Run metadata
AI simulationv0.3
| Field | Value |
|---|---|
| Run date (real) | 2026-08-31 |
| Standard under test | keystone branch readability at commit 2c88669 (v0.3 unreleased, after the phase-1 consistency/readability pass, before any simulation-driven change) |
| Scenario | ../../scenario.md (CertHub — wholly fictional) |
| Organisation | ../../../../organisations/fernbrook.md (Fernbrook Software) |
| Seats played by | AI role-play throughout (Claude agents, one per seat, orchestrated per PROTOCOL.md); no human seats this run |
| Stages run | S0, S1–2a, S2b, S3a, S3b, S4, S5a/b/c (parallel), S5d, S6, S7 (firewalled), S8 — 12 runs |
| Firewall | Held: S7 and S8 read artifacts/ only; findings files were never shared between seats |
| Outputs | 16 in-world artifacts, 12 findings files, findings/CONSOLIDATED.md, findings/NORMATIVE-CANDIDATES.md, REPORT.html |
Outcome
~110 findings → 10 converged themes. Dispositions: 27 reference-tier fixes applied
(commits 8bbe8ee…d5c977b); normative package approved by the maintainer and applied
to the unreleased v0.3 as FSD-GV-8 (owner transition / continuity re-verification),
FSD-GV-9 (service levels & reporting), FSD-SD-7 (conditions tracked to closure),
FSD-SA-6 (key/secret management), plus note-level items (commit 3862314); role
questions deferred to the Tier 3 guidepost; document-content guidance deferred to the
v0.4 worked example (BACKLOG).
Notes for the next run
- The historical
artifacts/here predate the v0.3 form changes this run caused — a re-run against the current standard should produce records with condition fields, register rows and operating-effort lines already in the templates. That delta is itself worth checking. - All-AI seats share blind spots; put a human in S5b or S7 next time as a control.
- This artefact set is candidate seed material for the published worked example (Roadmap, v0.4): fictional, organisation-neutral, complete idea-to-succession.