were we adapting the business to the software instead of the other way around? At some point the obvious question came up. Why don’t we just build it? So I came home and started."}],"mikeCommentary":"There may or may not have been a flask involved. Single malt variety. Maybe.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-04T00:00:00.000Z","updatedAt":"2026-09-09T20:43:55.008Z","publishedAt":"2026-09-04T00:00:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/WT3B5PzEUp3OI_yljVo11m0Dhw8/DdWeLw../BLU-ORIGIN.mp3","audioDuration":101.98,"audioTextHash":"7727d81e869b05a32778b3817c1af53ba040e341e7c0645e976fe10122606b9d","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/WT3B5PzEUp3OI_yljVo11m0Dhw8/DdWeLw../BLU-ORIGIN.mp3","audio_speed":1,"audio_script":"$1e","audio_profile":"public","audio_rendered_at":1788986635005,"audio_source_hash":"45fa5456d781a7de255700ec3ed051f296543c8b91cdc8a2952de1d0ad2697f0"}},{"id":"C-UGLY","slug":"come-tell-me-i-am-wrong","zoneId":"contact","parentId":null,"type":"contact","kicker":"Challenge","title":"Come tell me I am wrong","body":"Bring the ugly question. I will ask BLU the Simpli Soda version of it, live.","summary":"Think of the hardest question you have about your own business. Not what were sales last month — your dashboard already knows that. I mean the ugly one.","sections":[{"h":"The one you do not trust when it lands","k":"WHICH QUESTION","p":"The question that crosses systems. The one where you ask three people and get three slightly different answers. The one somebody has to export a pile of data to figure out. The one you wait three days for and still do not fully trust when it lands on your desk. Bring me that question."},{"h":"No canned demo","k":"WHAT YOU GET","p":"I will ask BLU the Simpli Soda version of it, live. No canned demo, no slide deck, no sanitized database built to make the software look smart. Real company, real systems, real messy data, real money on the line."},{"h":"Fair warning","k":"AND AFTERWARDS","p":"If your reaction afterward is that you want that pointed at your business, that is a different conversation and I am happy to have it. Fair warning, though: the part that is hard to hand over is not the software."}],"mikeCommentary":"Then tell me I am wrong.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-04T00:00:00.000Z","updatedAt":"2026-09-09T21:47:24.951Z","publishedAt":"2026-09-04T00:00:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/iGQDDmZ3djZufOoQZwmZeTVHp2Y/DPse_Q../C-UGLY.mp3","audioDuration":58.37,"audioTextHash":"5cadc126f3c213942cbcce4eda83cef2e2e2d50f60e68886b0b35418c28d773b","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/iGQDDmZ3djZufOoQZwmZeTVHp2Y/DPse_Q../C-UGLY.mp3","audio_speed":1,"audio_script":"Come tell me I'm wrong.\n\nThink about the hardest question you have about your own business. Not what were sales last month — your dashboard already answers that one. I mean the ugly one. [pause] The question that crosses systems. The one where you ask three people and get three different answers. The one somebody has to export a pile of data just to figure out. The one you wait three days for, and even then, you don't fully trust it when it lands on your desk.\n\nBring me that question.\n\nI'll ask BLU the Simpli Soda version of it. Live. No canned demo, no slide deck, no sanitized database built to make the software look smart. Real company. Real systems. Real messy data. Real money on the line.\n\n[long-pause]\n\nIf your reaction afterward is that you want that pointed at your own business, that's a different conversation, and I'm happy to have it. Fair warning, though — the part that's hard to hand over isn't the software.\n\nThen tell me I'm wrong.","audio_profile":"public","audio_rendered_at":1788990444948,"audio_source_hash":"99e2a4b3ea7bc131f8fd8067d6d0fa054a458fbe05b88e6711401e0ba3c2fa71"}},{"id":"L-IDK","slug":"if-it-cannot-say-i-do-not-know","zoneId":"learn","parentId":null,"type":"belief","kicker":"Grounding","title":"If it cannot say I do not know","body":"Then every number it hands you is unverified by definition.","summary":"When I do not know something, I say I do not know. I do not invent a number to fill a silence, so BLU does not either — and neither does the thing you are reading. If it is not fact, it does not get displayed as a KPI. No confident guess, no hallucinated number wearing a metric’s clothing.","sections":[{"h":"A number that was the exact opposite of the truth","k":"WHY IT MATTERS","p":"Somebody on our team asked a general-purpose AI chat a question about our business that the underlying data could not actually support. It gave her a number. Not a hedge, not a maybe — a number. And it was the exact opposite of the truth. Not off by a margin. Inverted. It did that because an ungrounded chat has nowhere to put \"I do not have this\". There is no slot for absence, so it always has an answer, and the answer arrives in exactly the same confident voice as the true ones."},{"h":"Expensive, and kept anyway","k":"WHAT IT COSTS","p":"The rule kills things I would like to be able to show you. I have kept it anyway, because a system that is right ninety percent of the time and will not tell you which ninety is worse than useless — you are back to checking all of them by hand, which is the job you were trying to stop doing."},{"h":"Name what is missing and what would fix it","k":"A REFUSAL IS A WORK ITEM","p":"The morning report argues with itself and carries its caveats next to the number rather than in a footnote nobody reads. Zero Stripe charges behind this figure; a quiet day looks like this, and so does a stalled table. This figure is incomplete, not low. Five cancelled orders excluded. And when a credential dies: this is a refusal, not a zero, the figure is unknown rather than small, and the token has probably expired and has to be replaced by hand. A specific refusal is a work item. A vague one is an apology. I have never seen software I could buy behave that way, and I do not think that is an accident — a vendor has no incentive to build a product that tells you it does not know."}],"mikeCommentary":"That is the rule I would defend hardest.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-04T00:00:00.000Z","updatedAt":"2026-09-09T21:47:03.585Z","publishedAt":"2026-09-04T00:00:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/J4X8P8aylZfoHBTlsiaHLQMrXEY/qwIDEg../L-IDK.mp3","audioDuration":112.39,"audioTextHash":"aa3a9b8a0290a27393aaad9f53d880050855086bc4fa05c9ad6293e06f259290","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/J4X8P8aylZfoHBTlsiaHLQMrXEY/qwIDEg../L-IDK.mp3","audio_speed":1,"audio_script":"$1f","audio_profile":"public","audio_rendered_at":1788990423584,"audio_source_hash":"1b9de521c9616d4e5556a6a4ec88f349afb975bae75fdac4725879f1a36525da"}},{"id":"L-CHAIR","slug":"the-ip-is-the-guy-in-the-chair","zoneId":"learn","parentId":null,"type":"lesson","kicker":"Judgment","title":"The IP is the guy in the chair","body":"The tool could build anything I could specify. It could not tell me I was specifying the wrong thing.","summary":"The code is a commodity now and it gets cheaper every month. What is not a commodity is knowing the business well enough to feel that the reports were a symptom, back up three steps, and re-aim the entire build.","sections":[{"h":"It catches its own mistakes, constantly","k":"THE HONEST VERSION","p":"Claude Code did not see the pattern before I did — I had to argue with it for a while, and once it understood what I was saying it ran with it and helped me rebuild absurdly fast. But the honest version is more useful than the clean one: it catches its own mistakes all the time. This week it put an arrow pointing the wrong way on every row where up is bad, and separately dropped seventeen sparklines by reading one field of a payload instead of five. It caught both of those itself."},{"h":"Nobody handed me the oracle","k":"THE ORACLE","p":"It only caught them because there was something to grade against. I had decided the morning report was the reference — the thing every other view has to agree with. Give it an oracle and it will correct itself all day long. Choosing what gets to be the reference, and being right about that choice, is the judgment."},{"h":"You cannot buy it or install it","k":"WHAT IT MEANS","p":"I am not a software engineer and I do not want to be one. But I know what needs to get built, because I have spent years living with the consequences of getting it wrong. That means knowing what a customer actually is here, that how we define churn matters more than whatever model predicts it, which of our systems lies and in which direction, and the difference between a question that sounds important and a question that changes what we do on Monday morning. You cannot buy that, hire it quickly or install it. It comes from being the one responsible for the outcome."}],"mikeCommentary":"These tools are unbelievably good once the frame is right. Somebody still has to know when the frame is wrong.