Register the event-slug data migration in the isolation inventory

[alexeygrigorev]

Product outcome

Restore the deterministic data-migration isolation gate after the accepted event-slug shortening added a historical RunPython migration without registering it in the gate's exact inventory. The gate must discover and inspect that migration like every other data migration, while event rows, aliases, canonical URLs, slug allocation, and migration execution remain byte-for-byte behaviorally unchanged.

Confirmed evidence

• Exact failed release anchor: face8e4808d65afbf0374d1ced7a88079950d663.
• Scheduled run 33315212334, migrations job 99267303151 ran make test-migrations and failed MigrationSeedContractTests.test_every_data_migration_uses_historical_apps_without_runtime_side_effects because the discovered set contained one unregistered path: events/migrations/0012_shorten_event_slugs.py.
• Commit ef19e48 added that migration. It uses historical apps.get_model("events", "Event") and apps.get_model("events", "EventAlias"), performs deterministic ordered database updates, and imports no current application model/service.
• test_support/tests/test_migrations.py maintains an exact allowlist of every file for which data_migration_functions() finds a RunPython callable. It then runs assert_data_migration_isolation(path) and requires the source to contain apps.get_model for every registered path. The new migration is discovered correctly; only the expected inventory entry is absent.
• The same CI job separately fails EditorialRouteMigrationContractTests.test_checked_manifest_is_bound_to_schema_projection_and_runtime on a projection digest mismatch. That failure belongs to Reconcile non-reproducible public projection after face8e4 #253 and is not part of this issue.
• Repository and GitHub issue search found no existing owner for the missing 0012_shorten_event_slugs.py inventory entry. Restore face8e4 formatting, typing, and focused content gates #261 owns a bounded formatting/typing/template recovery and does not include this file or failure.

Normative authority

• Development process, for truthful red-gate recovery, role separation, and terminal verification.
• Migration and rollout roadmap, for backend-portable migrations and migration gates.
• Event specification, for immutable Event identity and event behavior that this gate-only correction must preserve.
• test_support/tests/test_migrations.py, the current executable exact-inventory and historical-app isolation contract.

Scope

Change exactly one tracked file:

• test_support/tests/test_migrations.py

Add events/migrations/0012_shorten_event_slugs.py exactly once to the expected data-migration path set in test_every_data_migration_uses_historical_apps_without_runtime_side_effects. Retain exact-set equality so both an unreviewed new data migration and a stale/nonexistent inventory entry continue to fail closed.

The existing loop must then apply the existing isolation checker and historical-apps.get_model source assertion to migration 0012; do not special-case, skip, mock, or weaken either check.

Acceptance criteria

• The exact expected data-migration inventory contains events/migrations/0012_shorten_event_slugs.py once, in the established deterministic ordering, with no other inventory membership change.
• The focused inventory test discovers migration 0012 through the existing AST-based RunPython detection and runs the same assert_data_migration_isolation and apps.get_model requirements used for every other data migration.
• Removing the new inventory entry reproduces the exact unexpected-path failure; adding an unregistered synthetic numbered RunPython migration or a stale expected path still fails exact-set equality. No permissive glob-only assertion replaces the reviewed inventory.
• events/migrations/0012_shorten_event_slugs.py, events/slugs.py, event models/services/routes/templates/projections/manifests, and event tests are unchanged by this issue.
• Existing migration 0012 semantics remain unchanged: historical models only, deterministic id ordering, current slug/alias update behavior, no network/provider/runtime-model import, and the existing no-op reverse operation.
• The exact focused Django test passes on the frozen candidate. After Reconcile non-reproducible public projection after face8e4 #253 repairs the independent editorial projection digest mismatch, make test-migrations passes in full with 18 tests and no skip, retry, xfail, quarantine, assertion weakening, or accepted residual failure.
• Locked format/lint/type/migration-drift checks and the graph-selected versioned verification plan/report pass on the same frozen candidate.
• Engineer and independent tester publish their exact base/candidate, one-file diff digest, plan/report digests, terminal commands/counts, and confirm the event product surface is unchanged.

Required test scenarios

1. Run the exact focused test:
   uv run --frozen python manage.py test --noinput test_support.tests.test_migrations.MigrationSeedContractTests.test_every_data_migration_uses_historical_apps_without_runtime_side_effects.
2. Prove migration 0012 is in the discovered set and is evaluated by both existing historical-isolation assertions, without changing the migration source.
3. Prove the contract remains fail-closed for one unexpected numbered RunPython migration and one stale expected path using the smallest existing test technique; do not leave a synthetic migration in the repository.
4. On a base containing Reconcile non-reproducible public projection after face8e4 #253's accepted projection repair, run make test-migrations and require all 18 tests green. A run that still contains Reconcile non-reproducible public projection after face8e4 #253's unrelated digest failure is useful attribution evidence but is not terminal tester acceptance.
5. Run the change-selective plan and every selected repository/Django gate. Browser execution and screenshots are not_applicable only if the final one-file diff proves no render-impact path changed.

