This is what python engine/pipeline.py prints out of the box, with no model and no
API key, reading engine/example_activity.json. It is here so you can see the whole loop
before you run it. Every step below saw a small assembled window, never the whole knowledge base.
Read 5 raw activity items. Extracting signals...
======================================================================
RANKED SHORTLIST (drafted unattended, waiting for a human yes)
======================================================================
[75] Northwind | why now: Northwind posted a job: Growth Ops Manager
Saw the Growth Ops role went up right after you stepped in. Usually that hire
spends month one firefighting attribution instead of building, so we take that
half off the plate for teams your size. Worth a look, or too early?
[60] Acme Freight | why now: Acme Freight raised a $12M Series A
Saw the raise. The teams that put that money to work fastest usually hit the
next quarter's number a month early. Happy to share how, if it is useful.
[60] Bluebird | why now: Bluebird VP Marketing posted about hating their current ABM tool
Saw you are moving off your current setup. The gap most teams hit on the way out
is the one we were built to close. Worth a quick look, or too soon?
Two things to notice.
Quiet Co never appears. Its activity (a culture blog post) produced no signal, so it was never even scored. No signal, no touch. That is the system working, not a gap.
Northwind ranks first because two signals stacked in one week (a new Head of Growth and a Growth Ops req), and the opener says so. A stacked moment is a stronger opening than either fact.
Then the weekly refine pass runs and rewrites the context from outcomes:
{
"promote": ["Saw the Growth Ops role went up right after you stepped in. ..."],
"reweight": {"funding.raise": "kept going cold, read later in the cycle"},
"note": "1 promoted, 1 reweighted"
}
The winning line gets promoted into the proof examples, and the funding signal that kept going cold gets reweighted so the next window reads it later. Nobody touched the model. The context moved.