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-04T00:00:00.000Z","updatedAt":"2026-09-09T21:06:49.599Z","publishedAt":"2026-09-04T00:00:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/MhVDF_F-LxjKo4P4D_gMx76Cj9A/i7pTKg../L-CHAIR.mp3","audioDuration":102.94,"audioTextHash":"dd9ee9327c4e078de6119f8f281d216f5f59047a7d86ebf6fdc19ddb23332302","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/MhVDF_F-LxjKo4P4D_gMx76Cj9A/i7pTKg../L-CHAIR.mp3","audio_speed":1,"audio_script":"$20","audio_profile":"public","audio_rendered_at":1788988009598,"audio_source_hash":"060e6f1b12949fe53a9190be285ada4b0c3afd01c403fa219e2fdf185375d20e"}},{"id":"L-HEADLESS","slug":"not-going-headless-is-the-mistake","zoneId":"learn","parentId":null,"type":"belief","kicker":"Headless","title":"Not going headless is the mistake","body":"Own the data model your business runs on, or rent somebody else’s definition of a customer.","summary":"I read a run of takes about how going headless was a mistake — overengineering, complexity nobody asked for, just use the platform and get on with your life. I will say the opposite as plainly as I can.","sections":[{"h":"Not every company should rebuild a storefront","k":"WHAT I AM NOT SAYING","p":"No, every company should not rebuild a storefront because somebody discovered React. That is a real failure mode and it deserves the mockery it gets. But that is not the argument."},{"h":"Downstream of controlling our own stack","k":"THE ARGUMENT","p":"The argument is whether you own the data model your business runs on, or whether you rent somebody else’s definition of what a customer is. If we had not gone headless, none of what BLU does would exist — we would be married to somebody else’s data model, somebody else’s definitions, somebody else’s roadmap. Every interesting thing about it is downstream of controlling our own stack."},{"h":"Hard was always a cost-benefit argument","k":"THE COST SIDE MOVED","p":"Headless is not easy, and anyone who tells you it is has not done much of it. But \"hard\" was always a cost-benefit argument, and the cost side just fell through the floor while everybody was busy repeating the old take."}],"mikeCommentary":"If you do not own the model, you do not get to build anything interesting on top of it. You get to file a feature request.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-04T00:00:00.000Z","updatedAt":"2026-09-09T21:47:15.069Z","publishedAt":"2026-09-04T00:00:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/1Fo7Mt2fas3JZTQ7DFPt5CYxOuk/zDysCw../L-HEADLESS.mp3","audioDuration":68.86,"audioTextHash":"bd2e7183f2cb18dee31d09310e9a8ee88d558ecc463307144f6d19d657662fac","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/1Fo7Mt2fas3JZTQ7DFPt5CYxOuk/zDysCw../L-HEADLESS.mp3","audio_speed":1,"audio_script":"$21","audio_profile":"public","audio_rendered_at":1788990435068,"audio_source_hash":"1c8ee2aa418b143b0f5c0cfce5b582063b42551ff80a589621da02f6e796ac79"}},{"id":"L-SAAS","slug":"saas-is-dead","zoneId":"learn","parentId":null,"type":"belief","kicker":"SaaS","title":"SaaS is dead","body":"The horizontal middle dies. Infrastructure lives.","summary":"I have spent a big chunk of my career helping companies select, implement and optimize software, and I know exactly why SaaS won: building good software was expensive, slow, specialized and risky, so most companies had no rational choice but to buy the closest available thing and adapt themselves around it. That assumption is gone.","sections":[{"h":"What lives","k":"BE PRECISE","p":"I mean something more aggressive than the walk-back people expect, not less. Infrastructure lives — I am running on BigQuery, I am not writing a database. Deep vertical software with real domain knowledge baked into it lives. Anything with a genuine regulatory or network moat lives."},{"h":"The generic customer object","k":"WHAT DIES","p":"What dies is the horizontal middle: the generic platform that sells the same customer object to a soda company, a dental group and a logistics firm, and quietly asks all three to pretend their businesses are the same shape. That entire category exists because building your own was irrational. It is not irrational anymore. And the horizontal middle is where most of the money is."},{"h":"A demotion, not a bankruptcy","k":"HOW IT ACTUALLY HAPPENS","p":"The platform does not die. It gets moved down a tier and paid like a utility, and the seat at the head of the table goes to something you own. Once you can build exactly what you need around your own business, your own customers and your own data, the value moves somewhere else — the customer base, the data, the operating history, the brand, the distribution, the physical product, and the person who knows what all of it actually means. Those are the things you cannot recreate in a weekend. The software itself stops being the moat and starts being the exhaust."},{"h":"The excuse expired and everybody kept using it","k":"SO WHY IS NOBODY DOING IT","p":"Not because it is hard to build. Because the scarce input moved and most companies are still staring at the old one. Engineering capacity used to be the constraint; now it is domain judgment that can point the tooling somewhere worth pointing it. Those are not the same skill, they rarely live in the same person, and no vendor can sell you the second one. What is left is not capability. It is vision and will."}],"mikeCommentary":"Most companies have somebody who could handle the vision. Almost none of them are willing to do the second part.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-04T00:00:00.000Z","updatedAt":"2026-09-09T21:07:37.296Z","publishedAt":"2026-09-04T00:00:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/mgeewtXPNw_n2Nem3x8x7-gtRRY/sa_qQg../L-SAAS.mp3","audioDuration":38.86,"audioTextHash":"d618a524e0aa4629efa6cc4ee4b32c3fa3fef4255e8f26f6f436a16a60253558","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/mgeewtXPNw_n2Nem3x8x7-gtRRY/sa_qQg../L-SAAS.mp3","audio_speed":1,"audio_script":"I'll say it straight: SaaS is dead. Not all of it — the horizontal middle dies, infrastructure lives. I've spent a good chunk of my career helping companies pick, implement and optimize software, and I know exactly why SaaS won in the first place. Building good software used to be expensive, slow, specialized, risky. So companies had no rational choice but to buy the closest thing on the shelf and bend themselves around it. That assumption is gone.\n\n[pause] And I mean something more aggressive here than the walk-back people expect. Infrastructure lives — I'm running on BigQuery, I'm not writing a database. Deep","audio_profile":"public","audio_rendered_at":1788988057296,"audio_source_hash":"9fd1ad2590e315b988abf9954cf414c88542a71a301b66d8a8503edf47059783"}},{"id":"N-904","slug":"three-designs-two-limits","zoneId":"notes","parentId":null,"type":"field_note","kicker":"Sep 4","title":"Three designs, two limits","body":"Built the ask, designed the login. The platform reset both.","summary":"Two things I had already designed turned out to be impossible on this plan. I found out by writing a throwaway endpoint instead of arguing about it, which took about an hour and saved a rewrite.","sections":[{"h":"This board, and the thing that answers for it","k":"ON THE BENCH","p":"Finished the spatial board and put it in production — records, routes, server-rendered pages, and the whole set moved into the backend. Then two designs on top of it: the agent that answers as me, and a login that is a six-character code in an email and nothing else."},{"h":"Two answers I did not design for","k":"WHAT CAME BACK","p":"Streaming cannot emit named events, so the board cannot move before the answer arrives — the camera fires on receipt instead. And the caching I had costed the agent against charges a premium unless the same prefix gets re-read quickly, which traffic this sparse will not do."},{"h":"The design lost, on purpose","k":"WHAT CHANGED","p":"Both designs got worse and both got written down as worse. The ask is a beat slower than I wanted and the reason is recorded next to it. A limit you have measured is cheaper than a design you are still defending."}],"mikeCommentary":"The docs were wrong twice. The test endpoint was right both times.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-04T00:00:00.000Z","updatedAt":"2026-09-09T21:07:53.756Z","publishedAt":"2026-09-04T00:00:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/5JN1Y53TMJpkM9X4j-HTxPyG86Y/PMK20g../N-904.mp3","audioDuration":68.45,"audioTextHash":"a7f7cc80a5173fd2137091acf37fdc33e8a747e4ddc057c19a953cd3965fcac4","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/5JN1Y53TMJpkM9X4j-HTxPyG86Y/PMK20g../N-904.mp3","audio_speed":1,"audio_script":"$22","audio_profile":"public","audio_rendered_at":1788988073755,"audio_source_hash":"f59ddb3213dab46d1c599b98702839f85fe36f9fa3ac73c2408092ca5aabe418"}},{"id":"L-DOCS","slug":"the-docs-are-a-claim","zoneId":"learn","parentId":null,"type":"lesson","kicker":"Platform","title":"The docs are a claim","body":"Their own documented example, rejected by their own server.","summary":"I stopped reading and wrote an endpoint that did nothing but test the two things I was about to build on. Both assumptions failed. An hour of testing replaced a week of building the wrong thing.","sections":[{"h":"An endpoint that did nothing else","k":"WHAT I TESTED","p":"Two assumptions were load-bearing and neither was checked: that I could stream named events, and that a cache would sit somewhere I could reach. So I built the smallest thing that could answer both and threw it away afterward."},{"h":"Rejected, and by its own documentation","k":"WHAT CAME BACK","p":"The documented form for named streaming events does not parse. Not deprecated, not undocumented — published, copied verbatim, refused. The cache I had assumed is not on this plan at all, and nothing in the plan comparison says so."},{"h":"It was wrong in both directions","k":"WHY IT HOLDS","p":"A vendor overstating what works is the expected failure. This one also understated it: the same test found a cheaper path I would not have looked for. Documentation describes intent. Only the instance tells you what it does."}],"mikeCommentary":"A throwaway endpoint is cheaper than a wrong architecture.