Dependencies and recovery ordering

There is no implementation dependency on #253 or #261: an engineer may prepare and freeze this one-file correction independently from the exact face8e4 anchor. It does not overlap #261's allowed files and must not be folded into or mutate the already frozen #261 candidate.

Final tester acceptance and integration do require a clean shared migration component. #253 owns the second, independent make test-migrations failure and must supply its accepted repair before this issue can claim the full target green. Preserve the recovery order already assigned to #261; rebase this candidate after the accepted #253/#261 recovery base and land it immediately after them, before later recovery issues reuse or claim migration-component evidence. Any base or source change invalidates the versioned plan and requires recomputation.

Non-goals

• No edit to events/migrations/0012_shorten_event_slugs.py, event data, aliases, slug length/allocation, routes, redirects, public copy, visual design, or event behavior.
• No new migration, seed, reverse migration, data replay, database/provider/production access, or durable job.
• No Reconcile non-reproducible public projection after face8e4 #253 projection/source/pin/manifest/hash/terminology repair and no generated artifact change.
• No Restore face8e4 formatting, typing, and focused content gates #261 formatting, archive-row, normalization, or prose-template work.
• No refactor of data_migration_functions, assert_data_migration_isolation, the migration test suite, Make targets, CI workflow, tool version, or verification policy.
• No assertion weakening, broad exception, skip, retry, xfail, quarantine, timeout increase, test deletion/demotion, commit, push, merge, or deployment during engineering/testing.

Verification disposition

This is an internal migration-test contract correction with no product render impact. Playwright and screenshots are not_applicable only for the exact one-file final diff; the independent tester must confirm that disposition from the recomputed graph. The ordinary engineer → independent tester → PM acceptance → focused commit → local no-ff merge/push → on-call lifecycle still applies.

[alexeygrigorev]

Grooming Complete

PM confirmed this is a new, one-file P0 gate regression with no existing owner. Engineering is limited to registering events/migrations/0012_shorten_event_slugs.py in the exact inventory in test_support/tests/test_migrations.py; the existing AST discovery, historical-app isolation check, and exact-set fail-closed behavior must remain intact.

Event migration/runtime/public behavior is explicitly unchanged. #253 owns the separate editorial projection digest failure, and #261 remains the first assigned recovery issue. The one-file candidate may be prepared independently, but tester-final/integration must use the accepted #253/#261 recovery base and require the full migration component green. Browser and screenshots are not applicable only if the final diff remains exactly the internal test file.

[alexeygrigorev]

Software Engineer Report

Verdict

Implementation is complete and frozen uncommitted. The exact #295 inventory contract is green, including both fail-closed mutation probes. The candidate is not tester-final ready on face8e4: #253 still owns the sole full migration-suite failure, while #261 and other already-groomed recovery issues keep graph-required quality, Django, and Playwright components red. No residual failure is accepted as a #295 pass.

Frozen candidate

• Issue: Register the event-slug data migration in the isolation inventory #295
• Worktree: /home/alexey/git/dtc-website/.tmp/issue-295-engineer
• Branch: issue-295-migration-inventory
• Base / observed HEAD: face8e4808d65afbf0374d1ced7a88079950d663
• Status: one unstaged tracked modification; no commit, push, merge, acceptance, deployment, provider, or production action
• Patch SHA-256: d5d41fbec8f46ccc591f9ae40358c3833c419fd7b2c9663d561f39a062dc1709
• Changed file: test_support/tests/test_migrations.py — 1 insertion, 0 deletions

Implemented behavior

• Added events/migrations/0012_shorten_event_slugs.py exactly once, after the existing events/0008 entry, to the exact expected data-migration set.
• The existing AST discovery, exact-set equality, assert_data_migration_isolation(path) loop, and apps.get_model source assertion are unchanged.
• Final census: 15 discovered data migrations; 0012 is discovered through its RunPython(shorten_event_slugs, ...); the expected literal occurs once.
• An instrumented execution proved the existing isolation checker receives the actual 0012 path.
• Temporary mutation probes proved both exact-set directions remain fail-closed: an unregistered numbered RunPython migration failed as “first set but not second,” and a stale expected path failed as “second set but not first.” Both probes were removed before the final tests and are absent from the worktree.
• events/, event runtime behavior, migrations, slugs, models/services/routes/templates, public/generated projections, and every other product surface are byte-unchanged by this issue.

