Repair sync-content verification test after staged-import helper removal

[alexeygrigorev]

Problem

The maintained Django suite cannot collect because scripts/tests/test_sync_content_verify.py imports both content_sync.tests.helpers and scripts.prod.sync_content_verify, which were intentionally removed with the retired DataTalksClub/content staged-import pipeline in 1cd976a4. scripts/content.py still exposes checkout and drift commands that dispatch to that missing verifier.

This is not a request to resurrect the retired pipeline. It is a baseline repair: remove the orphaned test/CLI surface while retaining every verification invariant that still applies to the supported database-owned sync paths.

Authority

• _docs/architecture/database-only-content.md — public content is database-owned; sync may write rows directly, but a checked-in/staged projection is not a runtime authority.
• _docs/runbooks/data-ingest.md — current ingest entry points and source ownership.
• _docs/specs/10-verification-strategy.md — maintained Django gates must collect and pass; a failed gate is not waived.
• _docs/PROCESS.md — selective evidence does not permit a required failed tier to be skipped.

Product/architecture decision

Do not restore content_sync.tests.helpers, content_sync.dtc_content, or scripts/prod/sync_content_verify.py. The removed helper was coupled to the removed dtc_content fixture tree and verifier, not a supported reusable test boundary. Restoring it would make collection green by reviving code for the wrong architecture.

The orphaned verifier test may be removed or replaced only after its still-relevant guarantees are explicitly mapped to current tests and any real gaps are ported to the owning current boundary. Cases that existed solely to compare a retired ContentRelease staged projection with that source checkout are obsolete behavior, not coverage to recreate.

Scope

• Reconcile scripts/tests/test_sync_content_verify.py with the removal commit so no maintained test imports or executes deleted staged-import code.
• Audit the old test cases and record their disposition in the implementation handoff:
  • current equivalent already covered, with the exact test named;
  • ported to the current owning test module; or
  • obsolete with the retired staged verifier, with the reason.
• Preserve/port the applicable guarantees: source scoping; discover/create/update/delete behavior; revision/checksum provenance; idempotence; fail-closed invalid input/checkout behavior; atomic failure; current offline/no-network paths; and no cross-source deletion.
• Remove the stale scripts/content.py checkout and drift command paths and their parser expectations, or retarget them only if a currently supported equivalent already exists. No command advertised by scripts/content.py --help may point at a missing module.
• Update only directly affected evergreen command/runbook references so operators are not told to invoke removed commands.
• Add a bounded regression that catches dangling production-script/test imports or advertised scripts/content.py subcommands before the full suite reaches collection.

Non-goals

• Recreating the deleted staged-import adapter, fixtures, projection comparator, or ContentRelease-based editorial workflow.
• Designing a new generic drift-reporting product or changing the community_base.content_sync API.
• Changing public content, parser output, database rows, source registrations, course curriculum ingest, or production data.
• Weakening, skipping, quarantining, or excluding the maintained Django suite.
• Refactoring unrelated legacy importers or resolving all historical runbook drift.

Dependencies and blocked work

• No open implementation dependency. Commit 1cd976a4 is historical context, not a dependency to reverse.
• This repair blocks final full-Django evidence for Make current-course lessons and homework a coherent reading flow #425 and any other candidate currently encountering the same baseline collection failure. Those candidates must rebase/freeze and rerun against the repaired baseline; their selective-CI evidence cannot waive this failure.
• Keep the change in this repository. If implementation unexpectedly requires a community-base change, stop and re-scope first; that shared package would require its own contract and both consumer suites per AGENTS.md.

Acceptance criteria

• The maintained Django test suite collects without importing content_sync.tests.helpers, content_sync.dtc_content, or scripts.prod.sync_content_verify.
• The implementation handoff contains a case-by-case mapping of the old verifier coverage to named current tests, newly ported tests, or a precise retired-only rationale; the old test is not simply skipped, renamed out of discovery, or deleted without that mapping.
• Current tests prove source-local create/update/delete, checksum/revision provenance, idempotence, invalid-input atomicity, and protection from deleting another source's rows.
• Current checkout-backed paths retain their existing dirty/revision/refusal and offline/no-network coverage; no test reaches GitHub or production data.
• uv run --frozen python scripts/content.py --help advertises only executable, supported workflows, and every advertised subcommand has a focused dispatch/parser test.
• Repository search finds no live code, tests, or operator instructions that invoke the removed verifier/helper. Historical audit prose may name it as history when clearly marked.
• No retired staged-import module, fixture tree, hardcoded public-content fallback, or checked-in projection is restored.
• Focused current sync/parser/command tests pass from synthetic local fixtures.
• uv run --frozen python scripts/ci.py test collects and passes in full; no exclusion, quarantine, or collection-ignore is added.