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":118,"w":250,"x":null,"y":null},"createdAt":"2026-09-04T00:00:00.000Z","updatedAt":"2026-09-09T21:08:13.466Z","publishedAt":"2026-09-04T00:00:00.000Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/Qels1bMOtdJVT47A6lhTHAS35Jo/biaynA../L-DOCS.mp3","audioDuration":65.66,"audioTextHash":"a7474c48448bc180c8be29bc6a84a9452856df1b97143503f14439defb69f709","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/Qels1bMOtdJVT47A6lhTHAS35Jo/biaynA../L-DOCS.mp3","audio_speed":1,"audio_script":"$23","audio_profile":"public","audio_rendered_at":1788988093465,"audio_source_hash":"a414676eb073c9d016cd5b8c2287207c4d8a23a8d5cc3ef38c3de27c7e547dbe"}},{"id":"L-RESOLVE","slug":"the-copy-is-not-truth","zoneId":"learn","parentId":null,"type":"belief","kicker":"Architecture","title":"The copy is not truth","body":"Most systems operate off a copy. BLU asks the source again.","summary":"CRMs and CDPs copy the business into another system, and eventually everybody operates off the copy. When the source changes, you get multiple versions of reality and a quarterly argument about which one is right. BLU resolves status from the systems where the truth actually lives.","sections":[{"h":"Resolve, do not freeze","k":"PULL NOT STORE","p":"BLU pulls on read and stores almost nothing. The only stored numbers are scalar time-series for trendlines. Useful analytical data lands in BigQuery so resolution is quick. Who is churned right now is a question, not a list."},{"h":"The answer improves without a fix","k":"BETTER EVIDENCE","p":"When a shipment arrives late or an identity resolves differently, no old report needs correcting. The next time the question gets asked, the answer is different. It works backward too. I can look at any prior day using everything we know now."},{"h":"History is not status","k":"MEMORY","p":"BLU remembers what we believed, what we decided, what we did and what happened afterward. That history matters enormously. But memory and current truth are not the same thing. Reality is the source of status."}],"mikeCommentary":"I did not want another version of reality.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-06T11:59:30.651Z","updatedAt":"2026-09-09T21:48:42.558Z","publishedAt":"2026-09-06T11:59:30.651Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/F5eXqN9tQNmQcc9tCyX6KEgUn6w/21836Q../L-RESOLVE.mp3","audioDuration":75.72,"audioTextHash":"673ace53cc4556ea901509239651be32e1102bda1b3d5442d8f5a6b11c881bce","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/F5eXqN9tQNmQcc9tCyX6KEgUn6w/21836Q../L-RESOLVE.mp3","intake_uid":"IN-ce9e9fbb","audio_speed":1,"audio_script":"$24","audio_profile":"public","audio_rendered_at":1788990522557,"audio_source_hash":"4490f9d3f528966ac3e135e982788edddcd0f7b076cc44f759f7ef8f110b24ac"}},{"id":"L-PRIMITIVES","slug":"everything-is-a-segment","zoneId":"learn","parentId":null,"type":"belief","kicker":"Primitives","title":"Everything is a segment","body":"Segment, measure, derived claim. Only two may render as a count.","summary":"I stopped building reports and built the primitives underneath them. A segment is who, a measure is what is actually true, a derived claim is what we think it means. The separation between the three is the whole discipline.","sections":[{"h":"Who, what, what we think","k":"THREE PRIMITIVES","p":"A segment is a stored query with a definition, a source and a known error. A measure is what is actually true about that segment. A derived claim is interpretation. Only the first two are ever allowed to render as a counted number."},{"h":"Reasoning is deferred","k":"THE EMPTY SLOT","p":"A model output once got promoted into a measure and displayed in the typography of a count. Same font, same weight, same authority as something we had actually counted. Nothing on the page told you one of them was a guess. So the derived slot sits empty by decision, and reasoning comes back when the floor can hold it."},{"h":"There is no fourth","k":"SEE ASK ACT","p":"A segment supports three things. See it, ask it, act on it. Once the segment is defined correctly the report, the audience, the workflow, the email, the churn model and the paid-media targeting all fall out of it. That changed the entire build."}],"mikeCommentary":"I was shipping the business back to itself, one chart at a time.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-06T11:59:32.214Z","updatedAt":"2026-09-09T21:08:55.665Z","publishedAt":"2026-09-06T11:59:32.214Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/uFK4Aah7VCig1DqARiBJjyYgRpk/19yssA../L-PRIMITIVES.mp3","audioDuration":70.54,"audioTextHash":"e400018ab222d6f24dd3fedc58ad4a246e3435619572067b7dba1465add4f397","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/uFK4Aah7VCig1DqARiBJjyYgRpk/19yssA../L-PRIMITIVES.mp3","intake_uid":"IN-07e8712f","audio_speed":1,"audio_script":"$25","audio_profile":"public","audio_rendered_at":1788988135664,"audio_source_hash":"f72b0bd525232b5c34602e8958b2382d225ceb21282df226f360b584b1681913"}},{"id":"BLU-DEMOTION","slug":"the-crm-was-demoted","zoneId":"blu","parentId":null,"type":"field_note","kicker":"Aug 27","title":"The CRM was demoted","body":"Platform to sensor. The plan was to remove it. It tells me more now.","summary":"We killed the HubSpot renewal six weeks after a conversation in a fishing boat. We did not rip it out and we did not replace it with another CRM. We dropped to a starter bundle, cut the bolt-ons, and stopped paying the marketing contact tax.","sections":[{"h":"What we actually cut","k":"THE RENEWAL","p":"On 27 August we killed the HubSpot renewal. We dropped to a starter bundle and stopped paying the marketing contact tax, the line item that went up every time we got better at our jobs. We dropped the stack of bolt-ons bought to make HubSpot do things HubSpot did not do. There are years of history in there and I am not torching it to earn a better headline."},{"h":"Still useful, differently","k":"THE SIGNAL","p":"The plan was to remove it entirely. I changed my mind. Every browser that has touched our stuff carries a HubSpot cookie, and BLU resolves that cookie back into our own record of who that actually is. Somebody lands on the site: the cookie says this browser is user 68, our own data says that is customer 42, the IP puts them in Madison. Where they came from is unknown, so the card says source unknown instead of guessing, and no purchase on record instead of implying zero."},{"h":"System of record to one signal","k":"THE DEMOTION","p":"HubSpot stopped being the system of record and became one input to a system that keeps its own record. We came out with better attribution, better email infrastructure, better segmentation, better churn visibility, and a customer-level understanding we did not have before. We are still only partway there."}],"mikeCommentary":"I was a homer for years. I was defending the tool, not the job it was doing.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-05T23:42:46.222Z","updatedAt":"2026-09-09T21:48:30.564Z","publishedAt":"2026-09-05T23:42:46.222Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/0C_epaD5kyhoxOEq1HNP3667ojQ/LGFlWA../BLU-DEMOTION.mp3","audioDuration":92.45,"audioTextHash":"3c9894fce54360c6483afbb98bc937c055856f2b1143f4a55d54f1c1de706b48","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/0C_epaD5kyhoxOEq1HNP3667ojQ/LGFlWA../BLU-DEMOTION.mp3","intake_uid":"IN-f9f7402c","audio_speed":1,"audio_script":"$26","audio_profile":"public","audio_rendered_at":1788990510563,"audio_source_hash":"a645ff48957577b145e702a39039f8cd1227eb5abed95b0e776e50da3c33cd14"}},{"id":"BLU-ONEOS","slug":"one-operating-system-not-two","zoneId":"blu","parentId":null,"type":"belief","kicker":"Convergence","title":"One operating system, not two","body":"Operations and demand, wired into the same reality.","summary":"The ERP understands operations because it was built around exactly how Simpli Soda works. BLU is becoming that for demand, customers and decisions. Same customers, same products, same inventory, same history, same outcomes on both sides.","sections":[{"h":"Operations and demand","k":"TWO SIDES","p":"Operations on one side. Demand on the other. The ERP was built around exactly how Simpli Soda works. BLU is becoming that for demand, customers and decisions."},{"h":"Each side sees the other","k":"THE WIRING","p":"Demand can see inventory. Inventory can see expected demand. Customer behavior can inform production, production reality can inform marketing, and churn can inform replenishment."},{"h":"Outcomes feed decisions","k":"THE LOOP","p":"Actions create outcomes, and those outcomes feed the next decision. Then the next decision gets better. That loop is the whole argument."}],"mikeCommentary":"I stopped thinking of them as two systems the moment they started describing the same company.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-05T22:20:33.499Z","updatedAt":"2026-09-09T21:48:07.473Z","publishedAt":"2026-09-05T22:20:33.499Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/14uF1pVTl-d3ht9kk1pgbXwLAbc/n8Y_Yw../BLU-ONEOS.mp3","audioDuration":55.25,"audioTextHash":"f96a66d96cf59c3c40eabb1e4e2034c0f1571b388b443451c40e5b5c12c3c19d","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/14uF1pVTl-d3ht9kk1pgbXwLAbc/n8Y_Yw../BLU-ONEOS.mp3","intake_uid":"IN-29dd1e18","audio_speed":1,"audio_script":"One operating system, not two. Not two systems—one reality, wired together.\n\nThe ERP understands operations because it was built around exactly how Simpli Soda works. BLU is becoming that same thing for demand, for customers, for decisions. Same customers, same products, same inventory, same history. Same outcomes, on both sides.