CultivateOSwinnow refines scattered evidence
into cited recommendations.
Sales calls, decision memos, research transcripts, support tickets. Winnow ingests them, ties each fact to its source, and produces recommendations your team can hand to an AI builder without writing a brief.
Three steps, no theatre.
Winnow is a wiki that maintains itself. You feed it sources; it keeps the foundations up to date and produces recommendations cited back to the lines that justify them.
Sources come in.
Connect HubSpot, Notion, Zendesk, Linear. Or drop files directly. Every source keeps its origin and timestamp; nothing is laundered.
> winnow ingest acme-q1-call.md
✓ stored as [s/2025-q1/acme-call-7]
✓ 14 lines of evidence
✓ 3 hypotheses generated
The wiki updates itself.
New evidence revises the foundations — product proposition, target markets, constraints. Changes propagate; old hypotheses get demoted or refuted as the picture sharpens.
foundations/product_proposition → updated (3 new sources cited) → hypothesis demoted: [h/price-sensitivity] → state: refuted
Recommendations, cited.
Ask the wiki a question; get back a recommendation that links every claim to its source. Hand it to Claude Code, Cursor, Lovable — your AI builder has full strategic context.
> winnow recommend "fix cold-start"
✓ 1 recommendation, 3 sources
✓ exported to [r/2025-04/cold-start]
✓ MCP-ready
Every line carries its provenance.
Winnow tracks three states — and the transitions between them. Validated evidence becomes hypothesis again when challenged. Hypotheses become validated when sources back them up. Refuted beliefs stay refuted, struck through but kept on the page so the reasoning is visible.
The cold-start onboarding produces a usable evidence pool inside ten minutes, and the wiki compiles without further prompting [s/2025-q1/acme-call-7].
A hypothesis we tracked early — that buyers wouldn’t tolerate pricing above $200 — was refuted by four deals closed at $499 with no price objection [h/price-sensitivity].
The trail back to the source line is one click away. Always.
“The cold-start onboarding kills us — we ship evidence to the team but they have to spend an hour reading before they can act.”
If the wiki feels like a library, senior PMs spend longer reading and return more often.
Buyers won’t tolerate pricing above $200.
Winnow is the part of the cycle that refines.
It sits between the gathering and the building. Bushel takes Winnow’s recommendations and turns them into the nine artefacts a launch needs. The forthcoming Build agent takes them straight to an AI builder with full strategic context attached.
Lay the foundations.
Ingest sources.
Synthesise into cited recommendations.
Hand to an AI builder.
Launch artefacts.
Did it land?
Get a workspace. Bring your evidence.
Free for solo PMs evaluating the methodology. Pro for working teams. We onboard new workspaces in small batches.