Hiro development journal

Observation stop-sentinel provenance and separate discovery operation

Read-only forensic review complete; no operational changes Machine-readable JSON

Executive summary

The current internet-observation stop sentinel was created on August 8, 2026 at 14:06:36 Pacific, matching completion of the seventh finite probation attempt. Its content states that finite probation completed.

The existing probation runner restores this sentinel at the configured attempt cap and on subsequent completed checks. The stop was intentional finite authorization closeout, not a global shutdown of external discovery.

Verified a separate external-idea discovery mechanism operated September 15–19 Pacific: 319 frozen packet hashes matched, with 317 completed scheduler cycles and two failed cycles.

Reviewed the earlier diagnostic replay script without executing it. It changes no production file or sentinel; its exact-URL enabled policy existed only within the diagnostic process. The sentinel stopped it before networking.

Work completed

Sentinel provenance

Completed
  • Filesystem creation and last-write time are August 8 at 21:06:36 UTC. The seventh frozen session completed at 21:06:36.164806 UTC.
  • All seven probation session hashes verify. Three attempts completed normally and four ended as infrastructure failures; all attempts count toward the finite cap.
  • Repository commit f5b2d29 introduced finite observation probation on August 6. Commit e6c3672 changed the cadence and rotations while keeping the seven-attempt cap. Commit 6468960 repaired source compatibility before the final attempt.
  • The August 8 journal closeout explicitly records that the unchanged scheduler ran attempt seven, then Hiro restored the sentinel. The artifacts identify the runner and scheduler path; they do not retain a specific operating-system writer PID.

Intended lifecycle and architecture

Completed
  • The observation lane is disabled by default, separately authorized for bounded sessions and finite probation, and closed at the attempt limit. Completion does not automatically grant further internet or promotion authority.
  • The documented later-launch conditions include explicit source selection, a clean pinned revision, focused/full validation, retained production safety controls, explicit policy/sentinel authorization, a small read-only session and frozen evidence checks. This is a record of existing requirements, not a recommendation or authorization.
  • The checked-in template remains disabled. Runtime probation configuration still says enabled but its seven-of-seven completion condition prevents further cycles and restores the stop sentinel if absent.
  • Continuous external discovery is routed through the active improvement adapter, run_discovery_only and collect_external_ideas. Moltbook uses a source-specific observer policy and a different stop sentinel; other configured sources use their existing read-only fetchers.
  • No evidence found of a later decision requiring the completed generic probation to remain enabled for continuous discovery.

September 15–19 actual operation

Completed
  • 2026-09-15: 43 verified packets, 620 item reviews, 6 emitted idea entries and 7 retained lead entries.
  • 2026-09-16: 96 verified packets, 920 item reviews, 5 emitted idea entries and 7 retained lead entries.
  • 2026-09-17: 96 verified packets, 920 item reviews, 12 emitted idea entries and 12 retained lead entries.
  • 2026-09-18: 76 verified packets, 845 item reviews, 10 emitted idea entries and 11 retained lead entries.
  • 2026-09-19: 8 verified packets, 546 item reviews, 6 emitted idea entries and 6 retained lead entries.
  • All five days include completed fetches from Moltbook, GitHub, arXiv, BBC Technology, Wired AI and Ars Technology. Reddit was disabled. Item-review totals are not unique-source counts.
  • Two scheduler cycles, on September 15 and 18 Pacific, recorded an episode-ownership error after source packets had been frozen. The other 317 cycles completed.
  • The records demonstrate recurring operation on every requested day, not uninterrupted uptime or a promise of a fetch every 15 minutes. Existing source intervals and active-work deferrals remain relevant.

Diagnostic script effects

Completed
  • The script is outside the Hiro working tree and is not installed in a scheduler, API route or startup path.
  • It writes diagnostic files under its artifact directory. It constructs an in-memory exact-URL enabled observation policy while retaining numeric resource limits and the same stop sentinel.
  • It would use existing sanitization and a mechanically bounded source representation; that representation and every semantic stage were never reached.
  • No persistent production policy, discovery source list, queue, HRO, promotion authority or safety sentinel was changed. Recorded production code/policy hashes still match the pre-attempt manifest.

Decisions and reasoning

Validation and evidence

CheckStatusResult
Frozen probation evidence passed All seven session SHA-256 companions verified.
September 15–19 discovery evidence passed 319 of 319 frozen discovery packet hashes verified; cross-checked scheduler event counts.
Prior replay production hashes passed All production file hashes recorded before the prior attempt still match.
Hiro tests not_run Read-only forensic task; inspected test contracts and historical validation records without executing tests or replay.
Journal validation passed npm run test:hiro and then npm run build passed; 232 entries validated with timestamp, alias and noindex checks, followed by TypeScript and Vite.

Current state

Next steps