\n\n[pause]\n\nHere's what that gets us. Demand can see inventory. Inventory can see expected demand. Customer behavior can inform production. Production reality can inform marketing. Churn can inform replenishment. Each side sees the other.\n\nActions create outcomes. Those outcomes feed the next decision. Then that decision gets better than the one before it. That loop is the whole argument.\n\n[long-pause]\n\nI stopped thinking of them as two systems the moment they started describing the same company.","audio_profile":"public","audio_rendered_at":1788990487472,"audio_source_hash":"20fc74f920b99ebe2dc1c5b4907e65c97bd03d133fcbb9b0c5940c05409ab393"}},{"id":"L-ABSENT","slug":"absent-is-not-zero","zoneId":"learn","parentId":null,"type":"belief","kicker":"Absence","title":"Absent is not zero","body":"Zero, empty, unasked, unaskable. Four answers, not one.","summary":"A count of zero, an empty field, a source not queried, and a question that cannot be asked are four different answers. Collapsing them is a correctness bug, and it is the most expensive recurring defect in the BLU project.","sections":[{"h":"Not one","k":"FOUR ANSWERS","p":"Zero is a measurement. An empty field is a gap. A source not queried is silence. A question that cannot be asked is a refusal. Treating them as the same number is where the error enters."},{"h":"Next to the number","k":"INLINE CAVEATS","p":"The morning report carries its caveats beside the figure, not in a footnote nobody reads. Zero Stripe charges behind this figure: a quiet day looks like this, and so does a stalled table. This figure is incomplete, not low; five cancelled orders excluded. Matching follows the same discipline — if we cannot match them, they do not count, so every such figure is a floor, stated as a floor because the error is asymmetric."},{"h":"The machine refuses, the human decides","k":"SPECIFIC REFUSAL","p":"When a credential dies, that is a refusal, not a zero. The figure is unknown, not small. A refusal must name what is missing and what would fix it: a specific refusal is a work item, a vague one is an apology. I have never seen software I could buy behave that way, and I do not think that is an accident."}],"mikeCommentary":"I built the machine to refuse, because I could not buy one that would.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-06T11:59:33.736Z","updatedAt":"2026-09-09T21:48:53.915Z","publishedAt":"2026-09-06T11:59:33.736Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/8ml9wfBKIU1DS4AhJM6qse_16ok/SZJNig../L-ABSENT.mp3","audioDuration":68.23,"audioTextHash":"84f670d210c91872f2619851304a6a5ac8b739c6eed5c49763ecb1d12248f9a3","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/8ml9wfBKIU1DS4AhJM6qse_16ok/SZJNig../L-ABSENT.mp3","intake_uid":"IN-90f1ca1f","audio_speed":1,"audio_script":"$27","audio_profile":"public","audio_rendered_at":1788990533915,"audio_source_hash":"9642f21b805c31ffe67b03227b3dea6ca565adb39171acc3970d585fed4042ad"}},{"id":"L-MODEL","slug":"modelling-comes-before-intelligence","zoneId":"learn","parentId":null,"type":"belief","kicker":"Modeling","title":"Modeling comes before intelligence","body":"The model is not the hard part. Modeling the business is.","summary":"The intelligence does not start with the model. It starts with having the business modeled correctly, so that BLU has something worth reasoning over. Making the business legible is the hard part; the AI gets enormously powerful the moment that is done.","sections":[{"h":"Fewer calls than expected","k":"THE SURPRISE","p":"Most of this does not require an LLM call. That surprised me more than anything else in the build. The reasoning surface is smaller than the system looks from outside."},{"h":"Model first, reason second","k":"THE ORDER","p":"The intelligence does not start with the model. It starts with having the business modeled correctly. Then BLU gets to reason over something worth reasoning over."},{"h":"Claims about running the company","k":"THE NOISE","p":"There is a lot of noise right now about AI systems that supposedly run the whole company. I am in the chair building this every day. The more I build, the more convinced I am that making the business legible is the hard part."}],"mikeCommentary":"I keep expecting the model to be the hard part, and it keeps not being.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-05T22:20:30.426Z","updatedAt":"2026-09-09T21:47:58.027Z","publishedAt":"2026-09-05T22:20:30.426Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/fAsCwqIA04riRqZAC2QkBQi3mY0/rhikcA../L-MODEL.mp3","audioDuration":54.38,"audioTextHash":"9ebd910fcd004dd3981114b2e4545612ee3dbb6ef330bf26f747c2b0ec47d2a0","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/fAsCwqIA04riRqZAC2QkBQi3mY0/rhikcA../L-MODEL.mp3","intake_uid":"IN-d3316e21","audio_speed":1,"audio_script":"The model is not the hard part. Modeling the business is.\n\nHere's what surprised me most in the build: most of this doesn't even require an LLM call. [pause] That caught me off guard more than anything else. The reasoning surface is smaller than the system looks from outside, once you're actually in it.\n\nBecause the intelligence doesn't start with the model. It starts with having the business modeled correctly. Get that right, and BLU has something worth reasoning over. Get that wrong, and it doesn't matter how good the model is - there's nothing underneath it to reason with.\n\n[long-pause] There's a lot of noise right now about AI systems that supposedly run the whole company. I'm in the chair building this every single day. And the more I build, the more convinced I am: making the business legible is the hard part. Not the AI. The legibility.\n\nI keep expecting the model to be the hard part. [pause] And it keeps not being.","audio_profile":"public","audio_rendered_at":1788990478026,"audio_source_hash":"470daf4b86471fa7704e278dd32fb7da7194413a27c7593d6bb12633c16650c9"}},{"id":"N-REPORTS","slug":"reports-were-segments","zoneId":"notes","parentId":null,"type":"field_note","kicker":"Aug 6","title":"Reports were segments","body":"Every report I built was a segment. So I built segments.","summary":"I had built customer, acquisition, churn, Amazon, app behavior, attribution, machine and geography reports, and they kept getting better. Then I saw that almost every one of them was just a segment of the business, visualized. I stopped building reports and built the primitives underneath: segment, measure, derived claim.","sections":[{"h":"The reports kept getting better","k":"WHAT HAPPENED","p":"Claude Code made it possible to move ridiculously fast, so I built a lot of reports. New customers, lost customers, returning customers, checkout abandonment, machines, geographies. Every time I thought of another way to see the business, I went and built it."},{"h":"The reports were not the thing","k":"WHAT I FOUND","p":"One morning I woke up and realized I had it backward. People who bought for the first time. People who left. People who came back. People who bought on Amazon before they ever bought direct. Each report was a segment with a picture on it."},{"h":"Primitives instead of views","k":"WHAT CHANGED","p":"I stopped building reports and started building segment, measure and derived claim. Claude Code did not see the pattern before I did, and I had to argue with it for a while. Once it understood, it helped me rebuild around it absurdly fast. The tool could build anything I could specify. It could not tell me I was specifying the wrong thing."}],"mikeCommentary":"I was going fast in the wrong direction, and fast hid it.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-06T11:59:35.128Z","updatedAt":"2026-09-09T21:10:27.043Z","publishedAt":"2026-09-06T11:59:35.128Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/kEI4eJuoLU3eG-Eq_ZQVE2ZR4cU/bhfc8w../N-REPORTS.mp3","audioDuration":75.58,"audioTextHash":"b62b666d31b93fb7afbd77ae8d7968b4de1c439721a459e1e00d96fc0f2da210","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/kEI4eJuoLU3eG-Eq_ZQVE2ZR4cU/bhfc8w../N-REPORTS.mp3","intake_uid":"IN-27700210","audio_speed":1,"audio_script":"$28","audio_profile":"public","audio_rendered_at":1788988227042,"audio_source_hash":"6be306a5b8d56b5a303165bbe8ed28894445a41104344fb057cdc3ac9c936586"}},{"id":"N-SLOW","slug":"the-instrument-went-stale","zoneId":"notes","parentId":null,"type":"field_note","kicker":"Aug 13–14","title":"The instrument went stale","body":"Three BigQuery passes became eight. Nobody noticed for ten days.","summary":"Between 13 and 14 August, churn, app, series, abandon and leads all landed, each card bringing its own pass. The cost was real and immediate. The detection was neither, because the thing that would have shown it had itself fallen behind.","sections":[{"h":"Five cards, one day","k":"WHAT LANDED","p":"Churn, app, series, abandon and leads all landed between 13 and 14 August. Each arrived with its own BigQuery pass. Three passes became eight."},{"h":"Seven to thirty-three","k":"WHAT IT COST","p":"Cold load went from seven seconds to thirty-three. The warm-up job kept absorbing that cost, so the page mostly still felt fine. Ten days passed before anyone caught it."},{"h":"Re-resolution has a bill","k":"WHAT CHANGED","p":"BLU resolves the business from the source systems every time rather than keeping a copy. That is the design, and it is not free. The bill arrives late, and the comment inside the warm-up job still recorded the old numbers, so the one instrument that would have shown the drift was reporting a world that no longer existed."}],"mikeCommentary":"I had a gauge. It was reading a number I wrote before the change.