Repository and operations scenarios

1. From a clean test environment, collect and run the focused sync/parser/command tests; they use synthetic fixtures and no network or operator staging tree.
2. Exercise every scripts/content.py subcommand parser/dispatch contract and confirm none resolves a deleted file.
3. Run the full maintained Django tier and confirm collection reaches execution and passes.
4. Inspect the coverage-disposition record against every former test class (Agreement, Drift, MediaAsymmetry, RevisionStatus, Scope, Refusal, WorkingCopyIndependence, CheckoutPlan, ReadOnly, ReportShape, Convention). Reject unexplained loss.

Browser scenarios

Not applicable. This issue changes test/CLI wiring and directly affected operator documentation only; it must not alter a rendered route. Per _docs/PROCESS.md, the tester should classify screenshot evidence as not_applicable with this reason and run the applicable backend-only Playwright smoke tier selected by the frozen verification graph.

[alexeygrigorev]

Grooming Complete

PM decision: do not restore the deleted staged-import helper or verifier. This issue now requires a case-by-case disposition of the old verification suite, ports every still-live invariant to the current owning boundary, retires dangling scripts/content.py commands, and proves full Django collection/execution without skips. Priority is P0 because it blocks required delivery gates.

No open dependency. #425 and other affected candidates must rerun their frozen verification evidence after this baseline repair.

[alexeygrigorev]

Issue #427 engineer handoff

Scope and architecture decision

The retired staged ContentRelease comparator remains deleted. This candidate
removes its orphaned Django test and the singular scripts/content.py checkout
and drift wrappers. It does not restore content_sync.tests.helpers,
content_sync.dtc_content, scripts.prod.sync_content_verify, a staged fixture
tree, or a file-backed projection.

The supported owners are now explicit:

• editorial source rows: community_base.content_sync plus
  content/sync_parsers/ and SyncedDocument;
• course repositories: scripts/content.py sources, pull-plan, checkouts
  and pull, backed by the existing course-repository transport/import tests.

Former verifier case disposition

AgreementTests

• test_a_matching_catalogue_is_clean_and_exits_zero: obsolete report/exit
  behavior. The staged served-vs-checkout report no longer exists. Current
  convergence is covered by
  content.tests.test_sync_parsers.PeopleParserTests.test_discover_upsert_update_and_delete
  and the equivalent article/podcast/book parser tests.
• test_the_report_is_indented_sorted_json_with_no_record_body: obsolete
  report-shape behavior; no current command emits that report.
• test_an_article_that_moves_with_its_bytes_is_not_drift: obsolete drift
  comparator behavior. Current identity/checksum semantics are owned by the
  parser's stable key and checksum tests, including
  PeopleParserTests.test_discover_upsert_update_and_delete.

DriftTests

• test_an_upstream_only_article_is_missing: the old missing bucket is
  obsolete. Current create behavior is covered by
  ArticlesParserTests.test_discover_upsert_and_delete.
• test_a_served_book_with_no_upstream_file_is_extra: the old extra bucket
  is obsolete. Current source-local deletion is covered by
  BooksParserTests.test_discover_upsert_and_delete.
• test_changed_podcast_bytes_are_mismatched_and_nothing_else_moves: the old
  mismatched report is obsolete. Current checksum-driven update and source
  isolation are covered by
  PodcastsParserTests.test_discover_upsert_update_and_delete and the newly
  added PeopleParserTests.test_deletion_is_scoped_to_the_owning_source_and_kind.
• test_a_transcript_drifts_independently_of_its_episode: the report bucket is
  obsolete; episode/transcript parsing and checksum-driven updates remain in
  PodcastsParserTests.test_discover_upsert_update_and_delete.