Change-selective plan

• Plan: .tmp/verification-295/verification-plan.json
• Plan file SHA-256: 1631c687ed54d57b7f8c30ee27f80c2c708340ba064770f0a31e1da8ffea1157
• Selection SHA-256: c884d9bd245c6997566dc4a1c76c45d1004b8a0d47d63d92a10d677328f9287a
• Graph SHA-256: 05ffeee99244c9becd5e98e40be08188f1c5e2a61204daece984c3beb1d1bce6
• Source-tree manifest SHA-256: 18346c7c1e9a1f7fbd65e9b4440f751fcd5dded747f893df1edba2b027e8dbca
• Source tree OID: c8496fb2a37529eb7c63e1eeb9e35e29a2aadada
• Direct node: surface.test_support; no downstream or unknown nodes.
• Risk flag / profile: test_infrastructure / full; browser profile full.
• Render decision: impact=false, no routes or captures; render SHA-256 708da6bda70a0814bdfb3d18b83c292fb15ee295ffd62b047803704c03c71f9b.

Four disposition buckets:

• rerun: selector, quality, Django full, Playwright full, container, evidence validation — fresh full regression selected for test_infrastructure.
• reused: none.
• skipped: none in the computed plan.
• not_applicable: compatibility (outside_impact_closure), content invariants (outside_large_content_closure), screenshots (render_inputs_unchanged_without_reusable_visual).

No reuse provenance or expiry applies. Because required components terminate red on separately owned base regressions, no passing aggregate verification report exists and no report digest is fabricated. Rebase after accepted #253/#261 recovery invalidates this plan and requires recomputation.

Passing evidence on the final source

• Exact focused Django command from the issue — 1 passed; final rerun after removal of both probes is green.
• Instrumented AST/isolation execution — 0012 resolves to shorten_event_slugs, and assert_data_migration_isolation is called with its real path.
• Unexpected numbered RunPython mutation probe — expected failure names events/migrations/9999_issue_295_unregistered_probe.py as present only in the discovered set.
• Stale expected-path mutation probe — expected failure names events/migrations/9999_issue_295_stale_probe.py as present only in the expected set.
• uv run --frozen ruff format --check test_support/tests/test_migrations.py — pass, 1 file already formatted.
• uv run --frozen ruff check test_support/tests/test_migrations.py — pass.
• make lint — pass.
• make migrations-check — pass, no changes detected.
• make lock-check — pass, 63 packages resolved.
• make database-portability-check — pass.
• make security-check — pass, including canary artifact scan.
• make django-check — pass; OpenAPI, management parity, and system checks current.
• make deployment-check — pass.
• make test-ci — 607 passed.
• make verification-container VERIFY_CONTAINER_OUTPUT=.tmp/verification-295/container-check.json — pass, 13 assertions, including the full migration graph and events.0012. Artifact SHA-256 b3954adbc0731524e6d993d4c23d25d208c97a1d256118e79be796b7bb1f8647.
• git diff --check — pass; final diff is exactly one insertion in the allowed file.

Separate #253 migration failure

make test-migrations runs all 18 tests: 17 pass, 1 fails. The only failure is EditorialRouteMigrationContractTests.test_checked_manifest_is_bound_to_schema_projection_and_runtime, where required digest f399a8ed... differs from migration content digest a96bfcf.... This is the exact independently groomed #253 projection/source repair and is not caused or changed by #295. The #295 focused inventory test passes within that same component. Full migration acceptance must be rerun on the accepted #253/#261 recovery base.

Other inherited graph failures

• Full make format-check reports six untouched files; full make typecheck reports four untouched errors in content/event_speaker_bio_normalization.py. These are in Restore face8e4 formatting, typing, and focused content gates #261's recovery scope. Candidate-scoped Ruff format/lint are green; targeted mypy also reaches the same imported Restore face8e4 formatting, typing, and focused content gates #261 errors.
• make terminology-check reports frozen whole-file digest drift for content/public_projection/{podcasts,wiki,wiki_search}.json, part of Reconcile non-reproducible public projection after face8e4 #253.
• make test-django-full discovers 2,553 tests and terminates on the known Restore face8e4 formatting, typing, and focused content gates #261 pagination class-token assertion; Django's parallel runner then raises its secondary traceback-pickling error. No Register the event-slug data migration in the isolation inventory #295 path participates.
• Exact selected make test-playwright ran under an exclusive lease: 247 selected/attempted, 225 passed, 22 failed, 23 teardown errors, 3 deselected, 0 reruns, 0 quarantined, in 39m17s. Failures reproduce separately owned base clusters: stale assertions for the owner-approved Events and Community knowledgebase headings; Harden historical registration source preparation and publish aggregate event counts #112 registration wording; Fix undersized Random topic target in the homepage graph toolbar #236 33.9375px graph target sizing; Accept safe localized podcast resources without weakening external URL policy #270 localized podcast resource 500s; and Keep Spotify creator podcast embeds offline in Playwright #279 blocked Spotify embed requests. The exclusive browser lease was released at the terminal result; no foreign artifacts were inspected or accepted.

