PM Decision Archaeology Worksheet

Reconstruct how the last 8 to 12 significant decisions on your product actually got made. Not how the wiki says decisions get made. How they got made. The new PM who runs this dig learns the real operating model in two weeks, which is the thing that otherwise takes a year of bruises.

Run it in weeks two through four. Budget: 4 to 6 hours of digging plus 3 to 4 short conversations.

---

Part 1: The dig sites

Where decisions leave fossils. Work through these in order, cheapest first.

[ ] Merged PRs and their descriptions. Filter the repo to the biggest merges of
    the last two quarters. The PR description and review thread often contain
    the real debate the planning doc sanitized.
[ ] Launch posts and changelogs. Everything announced in the last year. Each
    launch was a decision. Note which ones had a stated "why" and which were
    pure "what."
[ ] Killed or stalled projects. Ask three people: "What got started in the last
    year and quietly stopped?" Kills are the highest-information decisions an
    org makes, and the least documented.
[ ] Escalations. Search Slack/email for threads where a decision went up a
    level. Who escalated, to whom, what came back down. This maps the real
    authority lines.
[ ] Pricing and packaging changes. Every price or plan change in two years.
    These decisions cross the most functions, so they expose the most about
    how the org negotiates.
[ ] Roadmap diffs. If old roadmap versions exist, diff this quarter's against
    two quarters ago. Everything that silently disappeared is a decision
    someone made without announcing it.

Pick 8 to 12 decisions total, mixed across sites. Include at least two kills and at least one escalation.

Part 2: The five questions per decision

For each decision, answer these five. From artifacts first, conversations second.

1. Who raised it? Where did the idea or the problem enter? A customer, an exec, an engineer, a competitor move. Note the entry point, orgs have habitual ones.
2. Where was it debated? A doc with comments, a meeting, a Slack thread, nowhere. "Nowhere" is a common and important answer.
3. Who actually decided? Not who signed off. Whose words show up in the final shape. Sometimes the decider is a staff engineer or a single large customer, and the org chart never mentions them.
4. What evidence was in the room? Data, customer quotes, a prototype, a vibe, an exec preference. Score it: rich / thin / none.
5. Did it stick? Was it relitigated, reversed, or quietly ignored? A decision that did not stick tells you where authority is contested.

Part 3: The timeline template

One row per decision. The patterns live in the columns, not the rows.

| # | Decision | Date | Raised by | Debated where | Real decider | Evidence (rich/thin/none) | Time to decide | Stuck? | Notes |
|---|----------|------|-----------|---------------|--------------|---------------------------|----------------|--------|-------|
| 1 | [One line: "Killed the v1 reporting module"] | [approx] | [entry point] | [doc/meeting/Slack/nowhere] | [name or role] | [score] | [days/weeks/months] | [Y/N/relitigated] | [anything odd] |

Reading the table (do this once all rows are in)

• Column "Real decider": if the same one or two names appear in most rows, you found the actual operating model. Plan accordingly.
• Column "Evidence": if most rows score thin or none, evidence-backed proposals are your differentiator here. If most score rich, match the bar fast.
• Column "Debated where": the venue that appears most is where you need to show up and write well.
• Column "Time to decide": the median is the org's real clock speed. Set your own expectations and your bet timelines to it.
• Column "Stuck?": decisions that did not stick cluster around certain topics. Those topics are political minefields. Walk carefully there until day 90.

Part 4: The agent prompt block (automate the digging)

Most of the fossil record is machine-readable. Point an agent at it. Paste this into Claude (or your agent stack) with repo and channel access, adjust the bracketed parts:

You are a decision archaeologist for [product name]. Reconstruct how significant
product decisions were actually made over the last [12] months.

SOURCES YOU HAVE ACCESS TO:
- [Git repo: PR titles, descriptions, review comments on merges over N lines]
- [Changelog / launch posts at URL or folder]
- [Slack channels: #product, #launch, #exec-escalations, adjust]
- [Docs folder: planning docs, PRDs, postmortems]

FOR EACH CANDIDATE DECISION YOU FIND, return one row:
| Decision (one line) | Approx date | Raised by | Debated where (link) | Apparent decider | Evidence present (rich/thin/none, one line why) | Did it stick (any later reversal/relitigating you can see) |

RULES:
- Candidates: launches, kills, pricing changes, big merges, escalations, and
  anything that appeared on a roadmap and later vanished.
- Cite a link or document for every cell you fill. If you cannot cite, write
  "needs human" in the cell, do not guess.
- Prefer kills and reversals, they carry the most information.
- Cap at 15 rows, ranked by how significant the decision appears.
- After the table, add a 5-line summary: the most common entry point, the most
  common venue, the apparent real deciders, the typical evidence level, and
  the one pattern that surprised you.

The agent gets you 70% of the table in an hour. The remaining 30% (the "real decider" column especially) needs the three or four short conversations. Ask people: "Walk me through how the [X] decision actually happened." People love telling this story. The wiki version and the told version never match, and the gap is the lesson.

The deliverable

A one-page summary: how decisions really get made here, in five sentences, with the table as backup. Do not publish it. It is your private map. The public version of this work is that your first proposal arrives shaped exactly like the proposals that historically won, with the evidence level that historically stuck, in the venue where debate actually happens.