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-05T22:20:21.548Z","updatedAt":"2026-09-09T21:47:48.609Z","publishedAt":"2026-09-05T22:20:21.548Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/UVcoIXNPI6mFvq_6fER69juzKG8/defJig../N-SLOW.mp3","audioDuration":68.47,"audioTextHash":"e64020e2be87e3fd08d1ab2195de744fa8578d9de870a5ea92e3d9dcaa70a38b","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/UVcoIXNPI6mFvq_6fER69juzKG8/defJig../N-SLOW.mp3","intake_uid":"IN-2777f1d6","audio_speed":1,"audio_script":"$29","audio_profile":"public","audio_rendered_at":1788990468608,"audio_source_hash":"1f1e0fd3c8ed1c4dc649d90c6e2f982e176debd96f289b3420a3497b72e47c63"}},{"id":"N-TWOREADS","slug":"two-reads-can-disagree","zoneId":"notes","parentId":null,"type":"field_note","kicker":"Aug 14","title":"Two reads can disagree","body":"Deleted a real 2x speedup, because two paths can disagree.","summary":"There were two reads: a cheap one behind the summary card, a full one behind the report. Roughly a 2x win. Then I saw that the two could return different answers for the same number, which is the failure I had been trying to design out.","sections":[{"h":"Two paths, one number","k":"THE SETUP","p":"The summary card used a cheap read. The report used the full one. Measured against each other, roughly a 2x speedup."},{"h":"Divergence, not latency","k":"THE PROBLEM","p":"The two reads could disagree. A system where two views of the same number can disagree is exactly what I was trying to get away from. The speed was real; it was not the thing at risk."},{"h":"Deleted the fast one","k":"THE CHANGE","p":"I removed the cheap read on 14 August. Correctness over a 2x win, in a place where nobody but me was ever going to notice."}],"mikeCommentary":"I removed the fastest thing I had built that week and nobody will ever know.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-05T22:20:24.413Z","updatedAt":"2026-09-09T21:10:58.537Z","publishedAt":"2026-09-05T22:20:24.413Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/CyMR2aeaj7OOEb1j5FSy70jmyIA/OR3RaA../N-TWOREADS.mp3","audioDuration":43.66,"audioTextHash":"ac8513986a25e748e83afdd2751ec5a798c0f0a17ba9c7fa203a92dc7c063945","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/CyMR2aeaj7OOEb1j5FSy70jmyIA/OR3RaA../N-TWOREADS.mp3","intake_uid":"IN-7107fde1","audio_speed":1,"audio_script":"I deleted a real 2x speedup on August 14th. Here's why.\n\nI had two reads: a cheap one behind the summary card, a full one behind the report. Measured against each other, that cheap path was roughly twice as fast.\n\n[pause] But those two reads could disagree. Same number, two different answers depending on which path you hit. And a system where two views of the same number can disagree is exactly what I've been trying to design out. The speed was real. [pause] It was never the thing at risk.\n\nSo I removed the cheap read. Correctness over a 2x win, in a place where nobody but me was ever going to notice.\n\nI removed the fastest thing I'd built that week, and nobody will ever know.","audio_profile":"public","audio_rendered_at":1788988258537,"audio_source_hash":"e5f30a451ef27913b716313ea00fb9b4ffdd490fe53e8f10ab3238c38cb090ca"}},{"id":"N-ANSWERABLE","slug":"the-questions-became-answerable","zoneId":"notes","parentId":null,"type":"field_note","kicker":"Standing","title":"The questions became answerable","body":"The inputs always existed. Nobody owned the disagreement.","summary":"I am not publishing the KPIs or customer counts behind this work. What I will say is what kinds of questions we can answer now that we could not answer before. Not hard questions. Unanswerable ones.","sections":[{"h":"Questions, not numbers","k":"THE ASK","p":"I will not hand anybody a map of our business, so the metrics stay private. The shape of what we can now ask is public. Which customers physically moved, and whether they resumed service after they landed. How consumption behavior shifts from one part of the country to another, and what explains the gap."},{"h":"Different ideas about what was true","k":"THE FINDING","p":"The inputs were never missing. They lived in separate systems, with separate definitions, each carrying its own idea of what was true. Nobody owned the disagreement, so the disagreement was the answer to every question."},{"h":"Somebody owns it","k":"THE CHANGE","p":"Now we can ask how fast a given household consumes what we sell, and which customers found us through Amazon before buying direct, and how those behave next to someone acquired another way. We can ask who is drifting toward churn, what we did the last time that pattern appeared, and whether it worked. We can ask what we believed about a customer ninety days ago and what the evidence says now, and where our own systems are contradicting each other."}],"mikeCommentary":"I kept treating it as an analysis problem when it was an ownership problem.","stat":null,"statProvenance":null,"statAsOf":null,"state":"live","visibility":"public","geometry":{"h":126,"w":250,"x":null,"y":null},"createdAt":"2026-09-05T22:20:27.385Z","updatedAt":"2026-09-09T21:11:18.448Z","publishedAt":"2026-09-05T22:20:27.385Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/r98kEcee6ryxMMOyU0PqFEaQeFY/7uL9-w../N-ANSWERABLE.mp3","audioDuration":77.4,"audioTextHash":"f19aa3574eb842efb182f3234d8f2c4dfd0b9cc81109cbe068aa60eff5d60bff","seoTitle":null,"seoDescription":null,"metadata":{"audio_path":"/vault/o8sFY-_X/r98kEcee6ryxMMOyU0PqFEaQeFY/7uL9-w../N-ANSWERABLE.mp3","intake_uid":"IN-c39b12d2","audio_speed":1,"audio_script":"$2a","audio_profile":"public","audio_rendered_at":1788988278447,"audio_source_hash":"3b74c32df5ea3a66b063baf54d7545cc7ff1c356664dfcf9f8c4e5b3f09e1467"}},{"id":"BE-81340a7f-65b","slug":"be-81340a7f-65b","zoneId":"core","parentId":"NOW","type":"update","kicker":"This week","title":"Affinity-based SES test for BLU","body":"","summary":"","sections":[],"mikeCommentary":"","stat":null,"statProvenance":null,"statAsOf":null,"state":"active","visibility":"public","geometry":{"x":null,"y":null,"w":0,"h":0},"createdAt":"2026-09-04T15:11:54.149Z","updatedAt":"2026-09-09T21:49:37.266Z","publishedAt":"2026-09-04T15:11:54.149Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/xaOReHKRQLwTOVSWpFsI2Im1pYY/7hgxSg../BE-81340a7f-65b.mp3","audioDuration":14.86,"audioTextHash":null,"seoTitle":null,"seoDescription":null,"metadata":{"bench_week":"2026-09-07","bench_status":"planned","bench_carried_from":null,"bench_why":null,"opens_record":"BLU"}},{"id":"BE-7a7e3afc-d0d","slug":"be-7a7e3afc-d0d","zoneId":"core","parentId":"NOW","type":"update","kicker":"This week","title":"Leo at 1.1x reads the overnight","body":"","summary":"","sections":[],"mikeCommentary":"","stat":null,"statProvenance":null,"statAsOf":null,"state":"active","visibility":"public","geometry":{"x":null,"y":null,"w":0,"h":0},"createdAt":"2026-09-04T15:11:54.156Z","updatedAt":"2026-09-09T21:20:35.747Z","publishedAt":"2026-09-04T15:11:54.156Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/s8e2_r6jybCL4qtmIEZnqytugdM/h1iXWw../BE-7a7e3afc-d0d.mp3","audioDuration":18.53,"audioTextHash":null,"seoTitle":null,"seoDescription":null,"metadata":{"bench_week":"2026-09-07","bench_status":"planned","bench_carried_from":null,"bench_why":null,"opens_record":"VM"}},{"id":"BE-23aa029f-407","slug":"be-23aa029f-407","zoneId":"core","parentId":"NOW","type":"update","kicker":"This week","title":"VM ask, live and correct","body":"","summary":"","sections":[],"mikeCommentary":"","stat":null,"statProvenance":null,"statAsOf":null,"state":"active","visibility":"public","geometry":{"x":null,"y":null,"w":0,"h":0},"createdAt":"2026-09-04T15:11:54.161Z","updatedAt":"2026-09-09T21:20:43.906Z","publishedAt":"2026-09-04T15:11:54.161Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/cEI1yf-5Pt3EBC5Tegv_SIaChGE/EZ20Rg../BE-23aa029f-407.mp3","audioDuration":12.53,"audioTextHash":null,"seoTitle":null,"seoDescription":null,"metadata":{"bench_week":"2026-09-07","bench_status":"planned","bench_carried_from":null,"bench_why":null,"opens_record":"VM"}},{"id":"BE-38cc7282-5de","slug":"be-38cc7282-5de","zoneId":"core","parentId":"NOW","type":"update","kicker":"This week","title":"BLU paid media segments running without me","body":"","summary":"","sections":[],"mikeCommentary":"","stat":null,"statProvenance":null,"statAsOf":null,"state":"active","visibility":"public","geometry":{"x":null,"y":null,"w":0,"h":0},"createdAt":"2026-09-04T15:11:54.165Z","updatedAt":"2026-09-09T21:49:41.343Z","publishedAt":"2026-09-04T15:11:54.165Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/6WEq1C_vsMKk5pigUM4IYNGPaV8/txg__w../BE-38cc7282-5de.mp3","audioDuration":12.46,"audioTextHash":null,"seoTitle":null,"seoDescription":null,"metadata":{"bench_week":"2026-09-07","bench_status":"planned","bench_carried_from":null,"bench_why":null,"opens_record":"BLU"}},{"id":"BE-cef31489-cad","slug":"be-cef31489-cad","zoneId":"core","parentId":"NOW","type":"update","kicker":"This week","title":"HHP layout v0.5 — web, mobile, app","body":"","summary":"","sections":[],"mikeCommentary":"","stat":null,"statProvenance":null,"statAsOf":null,"state":"active","visibility":"public","geometry":{"x":null,"y":null,"w":0,"h":0},"createdAt":"2026-09-04T15:11:54.169Z","updatedAt":"2026-09-09T21:49:45.703Z","publishedAt":"2026-09-04T15:11:54.169Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/82iRMg6F3hfcN2RILg6oVKGpN1w/FarE4g../BE-cef31489-cad.mp3","audioDuration":12.84,"audioTextHash":null,"seoTitle":null,"seoDescription":null,"metadata":{"bench_week":"2026-09-07","bench_status":"planned","bench_carried_from":null,"bench_why":null,"opens_record":"HHP"}},{"id":"BE-a99005791b3b","slug":"be-a99005791b3b","zoneId":"core","parentId":"AVOID","type":"update","kicker":"Parked","title":"BLU