Screenshot disposition and handoff

Screenshots are not_applicable for this exact one-file internal test-contract diff: the graph proves render.impact=false and selects no route/state capture. This is an engineer disposition only; the independent tester must recompute it after the required recovery-base rebase.

The candidate remains frozen uncommitted. Tester-final must begin only after accepted #253/#261 repairs, recompute the exact plan on that new base, require all 18 migration tests green, and independently verify the one-file scope. PM acceptance, focused commit, local no-ff merge/push, and on-call remain pending.

[alexeygrigorev]

Product Manager Lifecycle Review — HOLD

Disposition: HOLD at the frozen engineering handoff. Do not send this candidate to tester-final, accept it, commit it, or integrate it yet.

I reviewed the groomed contract, the engineer report, and the frozen worktree. The implementation itself matches the authorized narrow scope:

• observed base/HEAD is face8e4808d65afbf0374d1ced7a88079950d663;
• the worktree remains uncommitted with exactly one tracked modification;
• the diff is exactly one insertion in test_support/tests/test_migrations.py;
• the inserted path is events/migrations/0012_shorten_event_slugs.py, once, after the existing events/0008 entry;
• recomputed patch SHA-256 is d5d41fbec8f46ccc591f9ae40358c3833c419fd7b2c9663d561f39a062dc1709;
• the recorded plan file exists and recomputes to SHA-256 1631c687ed54d57b7f8c30ee27f80c2c708340ba064770f0a31e1da8ffea1157.

The engineer evidence is sufficient for a frozen implementation handoff, not for a tester or PM pass. The exact focused test is reported green; AST discovery reaches 0012; the ordinary isolation checker receives the real path; and both mutation directions remain fail-closed (one unexpected numbered RunPython path and one stale expected path). The probes were removed. This supports the intended exact-inventory correction without a special case or weakening.

The required terminal envelope is not green:

• make test-migrations is 17/18, with the sole failure the separately owned Reconcile non-reproducible public projection after face8e4 #253 editorial projection digest mismatch. That attribution is credible, but PROCESS forbids treating a known residual failure as acceptance. Reconcile non-reproducible public projection after face8e4 #253 is currently open and labelled needs grooming, so the prerequisite repair is not yet an accepted base.
• Restore face8e4 formatting, typing, and focused content gates #261 remains open and owns the reported whole-repository format/type/Django pagination failures. The established recovery order remains Reconcile non-reproducible public projection after face8e4 #253, then Restore face8e4 formatting, typing, and focused content gates #261, then Register the event-slug data migration in the isolation inventory #295.
• The graph selected full Playwright because this is test infrastructure. The reported terminal result is 247 attempted, 225 passed, 22 failed, 23 teardown errors, 3 deselected, with no reruns or quarantine. The stale assertions for the owner-approved Events and Community knowledgebase headings and the separately owned Harden historical registration source preparation and publish aggregate event counts #112/Fix undersized Random topic target in the homepage graph toolbar #236/Accept safe localized podcast resources without weakening external URL policy #270/Keep Spotify creator podcast embeds offline in Playwright #279 clusters are inherited, but the selected component is still red. They cannot be waived in tester-final evidence. Re-entry therefore requires a base on which every graph-required component can terminate green, or a newly groomed contract/graph result that validly changes the requirement; attribution alone is not a pass.
• No aggregate passing verification report exists. This is truthful and appropriate for the current HOLD, but it means there is no report digest for an independent tester to validate yet.

Screenshots are provisionally not applicable for this exact one-file internal-test diff: the current graph records render.impact=false with no routes/captures. That is not PM acceptance. After the recovery-base rebase, the independent tester must recompute the graph and independently confirm the same disposition. Any additional file or render-impact change cancels N/A and restores the required screenshot gate.

Conditions to advance

1. Reconcile non-reproducible public projection after face8e4 #253 is groomed, implemented, independently tested, PM-accepted, and present in the shared recovery base with all 18 migration tests capable of passing.
2. Restore face8e4 formatting, typing, and focused content gates #261 completes its earlier assigned recovery slot and is present in that base.
3. Rebase/reconstruct this exact one-line candidate without broadening scope; freeze a new base/head and patch digest.
4. Recompute the versioned plan/report and all digests. The face8e4 plan and evidence cannot be reused across that base change.
5. A separate tester verifies exact one-file scope, AST/isolation and both fail-closed boundaries, 18/18 migration tests, and every graph-selected gate with no skip, retry, quarantine, accepted residual, or pending evidence; it also independently confirms screenshot N/A if render impact remains false.
6. Only after a complete tester PASS may PM perform user-facing final acceptance. Commit, local no-ff merge/push, and on-call remain downstream and unauthorized now.

No acceptance criterion is checked by this review. The frozen candidate should remain unchanged until the dependency/base handoff is available.