Experiment · demonstrated · candidate
Qualification keeps a verified candidate distinct from production
LARI’s documented governance separates evidence, isolated candidates, qualification, active-model state, and explicit promotion. A current qualification receipt verifies one isolated candidate on two workspace languages: fail-before/pass-after, persisted-candidate reload, and exact-operator removal restoring failure. The active model, registry, and runtime remained read-only, the candidate was not promoted, and external-model calls were zero. The report separately describes rollback controls for bounded production lanes; those rollback checks are inherited report evidence, not a fresh rollback test in this receipt.
Research question
What does one current qualification receipt establish about candidate testing and active-production boundaries?
Why this matters
A candidate passing a contract should remain distinct from an active production capability until its named qualification and promotion gates are satisfied.
What was tested
Bind the candidate primitive to its structured qualification evidence and exercise the defined workspace contracts.
Baseline
An unqualified active legacy record fails closed; production model and registry are protected during the isolated candidate test.
Comparison conditions
JavaScript and Python workspace cases; isolated candidate; fail-before/pass-after; persisted reload; exact operator ablation; read-only production state.
Test environment
Isolated candidate qualification; active model, registry, and runtime were checked read-only.
Observed result
Fail-before/pass-after passed in JavaScript and Python. Candidate reload and exact ablation passed; active model, registry, and runtime remained read-only; candidate was not promoted.
Outcome
One verified isolated qualification lifecycle; no broad safety or universal rollback claim.
Measured results
Measurement conditions: One dated qualification-lifecycle receipt; the receipt does not itself verify the separate production rollback metrics summarized in the report.
- languages Tested
- 2
- fail Before Pass After Passed
- Yes
- persisted Candidate Reload
- Yes
- exact Operator Ablation Restores Failure
- Yes
- active Model Read Only
- Yes
- registry Read Only
- Yes
- runtime Read Only
- Yes
- external Model Calls
- 0
- candidate Promoted
- No
Evaluation method
Verify the isolated candidate against fail-before/pass-after workspace checks in JavaScript and Python; confirm persisted candidate reload, exact operator ablation, and unchanged active model, registry, and runtime.
External model usage
Zero external-model calls.
Cold reload
Persisted candidate reload passed.
Causal ablation
Removing the exact operator restored failure.
Transfer
The receipt covers the named JavaScript and Python workspace checks; no broader transfer is claimed.
Verification and lifecycle
verified · candidate · Measured 2026-10-31T10:00:58.566Z · Report LARI Full Report 2026-10-31T11:42:43.574Z
Limitations
One isolated primitive qualification. Does not prove universal correctness, safety, user-state persistence, or that every candidate type follows identical gates. Production rollback statements are attributed report evidence, not freshly re-tested here.
Architecture component: governance →
Related experiments
- Five bounded coding lanes have report-derived production qualification
- RECAP generates bounded language and composes existing families
Public provenance
Owner-reviewed summary of the dated qualification-lifecycle receipt and source-report governance description; private operator IDs, hashes, and paths are withheld.
Source Workspace primitive qualification-lifecycle receipt · 2026-10-31 · Published 2026-10-31T20:03:40.716Z
Back to experiments
Discuss this page
Read and join this page’s discussion here. Sign in to GitHub when you want to reply or react.
Loading discussion from GitHub…
The embedded discussion could not be loaded. Browse discussions on GitHub ↗