Prepare, qualify, and freeze the v0.1.0 release candidate from main #598

Outcome

Produce one exact, reviewable v0.1.0 release candidate from stabilized main, qualify it, and freeze its source and artifact identities for the paired Ultrafuzzbench experiment in #597.

This is release engineering, not feature development. Do not add feature scope while assembling the candidate.

Entry conditions

• Human strategy review Manually review every strategy and decide bug-search vs test-authoring framing #5 is complete and every approved shared-node change is merged to main.
• The human decision for the Invariant campaign can report complete when Recon smart-filter drops a generated property #519/Validate selected properties against executable backend oracles #524/Isolate invariant properties so one failure cannot retire unrelated backend tests #589 property ↔ oracle ↔ backend-entrypoint contract is recorded, and the approved bounded work is complete or explicitly deferred.
• Release-blocking bug issues in Stabilize, benchmark, and release Ultrafuzz v0.1.0 #184 are complete or have an explicit maintainer disposition.
• OpenRouter Integrate OpenRouter for v0.1.0 via main #584 / PR feat(openrouter): add catalogue-wide model support #586 is reviewed, green, and merged directly to main.
• No untracked feature work is being pulled into the candidate.

Candidate construction and qualification

• Start from the exact stabilized main revision; do not revive or merge the historical release/v0.1.0 branch or PR v0.1.0 release (open source candidate) #366.
• Carry all release-only metadata before the freeze, including the approved version convention, package versions where applicable, changelog/release notes, migrations, and compatibility notes.
• Repair the known changelog hygiene problems recorded in Stabilize, benchmark, and release Ultrafuzz v0.1.0 #184, including the duplicate R40 Aave v4 run exhausted Modal recovery after sandbox exits #267 entry, and make the v0.1.0 shipped-change record reviewable.
• Verify the final schema-v2 catalog contains both default and fuzz-only profiles.
• Prove that shared nodes, prompts, output contracts, settings, dependencies, and bug fixes are identical across the profiles; only the intended threat/database/goal-planning profile subgraph may differ.
• Pin and record the external vulnerability-database revision and retained manifest/record provenance without treating the corpus as an Ultrafuzz package.
• Run all deterministic release-validation, workspace/type, package/install, migration, and packed-artifact checks required by Stabilize, benchmark, and release Ultrafuzz v0.1.0 #184.
• Perform bounded operational qualification for both profiles on the pinned Ultrafuzzbench cohort. This qualification proves the candidate can run; it is not the scored paired experiment in Benchmark v0.1.0 default vs fuzz-only audit profiles on the frozen release candidate #597.
• Complete appsec and open-source readiness Complete appsec and open-source readiness review for v0.1.0 #527 against this exact candidate. If Complete appsec and open-source readiness review for v0.1.0 #527 requires a product or packaging change, apply it, invalidate the candidate, and repeat qualification before freezing again.
• Record the candidate source SHA, candidate/merge-tree identity, package/image identities and digests, catalog/profile/topology/prompt digests, external corpus revision, toolchain/runtime versions, and validation evidence in one candidate manifest or equivalent durable record.

Freeze rule

After the candidate is declared frozen, any product, prompt, topology, output-contract, dependency, model/evaluator configuration, release metadata, or package/image change invalidates the freeze. Re-run the affected qualification and #527 checks, issue a new candidate identity, and do not start #597 until the replacement is frozen.

Acceptance criteria

• Every entry condition has a recorded disposition.
• The versioning convention and all release metadata are maintainer-approved and present before freeze.
• Both profiles pass deterministic validation and bounded Ultrafuzzbench operational qualification.
• Complete appsec and open-source readiness review for v0.1.0 #527 is complete against the final candidate identity.
• Packed artifacts contain only intended files and reproduce the recorded identities.
• A candidate manifest records every identity and digest required by Benchmark v0.1.0 default vs fuzz-only audit profiles on the frozen release candidate #597 and the final release issue.
• A maintainer explicitly records that the candidate is frozen and ready for Benchmark v0.1.0 default vs fuzz-only audit profiles on the frozen release candidate #597.

Non-goals

• Running or interpreting the scored paired benchmark.
• Tagging or publishing v0.1.0.
• Adding benchmark-only product behavior or new release features.

Creating or completing this issue does not itself authorize merging a final release PR, tagging, or publishing.