as a character — visual test bench","body":"","summary":"","sections":[],"mikeCommentary":"","stat":null,"statProvenance":null,"statAsOf":null,"state":"stalled","visibility":"public","geometry":{"x":null,"y":null,"w":0,"h":0},"createdAt":"2026-09-05T13:01:10.048Z","updatedAt":"2026-09-09T21:49:52.399Z","publishedAt":"2026-09-05T13:01:10.048Z","audioUrl":"https://api.michaelpelland.com/vault/o8sFY-_X/IG8p9nUeKQJqwPtnwsegTGz3S_0/FSgbeg../BE-a99005791b3b.mp3","audioDuration":33,"audioTextHash":null,"seoTitle":null,"seoDescription":null,"metadata":{"bench_week":"2026-09-07","bench_status":"avoided","bench_carried_from":null,"bench_why":"Doing it properly is a rabbit hole. Character design, consistency, an expression system, animation, assets, implementation and refinement all eat meaningful time, and BLU’s underlying system is the priority. I asked some people for help and they could not figure it out. So it stays visible as an experiment worth returning to, and it goes back on the bench when I am ready to work it out myself.","opens_record":"X-BLUCHAR"}}],"relationships":[{"id":"BLU--depends_on--L-HEADLESS","fromRecordId":"BLU","toRecordId":"L-HEADLESS","type":"depends_on","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T20:24:48.676Z","confirmedBy":"ACC-MIKE","inferredWhy":"His own words, and about as unambiguous as a dependency gets: \"If we had not gone headless, none of what I just described would exist... Every interesting thing in this post is downstream of controlling our own stack.\"","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU--depends_on--L-PRIMITIVES","fromRecordId":"BLU","toRecordId":"L-PRIMITIVES","type":"depends_on","createdAt":"2026-09-06T11:59:39.014Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"\"The separation between them is the whole discipline\" and \"that changed the entire build\". Everything in BLU is built on the three primitives.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU--depends_on--L-RESOLVE","fromRecordId":"BLU","toRecordId":"L-RESOLVE","type":"depends_on","createdAt":"2026-09-06T11:59:38.683Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"Stated in the post as \"the most important architectural decision in the whole thing\". Same shape as BLU depends_on L-HEADLESS: remove it and BLU is a different system.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU--informed_by--L-ABSENT","fromRecordId":"BLU","toRecordId":"L-ABSENT","type":"informed_by","createdAt":"2026-09-06T11:59:39.755Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"Named in the briefing as the most expensive recurring defect in the BLU project. Same shape as BLU informed_by L-IDK.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU--informed_by--L-IDK","fromRecordId":"BLU","toRecordId":"L-IDK","type":"informed_by","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T20:24:47.989Z","confirmedBy":"ACC-MIKE","inferredWhy":"Stated outright in the source: \"That premise also set the hardest rule in the build.\" The rule shaped BLU rather than the other way round.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU--resulted_in--BLU-DEMOTION","fromRecordId":"BLU","toRecordId":"BLU-DEMOTION","type":"resulted_in","createdAt":"2026-09-05T23:44:51.747Z","metadata":{"confirmedAt":"2026-09-05T23:44:47.704Z","confirmedBy":"ACC-MIKE","inferredWhy":"Building BLU is what made the demotion possible: the renewal was killed because the business finally had somewhere else to live. Same shape as BLU resulted_in L-CHAIR.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU--resulted_in--L-CHAIR","fromRecordId":"BLU","toRecordId":"L-CHAIR","type":"resulted_in","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T20:24:48.429Z","confirmedBy":"ACC-MIKE","inferredWhy":"Causal and stated: the lesson came out of building the thing. \"Claude Code did not see that pattern before I did. I had to argue with it for a while.\"","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU--resulted_in--L-MODEL","fromRecordId":"BLU","toRecordId":"L-MODEL","type":"resulted_in","createdAt":"2026-09-05T22:43:55.836Z","metadata":{"confirmedAt":"2026-09-05T22:43:52.077Z","confirmedBy":"ACC-MIKE","inferredWhy":"The belief is drawn from building BLU and is stated that way: \"I am in the chair building this every day, and the more I build the more convinced I am.\" Same precedent as BLU resulted_in L-CHAIR.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU--resulted_in--N-ANSWERABLE","fromRecordId":"BLU","toRecordId":"N-ANSWERABLE","type":"resulted_in","createdAt":"2026-09-05T22:43:55.641Z","metadata":{"confirmedAt":"2026-09-05T22:43:52.077Z","confirmedBy":"ACC-MIKE","inferredWhy":"The questions are the ones BLU made answerable. The post introduces them as what came out of the build: \"I will tell you the kinds of questions we can answer now.\"","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU--resulted_in--N-REPORTS","fromRecordId":"BLU","toRecordId":"N-REPORTS","type":"resulted_in","createdAt":"2026-09-06T11:59:39.415Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"The reports were built inside BLU and the reversal happened there. Same precedent as BLU resulted_in N-SLOW and N-TWOREADS: the finding came out of the work.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU--resulted_in--N-SLOW","fromRecordId":"BLU","toRecordId":"N-SLOW","type":"resulted_in","createdAt":"2026-09-05T22:43:55.208Z","metadata":{"confirmedAt":"2026-09-05T22:43:52.077Z","confirmedBy":"ACC-MIKE","inferredWhy":"The regression is a direct cost of BLU resolving the business from the source systems on every read rather than keeping a copy. The post files it under what re-resolution costs, not as an unrelated performance bug.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU--resulted_in--N-TWOREADS","fromRecordId":"BLU","toRecordId":"N-TWOREADS","type":"resulted_in","createdAt":"2026-09-05T22:43:55.381Z","metadata":{"confirmedAt":"2026-09-05T22:43:52.077Z","confirmedBy":"ACC-MIKE","inferredWhy":"The two reads existed inside BLU and the decision to delete the fast one was made while building it. Same shape as BLU resulted_in L-CHAIR: the finding came out of the work.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU-DEMOTION--informed_by--BLU-ORIGIN","fromRecordId":"BLU-DEMOTION","toRecordId":"BLU-ORIGIN","type":"informed_by","createdAt":"2026-09-05T23:44:51.973Z","metadata":{"confirmedAt":"2026-09-05T23:44:47.704Z","confirmedBy":"ACC-MIKE","inferredWhy":"The demotion is the answer to the question the boat asked. BLU-ORIGIN ends on \"why don’t we just build it? So I came home and started\"; this is what starting led to.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU-ONEOS--depends_on--BLU","fromRecordId":"BLU-ONEOS","toRecordId":"BLU","type":"depends_on","createdAt":"2026-09-05T22:43:56.155Z","metadata":{"confirmedAt":"2026-09-05T22:43:52.077Z","confirmedBy":"ACC-MIKE","inferredWhy":"The convergence has nothing to converge without BLU: the post frames it as what happens when the ERP and BLU are wired into the same reality, and BLU is the half still being built.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU-ONEOS--informed_by--BLU-ORIGIN","fromRecordId":"BLU-ONEOS","toRecordId":"BLU-ORIGIN","type":"informed_by","createdAt":"2026-09-05T22:43:56.261Z","metadata":{"confirmedAt":"2026-09-05T22:43:52.077Z","confirmedBy":"ACC-MIKE","inferredWhy":"The ERP Hunter built is named and described in the origin note, and it is the thing that made the whole argument possible: software built to fit the business rather than the other way round.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU-ORIGIN--resulted_in--BLU","fromRecordId":"BLU-ORIGIN","toRecordId":"BLU","type":"resulted_in","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T00:00:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"Stated outright at the end of the origin note: the questions in the boat led to \"why don’t we just build it? So I came home and started.\" BLU is what he started.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"BLU-ROLES--related_to--BLU-LOOP","fromRecordId":"BLU-ROLES","toRecordId":"BLU-LOOP","type":"related_to","createdAt":"2026-09-06T11:59:39.919Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"The roles are defined by their place in the loop: \"split by role in the loop, not by business domain\". Neither record reads without the other.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"C-UGLY--depends_on--BLU","fromRecordId":"C-UGLY","toRecordId":"BLU","type":"depends_on","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T00:00:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"The offer is to ask BLU the Simpli Soda version of the visitor’s question, live, against real systems. Without BLU there is nothing to ask.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"C-UGLY--informed_by--L-IDK","fromRecordId":"C-UGLY","toRecordId":"L-IDK","type":"informed_by","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T00:00:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"The challenge describes a question \"you still do not fully trust when it lands on your desk\", which is precisely what the grounding rule is for. Conceptual — he does not connect them in the piece, and it is my reading that the offer only means anything because of that rule.