Hiro development journal

Adaptive-discovery feasibility stops at historical evidence gate

Requires a prospective design; no pilot authorized Machine-readable JSON

Executive summary

The approved feasibility stage reached its early stop: readily available evidence does not support the proposed historical query-policy comparison.

Verified 413 packet hashes in the proposed date window; every packet declares that raw external text was not retained. Partial archive records do not repair the missing historical input population.

Produced a prospective replacement draft. Scorer calibration, local model measurement, final preregistration and A/B/C/D execution were not performed after the gate failed.

No production discovery, query, source authority, ontology, promotion authority or observer safety changes were made. No new mission-governance mechanism was introduced.

Work completed

Bounded historical corpus check

Completed: admission not passed
  • Inspected configured discovery packet storage and producer schema. Of 802 enumerated JSON packets, 413 fall within August 28–September 19 Pacific. All verified against sidecars; no development-window packets were present in this tree.
  • GitHub evidence preserves response hashes and byte counts, but not enough source text and historical background metadata to evaluate unissued queries.
  • Common Crawl September catalog spans September 4–17. A bounded repository-search URL-pattern query returned 30 records and one reported page; none matches the narrow or six draft broad queries, and sorts are stars or unspecified.
  • GH Archive supplies events and some metadata, not a demonstrated continuous repository-state history. GHTorrent documentation cautions about dump-generation dates. Software Heritage supports source/metadata preservation but no immediately suitable series was established.
  • No bulk archive ingestion, historical-index engineering or renewed motivating-case search was undertaken. Failure to locate an admissible corpus is not proof that no such archive exists.

Prospective replacement

Draft for review
  • Proposes capturing actual synchronized read-only A/B/C/D query responses in an isolated environment after separate approval, with seven-day development and fourteen-day evaluation windows.
  • Uses a frozen independent reference observation frame and explicitly calls its metric reference-frame recall, not internet-wide recall. Additional arm-only discoveries remain separately measured.
  • Specifies future independent reviewer calibration, model fixture measurements, total cost accounting, capture completeness and authority gates. None of those gates is claimed passed.
  • Preserves the existing hypothesis and records the finding under external research synthesis, still blocked_on_measurement; no experiment created.

Decisions and reasoning

Validation and evidence

CheckStatusResult
Local corpus integrity passed 413 of 413 in-window packets matched existing SHA-256 sidecars; all raw-text persistence flags were false.
Historical corpus admission not_passed No defensible sealed historical input population was established within the bounded review.
Research state and protected files passed Canonical edit and HRO schema validated. Parent question/hypotheses/experiment links and protected source/sentinel hashes remain unchanged.
Scorer, model envelope and pilot not_run Skipped following historical early-stop gate; no inference, benchmark or final preregistration.
Journal tests/build passed npm run test:hiro and npm run build passed; 235 entries validated with timestamp, alias and noindex checks, followed by TypeScript/Vite build.

Current state

Next steps