test: prove complete local product parity on real imported history

[bradtaylorsf]

Summary

Run the final integrated acceptance suite after the OSS application and enterprise composition are merged, using a clean installer-managed stack plus the real local Claude, Codex, and Cowork corpus.

Why this matters

Page existence and unit tests do not prove the product is usable. Completion requires fresh-install behavior, real data, mutation workflows, grounded retrieval, responsive UI, integrity checks, and confirmation that enterprise consumes the same core.

Proposed approach

• Reinstall or upgrade through the supported local create flow without publishing npm.
• Verify database migration/idempotence, background sync, local MCP, API, and both configured local models.
• Exercise every restored UI workflow with agent-browser at desktop and 375px mobile.
• Validate the existing full history inventory and sample raw archive SHA-256 manifests.
• Run core/enterprise parity and boundary checks, then final project gates.

Acceptance criteria

• Fresh local install works without Auth0, billing credentials, or manual API-key creation.
• Complete UI workflows pass: content/import/tags/libraries/answers/recipes/artifacts/reports/dashboards/batch/settings/tokens/audit.
• Expected real inventory is present with no source failures: Claude 848 files, Codex 4105 files, Cowork 112 files.
• Every stored history row has a summary and raw archive manifest; sampled archives match SHA-256.
• API and MCP query/ask/lineage succeed with real records and grounded citations from all three sources.
• Background sync service is installed and verified.
• Enterprise issue the-answerai/answer-engine-enterprise#964 is merged and pins the verified OSS revision.
• pnpm verify, clean migration/init, agent-browser desktop/mobile checks, and alpha-loop run --verify-only 6 pass.

Out of scope

• npm publication.
• New features discovered during acceptance; file regressions as linked issues and fix them before retrying verification.

Related

• Parent epic: epic: Open-source the complete single-user Answer Engine application #6
• Depends on: feat: add enterprise extension contracts and product-parity guards #11 and the-answerai/answer-engine-enterprise#964
• Keep Graphify excluded.

Readiness note: intentionally not labeled ready until both dependencies merge. Add it to the ordered epic queue only at that point.

[bradtaylorsf]

Part of #6

[bradtaylorsf]

Acceptance execution plan

Dependencies are now satisfied: OSS #11 is merged and enterprise #964 merged via PR #966 with OSS pin 2cc52e9c53ab7641a8221df7d8b356d1ccfaab15.

Alpha Loop will process this as the final ordered child with batch size 1. Acceptance must use a clean installer-managed stack, verify the exact Claude/Codex/Cowork inventories, summaries and raw manifests, sampled archive hashes, background sync and local models, exercise every UI workflow on desktop and 375px mobile, prove API/MCP query/ask/lineage with grounded records, run both repository verifiers, and finish with the epic verify-only gate. Any discovered regression will stop the session and be repaired through its own tracked issue/branch/PR before resuming.

Graphify remains excluded; npm publication is out of scope.

[bradtaylorsf]

Preflight stopped — synced harness drift

alpha-loop sync correctly found one tracked difference: CLAUDE.md contained two lines that were not present in the project source of truth, .alpha-loop/templates/instructions.md. The sync result is internally consistent and alpha-loop sync --check is green, but the repository is now dirty, so I am reconciling this generated mirror through a small issue-linked branch/PR before the mandated dry-run.

Lesson Learned

Run the real sync—not only sync --check—during preflight, and treat a generated-output diff as a stop condition. Alpha Loop should never start from a dirty harness posture or from direct edits to a generated harness file.

[bradtaylorsf]

Alpha Loop stopped during preflight context refresh

The runner passed pnpm verify, then detected stale project context, regenerated .alpha-loop/context.md and .alpha-loop/templates/instructions.md, and correctly exited rather than pushing generated files to master. It cleaned the session worktree; #12 was not started and no learning file is expected yet. Session PR #35 remains open for the resumed run.

Review found the generated draft needs correction before merge: it misidentified root src/cli.ts as the CLI entry point and dropped the pinned browser workflow plus issue/branch/PR requirements. I am repairing the generated context on a dedicated #12-linked branch, extending the repository posture regression, syncing harness outputs, and merging that prerequisite before resuming the same session.

Lesson Learned

A successful context scan is not trustworthy by itself. Generated context/instructions must be reviewed against the actual workspace and AGENTS.md; preserve repository-specific browser and GitHub safety posture, and regression-test high-risk architectural claims before resuming.

[bradtaylorsf]

Context repair verified

The generated context is now corrected to the actual workspace (packages/cli owns the CLI), the issue/branch/PR and pinned browser posture is restored, the Graphify exclusion sentinel remains exact, both harness outputs are synced, and the 10-test Alpha Loop posture suite passes. I am publishing this prerequisite through its own PR before resuming session PR #35.