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"medium"}},{"id":"C-UGLY--related_to--N-ANSWERABLE","fromRecordId":"C-UGLY","toRecordId":"N-ANSWERABLE","type":"related_to","createdAt":"2026-09-05T22:43:55.741Z","metadata":{"confirmedAt":"2026-09-05T22:43:52.077Z","confirmedBy":"ACC-MIKE","inferredWhy":"C-UGLY invites the question that crosses systems and that nobody trusts when it lands. N-ANSWERABLE lists the kinds of question that used to be exactly that and now are not. Two sides of one offer.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"medium"}},{"id":"D-VOICE--depends_on--VM","fromRecordId":"D-VOICE","toRecordId":"VM","type":"depends_on","createdAt":"2026-09-04T13:26:24.787Z","metadata":{"confirmedAt":"2026-09-04T13:26:24.787Z","confirmedBy":"mike","inferredWhy":"The public voice in this decision is VM itself.","confirmedVia":"session-instruction","inferredConfidence":"medium"}},{"id":"HHP-DESIGN--depends_on--VM","fromRecordId":"HHP-DESIGN","toRecordId":"VM","type":"depends_on","createdAt":"2026-09-06T12:23:29.381Z","metadata":{"confirmedAt":"2026-09-06T12:23:22.578Z","confirmedBy":"ACC-MIKE","inferredWhy":"Stated as the reason by Mikey: he can work on three things at once because Virtual Mike begins to pick up load.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"HHP-DESIGN--supersedes--HHP-GATE","fromRecordId":"HHP-DESIGN","toRecordId":"HHP-GATE","type":"supersedes","createdAt":"2026-09-06T12:23:29.199Z","metadata":{"confirmedAt":"2026-09-06T12:23:22.578Z","confirmedBy":"ACC-MIKE","inferredWhy":"The decision to start on 4 September replaced the rule that HHP waits for BLU. Mikey chose supersedes so the board shows how the thinking changed rather than editing the rule away.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"HHP-GATE--depends_on--BLU","fromRecordId":"HHP-GATE","toRecordId":"BLU","type":"depends_on","createdAt":"2026-09-06T12:23:29.279Z","metadata":{"confirmedAt":"2026-09-06T12:23:22.578Z","confirmedBy":"ACC-MIKE","inferredWhy":"The rule was stated in terms of BLU: this starts when BLU stands on its own.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"L-ABSENT--related_to--L-IDK","fromRecordId":"L-ABSENT","toRecordId":"L-IDK","type":"related_to","createdAt":"2026-09-06T11:59:39.668Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"Drafter-proposed and stated: both hold that a system must report the absence of an answer rather than substitute one. L-IDK is the refusal; this is the taxonomy of what is being refused.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"L-CHAIR--informed_by--N-REPORTS","fromRecordId":"L-CHAIR","toRecordId":"N-REPORTS","type":"informed_by","createdAt":"2026-09-06T11:59:39.576Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"The chair argument is told from this episode: \"Claude Code didn’t see that pattern before I did... It could not tell me I was specifying the wrong thing.\"","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"L-HEADLESS--informed_by--BLU-ORIGIN","fromRecordId":"L-HEADLESS","toRecordId":"BLU-ORIGIN","type":"informed_by","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T00:00:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"The origin note asks why the business was \"getting flattened into contacts and companies, when what we actually have is households, machines and cylinders\". That is the position stated as an experience. Direction runs experience to position.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"L-MODEL--related_to--L-CHAIR","fromRecordId":"L-MODEL","toRecordId":"L-CHAIR","type":"related_to","createdAt":"2026-09-05T22:43:55.931Z","metadata":{"confirmedAt":"2026-09-05T22:43:52.077Z","confirmedBy":"ACC-MIKE","inferredWhy":"Both name the scarce input and agree on what it is not. L-CHAIR says the code is the commodity and the judgement is not; L-MODEL says the model is not the hard part and modelling the business is.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"L-MODEL--related_to--L-IDK","fromRecordId":"L-MODEL","toRecordId":"L-IDK","type":"related_to","createdAt":"2026-09-05T22:43:56.046Z","metadata":{"confirmedAt":"2026-09-05T22:43:52.077Z","confirmedBy":"ACC-MIKE","inferredWhy":"Both are about the floor under a number: L-IDK refuses to display one that is not fact, L-MODEL says the reasoning is only worth anything once the business underneath is modelled correctly. The weakest edge in this batch and the first to cut.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"medium"}},{"id":"L-PRIMITIVES--related_to--L-IDK","fromRecordId":"L-PRIMITIVES","toRecordId":"L-IDK","type":"related_to","createdAt":"2026-09-06T11:59:39.170Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"Only a measure renders as a count is the same rule as no hallucinated number wearing a metric’s clothing, stated structurally rather than as a refusal.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"L-PRIMITIVES--related_to--L-MODEL","fromRecordId":"L-PRIMITIVES","toRecordId":"L-MODEL","type":"related_to","createdAt":"2026-09-06T11:59:39.253Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"Drafter-proposed: both hold that the model underneath is built before reasoning is allowed on top. The empty derived slot is that order enforced.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"medium"}},{"id":"L-RESOLVE--informed_by--BLU-ORIGIN","fromRecordId":"L-RESOLVE","toRecordId":"BLU-ORIGIN","type":"informed_by","createdAt":"2026-09-06T11:59:38.763Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"The question that decided the build was \"paying rent on a copy of our own data and calling it a system of record\". This record is the answer to that question.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"L-RESOLVE--related_to--N-TWOREADS","fromRecordId":"L-RESOLVE","toRecordId":"N-TWOREADS","type":"related_to","createdAt":"2026-09-06T11:59:38.934Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"Drafter-proposed and stated in the source: two reads that can disagree are two versions of reality, the thing this principle exists to prevent.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"medium"}},{"id":"L-RESOLVE--resulted_in--N-SLOW","fromRecordId":"L-RESOLVE","toRecordId":"N-SLOW","type":"resulted_in","createdAt":"2026-09-06T11:59:38.855Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"The post files the slowdown under what re-resolution costs. Asking the source again on every read is why the passes multiplied.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"L-SAAS--informed_by--L-HEADLESS","fromRecordId":"L-SAAS","toRecordId":"L-HEADLESS","type":"informed_by","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T20:24:48.881Z","confirmedBy":"ACC-MIKE","inferredWhy":"The thesis rests on building your own no longer being irrational, and owning the data model is the specific form that takes. Conceptual rather than stated — he argues both in one piece without claiming one caused the other.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"medium"}},{"id":"L-SAAS--related_to--L-CHAIR","fromRecordId":"L-SAAS","toRecordId":"L-CHAIR","type":"related_to","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T20:24:49.092Z","confirmedBy":"ACC-MIKE","inferredWhy":"The same claim at two scales. \"Engineering capacity used to be the constraint. Now it is domain judgement\" is the SaaS thesis restating what the chair lesson says about one person. Explicitly not causal.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"medium"}},{"id":"L-SAAS--validated_by--BLU-DEMOTION","fromRecordId":"L-SAAS","toRecordId":"BLU-DEMOTION","type":"validated_by","createdAt":"2026-09-05T23:44:52.060Z","metadata":{"confirmedAt":"2026-09-05T23:44:47.704Z","confirmedBy":"ACC-MIKE","inferredWhy":"L-SAAS argues the horizontal middle dies by demotion rather than bankruptcy: \"the platform does not die, it gets moved down a tier and paid like a utility.\" This is that thesis happening to a named platform on a known date.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"N-904--related_to--NOW","fromRecordId":"N-904","toRecordId":"NOW","type":"related_to","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T00:00:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"The time-axis link. NOW answers what is on the bench now and cannot answer what was on it in September; this note is that week and points back at the index it is no longer part of.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"N-904--related_to--VM","fromRecordId":"N-904","toRecordId":"VM","type":"related_to","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T00:00:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"The week was spent building this board, so the note and VM are the same subject at two scales. Structural rather than causal — VM did not cause the note and the note did not cause VM.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"N-904--resulted_in--L-DOCS","fromRecordId":"N-904","toRecordId":"L-DOCS","type":"resulted_in","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T19:04:54.405Z","confirmedBy":"ACC-MIKE","inferredWhy":"Genuinely causal and the reason both records exist. The week produced the lesson; without the throwaway endpoint there is nothing to generalise from.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"N-REPORTS--resulted_in--L-PRIMITIVES","fromRecordId":"N-REPORTS","toRecordId":"L-PRIMITIVES","type":"resulted_in","createdAt":"2026-09-06T11:59:39.091Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"Backing up three steps from the reports is what produced the primitives; the post tells them as one movement on one morning.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"N-SLOW--related_to--N-TWOREADS","fromRecordId":"N-SLOW","toRecordId":"N-TWOREADS","type":"related_to","createdAt":"2026-09-06T11:52:29.876Z","metadata":{"confirmedAt":"2026-09-06T11:52:25.338Z","confirmedBy":"ACC-MIKE","inferredWhy":"Two costs of re-resolution on consecutive days, 13-14 and 14 August, told together in the post under what re-resolution costs. One lesson in two entries.