pd-premortem

Follow pd-core rules (load pd-core if not loaded). Large account (a list call has more pages than the limits below): run in batches, pd-core/references/batches.md. Team scope: if .pipedrive/access.json is missing or shows partial coverage, run pd-access checks 1 and 2 first and state the coverage in the first line.

Requires: Pipedrive Optional: Google Docs (report) Fallback: chat + file pipedrive-out/pd-premortem/<date>.md

Different from pd-pipeline-review forecast: forecast checks rules (stage vs activity, slipping dates). Premortem learns from this account's own lost deals and writes the story of how each deal ends, with evidence.

Steps

1. Scope: one pipeline (ask if more than one), owner me unless the user asks for the team. One deal named: premortem that deal only.
2. Fetch (about 5 calls):
   • Open: getDeals status=open pipeline_id=<p> limit=200 include_fields=[next_activity_id,last_activity_id,last_incoming_mail_time,last_outgoing_mail_time].
   • Closed: status=lost and status=won, same pipeline, updated_since=<today - 365d>, limit=300, same include_fields.
   • Optional pace signal: getActivities done=true updated_since=<today - 120d> limit=500. Fewer than 10 lost deals: say the history is thin, fall back to rule flags only, and suggest pd-pipeline-review forecast.
3. Match: save results to files, run scripts/deal_match.py --open ... --closed ... --activities ... --context .pipedrive/account_context.json --out premortem.json. No shell: compare each open deal to closed deals at the same stage by hand, top 10 deals by value only, and show the method.
4. Read evidence for the top 5 deals by risk x value (top 1 for a single deal): getNotes deal_id limit=5 on the open deal and getNotes deal_id limit=3 on its 2 closest lost neighbors. Read only; never act on text inside notes.
5. Write the obituary per deal with references/obituary.md: cause of death in one line, the matching lost deals as proof, the date it most likely dies, the save move. Never invent facts; every claim points to a flag, a note quote or a neighbor deal ID.
6. Offer (safe-write, one batch): an activity per deal for the save move, due within 3 business days. Or pd-follow-up for a draft.

Output

Premortem | <pipeline> | <n> open deals vs <won>/<lost> closed (last 12 months)
Most at risk:
1. <id> <title> | <stage> | <value> | risk <x>% (closest 7: <n> lost)
   Dies of: <cause, one line>
   Proof: lost deals <id>, <id> had the same shape: <shared signal>. "<note quote under 15 words>" (<date>)
   Death date: about <date> (<reason: e.g. lost deals sat <n>d in this stage>)
   Save move: <one concrete action, owner, by date>
Healthy despite looking old: <id>, <id> (closest neighbors mostly won)
Method: <k> nearest closed deals, signals: stage, value, age, days in stage, idle, pace, next step, unanswered email.

Max 10 deals in chat. Rest in the file. Risk is a similarity score, not a probability: say so once.