/ field notes
CORE
BLU
HAPPI HOME PRO
RESULTS
EXPERIMENTS
FIELD NOTES
WHAT HOLDS UP
INPUTS
CONTACT
LOOP
The product is a loop
It starts when two numbers disagree. Reports fall out along the way.
ARCHITECTURE
Split by role, not domain
Lens, brain, guard, ledger. Four roles, no domain modules.
VOICE
Two voices, one auth
Leo never lives outside my auth, so the site hears VM and I hear Leo.
SEP 4
Three things at once
The repo went live before BLU stood. Virtual Mike picks up the load.
WAS THE RULE
Starts when BLU stands
Two unfinished things is one too many. That was the rule.
LiveVM
Virtual Mike
The interface you are using. Not a chatbot on a site — the site.
2 INSIDE
VIRTUAL MIKE
The reflex is the product
The reflex is worth virtualizing. The knowledge is not.
ProjectBLU
BLU
Virtual Mike, running on Simpli Soda's stack.
2 INSIDE
HubNOW
On the bench this week
Five buckets. Two are BLU, one is the thing you are reading.
5 INSIDE
HonestAVOID
What I am avoiding
Naming the price. Dressed up as a design problem. It is not.
1 INSIDE
DesigningHHP
Happi Home Pro
Designed before built. No app code yet. Listed anyway.
2 INSIDE
14 daysR-SHIP
0
Changes noticed
Somebody outside my head noticed. That is the bar.
UnattendedR-AUTO
0
Running without me
Two weeks unattended, or it does not count.
ReversedR-BACK
4
Decisions taken back
Counted, because pretending they were right is expensive.
OpenSAY
Say something
mike@michaelpelland.com — partners, builders, people comparing notes.
ParkedX-BLUCHAR
Manifest BLU as an actual character
A visual and personality layer for the intelligence that already exists underneath it.
OriginBLU-ORIGIN
Why don’t we just build it
A boat, a flask, and a question nobody had a good answer to.
ChallengeC-UGLY
Come tell me I am wrong
Bring the ugly question. I will ask BLU the Simpli Soda version of it, live.
GroundingL-IDK
If it cannot say I do not know
Then every number it hands you is unverified by definition.
JudgmentL-CHAIR
The IP is the guy in the chair
The tool could build anything I could specify. It could not tell me I was specifying the wrong thing.
HeadlessL-HEADLESS
Not going headless is the mistake
Own the data model your business runs on, or rent somebody else’s definition of a customer.
SaaSL-SAAS
SaaS is dead
The horizontal middle dies. Infrastructure lives.
Sep 4N-904
Three designs, two limits
Built the ask, designed the login. The platform reset both.
PlatformL-DOCS
The docs are a claim
Their own documented example, rejected by their own server.
ArchitectureL-RESOLVE
The copy is not truth
Most systems operate off a copy. BLU asks the source again.
PrimitivesL-PRIMITIVES
Everything is a segment
Segment, measure, derived claim. Only two may render as a count.
Aug 27BLU-DEMOTION
The CRM was demoted
Platform to sensor. The plan was to remove it. It tells me more now.
ConvergenceBLU-ONEOS
One operating system, not two
Operations and demand, wired into the same reality.
AbsenceL-ABSENT
Absent is not zero
Zero, empty, unasked, unaskable. Four answers, not one.
ModelingL-MODEL
Modeling comes before intelligence
The model is not the hard part. Modeling the business is.
Aug 6N-REPORTS
Reports were segments
Every report I built was a segment. So I built segments.
Aug 13–14N-SLOW
The instrument went stale
Three BigQuery passes became eight. Nobody noticed for ten days.
Aug 14N-TWOREADS
Two reads can disagree
Deleted a real 2x speedup, because two paths can disagree.
StandingN-ANSWERABLE
The questions became answerable
The inputs always existed. Nobody owned the disagreement.
Focus modeAug 14 · FIELD NOTESCLOSE ✕

Two reads can disagree

Deleted a real 2x speedup, because two paths can disagree.

HEAR ITIN MIKE’S VOICE0:00 / 0:43

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.

I removed the fastest thing I had built that week and nobody will ever know.

THE SETUP

Two paths, one number

The summary card used a cheap read. The report used the full one. Measured against each other, roughly a 2x speedup.

THE PROBLEM

Divergence, not latency

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.

THE CHANGE

Deleted the fast one

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.

Published Updated N-TWOREADS
ASK