Part 3 · Operate
Triage and synthesis
Two stages: a cheap filter that kills most events, and an expensive conclusion for the rest.
01What it is
Every event is triaged on the cheap model tier and most die there. What survives, and only when enough of it lands inside the window, is synthesised into a read on the strong tier.
02Why it exists
Running the expensive stage on every event would cost more than the channel returns. Most events at most companies mean nothing, and deciding that is a cheap judgement.
The two-stage shape is also why reads are rare. A read that fires weekly on every company is a read nobody believes. The thresholds are set so that a read means something happened.
03How to use it
- 01Nothing to do. Both stages run on a schedule.
- 02Read the timeline on a read before the draft. The spine shows which events carried it.
- 03Mark a read wrong when you disagree. That feeds allocation.
04What good looks like
- You give it
- Three meaningful events at one company inside ninety days, on one theme.
- You get
- A read with a dated spine, a confidence figure and a verdict.
- Why
- Enough meaningful events inside the synthesis window, and the confidence clears the floor to publish.
Derived from the rules in the code
05What weak looks like, and what it costs
- You give it
- Two meaningful events four months apart.
- You get
- No read. Both stay in the archive.
- Why
- The synthesis window is ninety days, and recency decay has halved the older one’s contribution before that.
Derived from the rules in the code