koala73 / koala73/worldmonitor
ops(company-monitoring): run Stage 1A — two-external-customer historical-usefulness gate
- Dominant language
- TypeScript
- Stars
- 86.6k
- Forks
- 13.1k
- Avg merge
- 8h 4m
- Merged PRs (30d)
- 825
Description
## Parent
#6002
## Why this issue exists
The epic names Stage 1A as a promotion stage, #6003 froze its protocol, and #6015/#6016 activation waits on it — but no issue owned *running* it. This issue owns the Stage 1A execution and its evidence.
## What to run
Execute the frozen two-external-customer historical-usefulness protocol from #6003 (`tests/fixtures/company-monitoring-evaluation/protocol.json`, `cm_eval_v1`).
## Acceptance criteria
- [ ] Two external target customers recruited and recorded internally. Internal analysts cannot approve usefulness. **Recruitment is the longest external lead in the whole epic — start it immediately; it is not blocked by anything.**
- [ ] Both customers evaluate the same ten admitted impacts, including at least one positive, one negative, and one mixed, per the frozen protocol.
- [ ] Each customer independently rates at least 70% useful; no self-selected feedback.
- [ ] Results recorded as aggregate counts and opaque IDs only — no portfolio names, company claims, customer queries, or evidence text.
- [ ] Pass/fail recorded against the frozen thresholds with named product-owner sign-off.
## Blocked by
- #6003 (frozen protocol — approved 2026-08-05; Stage 0 measurements still pending)
- #6011 (produces the admitted impacts the customers evaluate)
Customer recruitment is explicitly NOT blocked and should begin now.
## Stop condition
A failed usefulness gate fails Stage 1: the workspace (#6015) and fenced alerts (#6016) stay dark, and promotion stops pending a product decision. Threshold changes cannot rescue a running evaluation (`currentRunMayNotBeRescuedByAmendment`).
Contributor guide
Research direction
Start by reading tests/fixtures/company-monitoring-evaluation/protocol.json for the frozen cm_eval_v1 procedure, then review blockers #6003 and #6011. Recruit two external customers and run the same ten admitted impacts; done means aggregate opaque-ID results, threshold pass/fail, and named product-owner sign-off are recorded without customer or portfolio details.
Written by the indexing model from the issue text.
Assessment
- Domain
- observability
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 30/100