Hiro development journal

Canonical research management integrated into the existing registry

2026.09.19 at 21:45:49 America/Los_Angeles Implementation and local continuity validation complete; research autonomy remains unqualified Machine-readable JSON

Executive summary

Following explicit approval of the preceding design, implemented the minimal canonical Research Map and Hiro Research Object layer in the existing knowledge-base index and Open Problems Registry.

The implementation preserves existing problem identity, hypotheses, experiment links, evidence, agenda and production controls. It adds research context, conflict-checked local edits, complete new revision events, artifact receipts and read-only views.

Imported the approved evaluative-dimension-discovery example into the canonical registry as observing-only. No experiment was admitted and no promotion authority was added.

Validated deterministic research-memory continuity across three separate CLI processes. This demonstrates local persistence and handoff behavior, not independent scientific competence or recursive improvement.

Work completed

Single canonical registry and migration

Completed

Extended open_problems with one additive research_json column; retained existing canonical hypotheses, experiments, evidence and append-only event tables. Initialization is idempotent and migration preserves legacy data.
Local edits require expected event revision, expected research-question hash and an attributed reason. Transactions serialize competing edits and reject stale updates.
New hypothesis, experiment and research edits preserve full before/after reasoning. Replaced replacement-style writes with upserts that retain foreign-key lineage.
Managed research originator is immutable, context cannot be erased, related problem references and parent cycles are checked, and interpretations can cite only evidence belonging to the same problem.
Created a SQLite backup before migrating the actual registry. The original ten problems remain; one observing-only example was added with its original provenance and no experiment link.

Map, sources and deduplication

Completed

Extended the existing knowledge-base JSON with the established objective, five RSI levels, terminology and aliases, sourced historical capabilities, categorized limitations, negative evidence, prior art, deprioritized ideas and a small frontier.
Current unresolved problems and hypotheses come from a consistent live registry snapshot and watermark rather than a second maintained list.
A bounded lexical/curated-alias check searches existing and retired problems, related concepts, hypotheses, failed evidence and prior art. It explicitly cannot establish novelty.
Historical repository hashes expose changed-since-review or unavailable sources. Old qualifications remain historically dated and are not silently upgraded to current reproduction.
Public projection excludes all runtime row text, including titles, source strings and evidence. Only the reviewed static map and counts are eligible for public export.

Research to experiment to evidence

Completed

Managed research readiness requires falsification, controls, preregistered success criteria, a measured resource envelope, implementation plan, reviewed deduplication and no declared blockers.
Planner and existing executor freeze research intent. Admission, execution, retry, replication, calibration and independent challenge reject incompatible intent changes instead of silently substituting a new question.
Ordinary terminal executor reports return evidence to the same problem. Infrastructure failures remain inconclusive rather than scientific rejection.
Local receipt import independently hashes the referenced artifact, validates experiment ownership, and uses experiment/hash/result-version identity for idempotency. Conflicting reuse is rejected; corrections append and may supersede prior evidence.
Evidence does not automatically resolve the problem, reinterpret the hypothesis or authorize promotion. Interpretation, research decision and production state remain separate.

Local handoff and Observatory

Completed

Added validated map, export, apply, receipt, search and paginated-history CLI operations with JSON and readable map/HRO Markdown output.
Added read-only local API research map/HRO routes and research content in the existing Observatory Open problems view. No public write route was added.
Documented exact edit versus export contracts, actor attribution, conflict recovery, legacy compatibility, readiness and authority boundaries in the existing knowledge-base guide.
Added a pinned JSON Schema validation dependency to the existing environment requirements and installed it in the configured local runtime.
Retained the approved example as proposed research with unknown confidence, alternatives, falsification, controls, resource blockers and original contribution attribution. No claim that Hiro originated or executed it is made.

Decisions and reasoning

Reuse problem_id as HRO identity and the existing registry as the single store. No second queue, ontology, scheduler or promotion lifecycle.
Keep optional research fields rigorous when present while allowing legacy unmanaged rows. Model authorship labels are attribution rather than authorization tokens.
Keep artifact verification distinct from scientific validity and ordinary executor status reports.
Treat the verified three-process handoff as deterministic integration evidence. Further independent model-based research qualification remains necessary.
Do not execute dimension discovery until its explicit blockers are resolved and it passes the existing admission process.

Validation and evidence

Check	Status	Result
Research registry, agenda and executor focused suite	passed	38 passed in 5.60 seconds after final planner/calibration and infrastructure-evidence integration.
Separate-process local CLI continuity	passed	Three fresh CLI processes exported, applied interpretation and reread the same question and failed result. No automatic resolution occurred; fixture only.
Schema, Observatory JavaScript and whitespace	passed	Draft 2020-12 schema self-validation, node --check on extracted Observatory JavaScript, and git diff --check passed.
Approved example import	passed	Actual registry backup and additive migration completed; example export validates, remains observing, preserves null research confidence and has zero linked experiments.
Hiro full suite	incomplete	Started the full suite; output reached 12% without a final result before the overnight session interruption. The process is no longer running. No full-suite pass is claimed; focused integration validation passed separately.
Journal tests and build	passed	npm run test:hiro passed, followed by npm run build: 230 entries validated with timestamp/alias/noindex checks, TypeScript and Vite completed.

Current state

Hiro source changes are local implementation work. The journal publishes only this reviewed summary and generated journal pages.
No production activation, promotion, autonomous research campaign or example experiment was run as part of this integration.
Historical event payloads cannot recover details never recorded. Complete before/after reasoning begins with the new implementation.
Lexical deduplication, static literature synthesis and contributor counts are not novelty proof or research-autonomy qualification.
Existing API process needs its normal reload to serve updated Observatory routes. This task did not restart autonomous operation.
Resource feasibility for the scientific example remains unmeasured; consciousness remains a separate unproven hypothesis rather than a system success criterion.

Next steps

Use canonical HRO exports for subsequent research discussions and return reviewed interpretations through conflict-checked edits.
Independently qualify whether a model can use negative research memory to avoid rediscovery and design valid experiments under matched budgets.
Resolve evaluator validity, source freshness, prior-art and measured resource blockers before admitting dimension discovery.
Measure research responsibility transfer through valid designs, reproducibility and comparable gains per compute, preserving assistance attribution.