Hiro development journal

Broader static GitHub discovery activated with selection receipts

Hiro running; static discovery active; adaptive research remains unresolved Machine-readable JSON

Executive summary

Implemented and activated the explicitly authorized production remediation: six separate static GitHub queries rotate within the existing two-hour request cadence.

Added bounded result-level identity, hashing and selection decisions to existing frozen packets, with lead/lineage links and discovery-event packet references.

Focused tests and broader regressions passed. Hiro restarted healthy on the committed revision and passed its existing automatic pipeline qualification.

No adaptive queries, source expansion, ontology or mission changes, historical reconstruction or A/B/C/D benchmark were performed.

Work completed

Bounded static retrieval

Active
  • Exact rotation: agentic; AI evaluation; AI reliability; AI memory; AI tools; language model. Each is a separate query, with no mandatory language qualifier.
  • One request per eligible source attempt, at least 120 minutes apart, update-descending first page, at most 20 records. A complete rotation takes at least twelve hours.
  • A persisted cursor and attempt timestamp survive restarts. Failed attempts consume the interval; migration respects the prior successful-fetch timestamp.
  • The approved endpoint, response and time limits, other source policies, relevance machinery and reservoir bounds remain unchanged.

Selection provenance

Active for future ordinary fetches
  • Packets add query ID/text/hash, timing, request count, rank-ordered repository identities and canonical URLs, bounded metadata/input/representation hashes and incomplete-result status.
  • Receipts distinguish invalid records, injection/relevance rejection, duplicate content, retained leads and reservoir/idea selection limits. No raw source prose is newly persisted.
  • Existing lead and idea lineage IDs connect collection to continuous-queue history; collection receipts do not pretend downstream work has already happened.
  • The discovery summary/event references its frozen packet. No separate telemetry service was introduced.

Activation and continuity

Verified
  • Committed the previously approved research-management work separately to satisfy the existing clean-source qualification gate, then committed the static discovery remediation.
  • Discovery revision: 95713f5a21366e09c544bfe0144acc67f1fd8755
  • Used the existing exact-process stop and supported detached launcher. Health reports the matching loaded/checkout revision and connected local model.
  • Normal scheduler qualification: 5 of 5 scenarios passed.
  • GitHub cadence was not bypassed. Next eligibility at verification: 2026-09-20T17:23:42.570837+00:00. An ordinary no-fetch discovery packet is not counted as a new GitHub response.

Decisions and reasoning

Validation and evidence

CheckStatusResult
Focused discovery/integration tests passed 52 passed, including schedule validation, reload/cadence/failure budgeting, adapter behavior, dedup, receipt safety and integration. Initial overlong Windows test paths were corrected by using a shorter test directory.
Broader regression tests passed 190 passed across continuous engine/governor/state durability, internet observation/probation, source claims and research registry.
Normal automatic pipeline qualification passed 5 of 5 scenarios passed on the activated revision.
Health and authority passed Health ok, model connected, revision matches. Two health probes timed out during verification; subsequent HTTP and Observatory health checks both returned 200 in 0.09 seconds. Protected policy/sentinel hashes and unrelated source-policy values unchanged.
Journal tests/build passed npm run test:hiro then npm run build passed; timestamp, aliases and noindex validation plus TypeScript/Vite build complete.

Current state

Next steps