Hiro development journal

Phase 3F-VC finds a viability-to-transfer disconnect

2026.08.29 at 14:43:13 America/Los_Angeles Phase 3F-VC implementation, full Hiro validation, public journal tests, and the production journal build completed successfully Machine-readable JSON

Executive summary

Phase 3F-VC tested whether the two findings selected by measured-gap viability could produce legitimate Hiro candidates. It used only the two accepted Phase 3F-VR inputs and ran no new discovery.

Both source mechanisms compiled through a new claim-independent behavior-relevant verification primitive, executed in isolation, retained immutable evidence, passed independent metric reconstruction, and returned REPRODUCTION_SUPPORTED.

In both experiments, regression-detection accuracy remained 1.0 while mean evaluation cost fell from six suites to one, an 83.33% reduction against a frozen minimum reduction of 20%.

The unchanged corroboration truth table returned CORROBORATED for both validated evidence packages.

Transfer reassessment then found that the current gap had actually been measured in adaptive suite selection, while neither finding's frozen production surface contained or invoked that measured interface.

Both findings terminated TRANSFER_NOT_JUSTIFIED with the normalized reason MEASURED_GAP_TARGET_NOT_IN_FROZEN_TRANSFER_SURFACE.

No candidate specification was legitimate, so candidate construction, fidelity, target evaluation, holdout, candidate resources, complete candidate regression, governor review, and promotion eligibility were correctly NOT_REACHED rather than waived.

The primary result is a viability false positive: the gaps and research mechanism were real, but Phase 3F-V had mapped them to production surfaces not connected to the measured behavior.

The complete Hiro repository suite passed 876 tests with one expected skip. Hiro is healthy with the local model connected and its loaded and checkout revisions aligned at 045d77b8bbefa8f1265b48a80594c0420353a586.

No promotion transaction was created, no candidate activation occurred, no fresh discovery ran, and no meta-improvement began.

Work completed

Immutable two-input bridge

Completed

Verified the accepted Phase 3F-VR final report against its expected SHA-256 and selected exactly the two findings classified as both MEASURED_GAP and VIABLE_FOR_REPRODUCTION.
Frozen source provenance, source claim, transfer hypothesis, viability assessment, probe specification, probe evidence identity, exact metric and threshold, transfer surface, and evaluation proposal before reproduction.
Generated a feasibility subset by deterministic disposition filtering only; no source, claim, or easier replacement was introduced.

Generic mechanism reproduction

Completed

The first preserved attempt correctly reported UNSUPPORTED_EXPERIMENT_TYPE because the generic compiler lacked a behavior-aware selective-verification primitive.
Added one safe primitive selected from semantic mechanism fields rather than claim IDs or source titles. Its cases are bounded data-only candidate behaviors and tagged evaluation suites; regression labels are not selection inputs.
The baseline executes all six suites. The treatment selects suites through behavior-tag overlap. The frozen criterion preserves detection accuracy within five percentage points and requires at least 20% cost reduction.
Both final reproductions used 12 frozen cases, kept accuracy at 1.0, reduced mean suite executions from 6.0 to 1.0, and returned REPRODUCTION_SUPPORTED.

Independent validation and corroboration

Completed

Both evidence packages passed identity, case-count, control, treatment, raw-artifact, metric-reconstruction, threshold, resource, and isolation checks with zero reason codes.
The isolated calculations were network denied, used no model bridge, and retained no candidate, production-write, or promotion authority.
The unchanged corroboration gate returned pass with reason CORROBORATED for both findings.

Production transfer reassessment

Completed

A first transfer attempt exposed an infrastructure semantic defect: contextual repository references were being treated as though they had generated the measured observations.
The minimum repair made the connection audit primitive-aware and identified the actual measured subject for adaptive-suite relevance as adaptive_suite_plan.
The first finding's frozen target was variant laboratory and promotion-gate code; the second finding's was agent and metacognition code. Neither target directly contains or invokes the measured adaptive-suite selector.
Both findings therefore terminated TRANSFER_NOT_JUSTIFIED before candidate specification.

Decisions and reasoning

Do not substitute current-gap evidence for mechanism reproduction; preserve the two evidence stages separately.
Add a reusable safe experiment primitive only after the original compiler failure was preserved, and do not change the plans' falsification semantics.
Treat the short execution path required by the Windows/isolated-output boundary as an execution-location repair, not permission to change cases, metrics, or criteria.
Credit a production transfer only when the component that generated the measured behavior is directly in, or invoked by, the frozen permitted surface.
Do not build an unused helper or evaluate an easier proxy merely to manufacture a candidate.
Stop before candidate construction and promotion because both findings legitimately failed transfer reassessment.

Validation and evidence

Check	Status	Result
Focused reproduction, bridge, and prior-boundary tests	passed	29 tests passed. The only warning was the existing inaccessible pytest cache directory.
Generic reproduction evidence	passed	Two plans preregistered, compiled, executed, and independently validated. Both returned REPRODUCTION_SUPPORTED and CORROBORATED; the report verifier returned valid with zero reason codes.
Transfer report verification	passed	The corrected report verified with zero reason codes, two TRANSFER_NOT_JUSTIFIED outcomes, zero candidate specifications, zero promotion transactions, and zero activations.
Complete Hiro repository suite	passed	876 tests passed, one expected test was skipped, and seven non-failing warnings were reported in 402.76 seconds.
Live runtime identity	passed	Hiro reported healthy with the local model connected and checkout and loaded runtime both at 045d77b8bbefa8f1265b48a80594c0420353a586.
Public journal tests and production build	passed	npm run test:hiro passed. npm run build generated and validated 181 journal pages, compiled TypeScript, and completed the Vite production bundle. The clean checkout first required npm ci; installation reported one existing high-severity dependency advisory, which was not modified because dependency maintenance was outside this session.

Current state

Phase 3F-VC is terminal with disposition PHASE 3F-VC NOT DEMONSTRATED — VIABILITY FALSE POSITIVE.
Two measured-gap findings reproduced and corroborated, but neither had a valid production transfer connection.
There are zero candidate specifications, candidate revisions, governor passes, and PROMOTION_ELIGIBLE candidates from this phase.
Hiro is healthy and revision-aligned at 045d77b8bbefa8f1265b48a80594c0420353a586.
No new discovery, candidate promotion, activation, or meta-improvement is underway.

Next steps

Under separate authorization, refine Phase 3F-V so the proposed transfer surface must be connected to the component and interface that produced the measured current-gap evidence.
Reassess these preserved findings only if a new frozen transfer proposal is derived without changing their measured-gap evidence or reproduction result.
Do not promote anything or start fresh discovery from this terminal qualification.