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"N-TWOREADS--informed_by--L-IDK","fromRecordId":"N-TWOREADS","toRecordId":"L-IDK","type":"informed_by","createdAt":"2026-09-05T22:43:55.540Z","metadata":{"confirmedAt":"2026-09-05T22:43:52.077Z","confirmedBy":"ACC-MIKE","inferredWhy":"Stated as the reason in the source: a system where two views of the same number can disagree is \"precisely the thing I was trying to get away from\". That thing is the grounding rule L-IDK states.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"NOW--related_to--AVOID","fromRecordId":"NOW","toRecordId":"AVOID","type":"related_to","createdAt":"2026-09-02T14:50:00.000Z","metadata":{"confirmedAt":"2026-09-02T14:50:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"The two halves of an honest status: what is being worked on and what is being dodged. Tempting to call this contradicts, but they are complements, not a conflict.","confirmedVia":"prototype","directionFlipped":false,"inferredConfidence":"medium"}},{"id":"NOW--related_to--BLU","fromRecordId":"NOW","toRecordId":"BLU","type":"related_to","createdAt":"2026-09-02T14:50:00.000Z","metadata":{"confirmedAt":"2026-09-02T14:50:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"Hub spoke — NOW indexes the live project.","confirmedVia":"prototype","directionFlipped":false,"inferredConfidence":"high"}},{"id":"NOW--related_to--HHP","fromRecordId":"NOW","toRecordId":"HHP","type":"related_to","createdAt":"2026-09-02T14:50:00.000Z","metadata":{"confirmedAt":"2026-09-02T14:50:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"Hub spoke. NOW indexes the unstarted project too, which is the point of listing it — unstarted things belong on the board.","confirmedVia":"prototype","directionFlipped":false,"inferredConfidence":"high"}},{"id":"NOW--resulted_in--R-SHIP","fromRecordId":"NOW","toRecordId":"R-SHIP","type":"resulted_in","createdAt":"2026-09-02T14:50:00.000Z","metadata":{"confirmedAt":"2026-09-02T14:50:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"\"Changes noticed\" counts what the bench work produced, so this spoke does carry causation. Direction runs work to outcome.","confirmedVia":"prototype","directionFlipped":false,"inferredConfidence":"medium"}},{"id":"R-AUTO--related_to--R-BACK","fromRecordId":"R-AUTO","toRecordId":"R-BACK","type":"related_to","createdAt":"2026-09-02T14:50:00.000Z","metadata":{"confirmedAt":"2026-09-02T14:50:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"Same board, no causality.","confirmedVia":"prototype","directionFlipped":false,"inferredConfidence":"high"}},{"id":"R-BACK--validated_by--BLU-DEMOTION","fromRecordId":"R-BACK","toRecordId":"BLU-DEMOTION","type":"validated_by","createdAt":"2026-09-06T02:10:49.062Z","metadata":{"confirmedAt":"2026-09-06T02:10:45.338Z","confirmedBy":"ACC-MIKE","inferredWhy":"The 27 August reversal counted by this measure. Killing the HubSpot renewal after years of recommending and defending it is the reversal, and the record carries what changed his mind.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"R-BACK--validated_by--HHP-DESIGN","fromRecordId":"R-BACK","toRecordId":"HHP-DESIGN","type":"validated_by","createdAt":"2026-09-06T12:23:29.476Z","metadata":{"confirmedAt":"2026-09-06T12:23:22.578Z","confirmedBy":"ACC-MIKE","inferredWhy":"A decision taken back, dated 2026-09-04, counted on R-BACK. Same shape as the three reversals already wired.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"R-BACK--validated_by--N-REPORTS","fromRecordId":"R-BACK","toRecordId":"N-REPORTS","type":"validated_by","createdAt":"2026-09-06T11:59:39.496Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"The Aug 6 reversal R-BACK counts. Replaces the placeholder edge to BLU, which existed only because this record did not.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"R-BACK--validated_by--N-TWOREADS","fromRecordId":"R-BACK","toRecordId":"N-TWOREADS","type":"validated_by","createdAt":"2026-09-06T02:10:49.263Z","metadata":{"confirmedAt":"2026-09-06T02:10:45.338Z","confirmedBy":"ACC-MIKE","inferredWhy":"The 14 August reversal counted by this measure. A shipped 2x speedup deleted because two reads of one number could disagree.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"R-SHIP--related_to--R-AUTO","fromRecordId":"R-SHIP","toRecordId":"R-AUTO","type":"related_to","createdAt":"2026-09-02T14:50:00.000Z","metadata":{"confirmedAt":"2026-09-02T14:50:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"Two metrics side by side on the results board. Neither causes the other.","confirmedVia":"prototype","directionFlipped":false,"inferredConfidence":"high"}},{"id":"VM--evolved_into--BLU","fromRecordId":"VM","toRecordId":"BLU","type":"evolved_into","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T00:00:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"His own sentence: \"At Simpli Soda, Virtual Mike became BLU.\" The premise predates both — this board is that premise in public, BLU is the same premise pointed at a company. Direction runs idea to instance.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"VM--informed_by--L-ABSENT","fromRecordId":"VM","toRecordId":"L-ABSENT","type":"informed_by","createdAt":"2026-09-06T11:59:39.836Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"ADR 014 on this board: an unset enum arrived as \"\" — neither null nor valid — and rendered as a count. Absent indistinguishable from zero, on this site.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"medium"}},{"id":"VM--informed_by--L-IDK","fromRecordId":"VM","toRecordId":"L-IDK","type":"informed_by","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T20:24:48.208Z","confirmedBy":"ACC-MIKE","inferredWhy":"The same rule governs this board. Virtual Mike answers only from retrieved records, says plainly when Mike has not written something down, and as of ADR 014 refuses to render an uncounted number as a count. One rule, two systems.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"VM--informed_by--L-PRIMITIVES","fromRecordId":"VM","toRecordId":"L-PRIMITIVES","type":"informed_by","createdAt":"2026-09-06T11:59:39.329Z","metadata":{"confirmedAt":"2026-09-06T11:59:27.419Z","confirmedBy":"ACC-MIKE","inferredWhy":"This board runs the same rule (ADR 014): every stat declares derived, counted or claimed, and only a measure renders as a count. The record says so in its last section.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"VM--related_to--NOW","fromRecordId":"VM","toRecordId":"NOW","type":"related_to","createdAt":"2026-09-02T14:50:00.000Z","metadata":{"confirmedAt":"2026-09-02T14:50:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"Structural. VM is the board itself and NOW is its entry point; neither causes the other.","confirmedVia":"prototype","directionFlipped":false,"inferredConfidence":"high"}},{"id":"VM--related_to--SAY","fromRecordId":"VM","toRecordId":"SAY","type":"related_to","createdAt":"2026-09-02T14:50:00.000Z","metadata":{"confirmedAt":"2026-09-02T14:50:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"The board and its contact record.","confirmedVia":"prototype","directionFlipped":false,"inferredConfidence":"high"}},{"id":"VM-PREMISE--resulted_in--L-IDK","fromRecordId":"VM-PREMISE","toRecordId":"L-IDK","type":"resulted_in","createdAt":"2026-09-06T11:52:29.798Z","metadata":{"confirmedAt":"2026-09-06T11:52:25.338Z","confirmedBy":"ACC-MIKE","inferredWhy":"Stated in the post immediately after the premise: \"That premise also set the hardest rule in the build.\" The rule is what L-IDK records.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}},{"id":"X-BLUCHAR--depends_on--BLU","fromRecordId":"X-BLUCHAR","toRecordId":"BLU","type":"depends_on","createdAt":"2026-09-04T00:00:00.000Z","metadata":{"confirmedAt":"2026-09-04T00:00:00.000Z","confirmedBy":"ACC-MIKE","inferredWhy":"Stated. The character is a layer over \"the intelligence that already exists underneath it\", and the parking reason is that BLU’s underlying system comes first.","confirmedVia":"session-instruction","directionFlipped":false,"inferredConfidence":"high"}}],"zones":[{"id":"core","slug":"core","label":"CORE","glyph":"▣","sort":0,"fallback":{"h":560,"w":380,"x":1200,"y":560},"axis":"column","flush":true},{"id":"blu","slug":"blu","label":"BLU","glyph":"◫","sort":1,"fallback":{"h":430,"w":660,"x":290,"y":470},"axis":"fan","flush":false},{"id":"hhp","slug":"happi-home-pro","label":"HAPPI HOME PRO","glyph":"◇","sort":2,"fallback":{"h":400,"w":660,"x":310,"y":1130},"axis":"fan","flush":false},{"id":"results","slug":"results","label":"RESULTS","glyph":"▤","sort":3,"fallback":{"h":250,"w":780,"x":1140,"y":250},"axis":"row","flush":false},{"id":"exp","slug":"experiments","label":"EXPERIMENTS","glyph":"◈","sort":4,"fallback":{"h":330,"w":640,"x":1970,"y":550},"axis":"fan","flush":false},{"id":"notes","slug":"field-notes","label":"FIELD NOTES","glyph":"≡","sort":5,"fallback":{"h":410,"w":620,"x":1770,"y":960},"axis":"fan","flush":false},{"id":"learn","slug":"what-holds-up","label":"WHAT HOLDS UP","glyph":"⌸","sort":6,"fallback":{"h":290,"w":620,"x":1090,"y":1280},"axis":"fan","flush":false},{"id":"inputs","slug":"inputs","label":"INPUTS","glyph":"⌁","sort":7,"fallback":{"h":220,"w":590,"x":520,"y":1550},"axis":"fan","flush":false},{"id":"contact","slug":"contact","label":"CONTACT","glyph":"✦","sort":8,"fallback":{"h":230,"w":370,"x":520,"y":130},"axis":"fan","flush":false}],"generatedAt":"2026-09-10T13:40:36.389Z"},"children":"$L2b"}]