Wins curation — worked example

The scenario

It’s mid-Q3. You’re an AI PM. You’ve been logging wins since the start of Q2 using wins-log, nudged by the Friday wins-due schedule. Promo cycle is in 8 weeks. You’ve never had a clean promo packet before — usually it’s a frantic two-hour scramble the night before the deadline.

What’s in the log (8 wins over 14 weeks)

A mix of types:

2026-04-12 — Killed the LLM reranker proposal before engineering started
  Type: judgment · Scope: quarter
  Impact: avoided ~6 eng-weeks of work on a problem that was data quality, not intelligence

2026-04-28 — Onboarding completion +14pp in 6 weeks
  Type: delivery · Scope: quarter
  Solo vs. shared: led the experiment design; EM and 2 devs did execution

2026-05-09 — Coached Yulia through her first stakeholder presentation
  Type: mentorship · Scope: sprint
  Impact: she ran it solo, exec sponsor called it "unusually clear for an intern"

2026-05-23 — Recovered launch after CS reported wrong output
  Type: recovery · Scope: sprint
  Impact: 48-hour incident-to-resolution; no churn; built eval harness that caught 3 more cases

... 4 more entries

I have a promo cycle in 8 weeks. Help me build a promo packet.

wins-curate in promo-packet mode:

Reads the full log
Maps wins to the dimensions a panel cares about (scope, impact, judgment, leadership, growth)
Surfaces gaps before you’re surprised by them

What it produces:

Strengths from the log:

Judgment (strong): The LLM reranker kill is the clearest judgment entry. Scope is quarter-level. Impact is quantified. Solo credit is honest. Panel-ready.
Delivery (strong): The onboarding +14pp. Well-framed solo/shared credit. Could use a stronger baseline (what was completion rate before?).
Recovery (solid): The 48-hour incident recovery. Needs the scale added — how many customers affected? What was the revenue risk?
Mentorship (light): One entry. A panel can accept one; more would be better if you have more. The Yulia entry is specific and outcome-cited, which helps.

Gaps:

Leadership: No entries. For the PM→Senior PM bar, leadership (influencing without authority, driving cross-team alignment) is often a gap. Do you have examples not in the log? If so, they should be added before submission.
Range: No entries. If you’ve done work outside your core PM scope (a strategy doc, a technical design review, a hiring loop), log it now — that’s range.

Pre-empted objections:

“The onboarding metric could be seasonal.” The log entry doesn’t address this. Add context: was Q2 a similar period last year? What controlled for seasonality?
“The LLM kill is judgment — what did you build?” Useful to have a delivery win from this same period that’s clearly yours.

Step 2: Log the gaps before submission

Armed with the gap analysis, you run wins-log twice more:

I helped drive the architecture decision for the contract-summarization eval harness
→ wins-log: Type: judgment + craft. You push back on "helped drive" — what specifically did you do?

The skill pushes until you articulate: “I proposed the golden-dataset structure, ran the first labeling session, and got engineering to agree to a 200-case minimum before ship.” That goes in the entry — specific, evidenced.

I ran the loop design for the intern hire last month
→ wins-log: Type: leadership + range. Scope: sprint. You write the signal map and the hiring rubric contribution.

Two new entries. Leadership gap partially filled. Range gap filled.

Step 3: Curate again with the fuller log

Run the promo packet again with the new entries.

Updated assessment: leadership is now “light but present,” range is covered, delivery and judgment are strong. The skill produces a citation-backed packet ready for the panel, with specific win references per section and the calibration notes already written in.

What made this work

Logging in-week. The incident recovery entry had the 48-hour timeline, the customer impact estimate, and the Slack thread as evidence — because it was logged that week. A retroactive entry wouldn’t have that.
The Friday nudge. The mentorship entry came from wins-due probing “any mentorship this week?” on a Friday when you would have said nothing notable. Yulia’s presentation had been that Thursday.
The gap analysis. Knowing eight weeks out that leadership was thin gave you time to log existing work and identify the intern hiring contribution you’d forgotten about.
Calibration. The onboarding metric entry was honestly framed as shared credit from the start. The panel question (“was this seasonal?”) was surfaced before the panel saw it — not after.