MediaAsymmetryTests

• All three cases (unreferenced upstream, served image deleted, changed image bytes) described the retired staged projection's asymmetric report.
  That projection rule is obsolete. Current media create/update/delete and
  provenance are covered by
  MediaParserTests.test_discover_upsert_and_delete; source-local deletion is
  additionally protected by the new PeopleParserTests scoping regression.

RevisionStatusTests

• tracking head clean, reachable older revision behind, and revision upstream never had is unreachable were fields of the retired report.
  Current revision provenance is asserted in
  PeopleParserTests.test_discover_upsert_update_and_delete and the other
  parser tests. The still-supported public-commit boundary is covered by
  SourceCommitReachabilityTests.test_commit_on_the_public_remote_branch_is_reachable,
  test_commit_that_exists_only_on_local_main_is_not_reachable, and
  test_unknown_commits_and_broken_checkouts_are_not_reachable.

ScopeTests

• test_records_owned_by_another_repository_are_ignored: live guarantee,
  covered by each parser's test_ignores_other_sources* and strengthened by
  the new PeopleParserTests.test_deletion_is_scoped_to_the_owning_source_and_kind.
• test_a_draft_upstream_is_not_reported_as_missing: the report assertion is
  obsolete; selection rules remain parser-owned and are exercised by their
  discover tests.

RefusalTests

• Wrong-repository, relative checkout, symlink checkout, unresolvable revision,
  missing editorial source, missing active release, and no-target argparse
  cases belonged to the removed verifier entry point and are obsolete there.
• Current checkout refusal remains covered by
  scripts.tests.test_checkout_refresh (wrong_origin, linked_worktree,
  outside_scratch_root, non_git_directory, malformed plan) and
  content_sync.tests.test_sync_course_repositories (dirty_checkout,
  branch_mismatch, missing_checkout, unknown_stable_id).
• Current invalid authored content is fail-closed without changing an existing
  row in the newly added
  ArticlesParserTests.test_invalid_input_leaves_existing_source_rows_unchanged.
  Course-import atomicity remains covered by
  CurriculumImportServiceTests.test_missing_project_rejects_atomically_with_redacted_diagnostic
  and test_history_breaking_question_change_rolls_back_all_source_updates.

WorkingCopyIndependenceTests

• The retired verifier intentionally ignored a dirty working copy. The current
  checkout-backed course path instead refuses dirty input by default in
  PullCourseRepositoriesCommandTests.test_a_dirty_checkout_is_refused_and_names_the_change;
  its explicit waiver still imports the committed snapshot, proven by
  test_a_waived_dirty_checkout_still_imports_the_commit.

CheckoutPlanTests

• Both singular editorial checkout-plan cases are obsolete with that command.
  The supported registered-data plan is covered by
  PullCourseRepositoriesCommandTests.test_the_repository_list_is_registered_data_not_a_hardcoded_list,
  test_naming_one_checkout_names_the_run, and
  test_from_disk_keeps_an_explicit_checkout_as_an_override.

ReadOnlyTests

• test_a_drifted_run_leaves_every_row_exactly_as_it_found_it: obsolete
  read-only report behavior. The current invalid-input no-mutation guarantee is
  ported to ArticlesParserTests.test_invalid_input_leaves_existing_source_rows_unchanged.
• test_the_script_names_no_networked_git_subcommand and
  test_a_full_run_completes_with_no_reachable_remote: the removed verifier has
  no successor command. Current pull offline behavior is covered by
  _TransportParityContract.test_the_pull_transport_makes_no_network_call;
  checkout refresh is the separately tested network boundary.

ReportShapeTests

• Bucket truncation and media asymmetric-bucket cases are obsolete because the
  report was retired and no supported equivalent exists.

ConventionTests

• The removed module's sync-model declaration and database-selected editorial
  source were entry-point conventions for a module that no longer exists.
  scripts/tests/test_content.py now enumerates exactly the four supported
  commands, drives every dispatch, verifies every dispatched script exists,
  and scans live Python for imports of the three retired module boundaries.

Live invariant summary

• create/update/delete: parser test_discover_upsert*_and_delete cases;
• checksum/revision provenance: people, media, podcast, book, docs, FAQ,
  podwiki, platform, Slack and course parser assertions;
• idempotence: repeated upsert assertions plus
  CurriculumImportServiceTests.test_exact_source_commit_and_parser_replay_is_a_no_op;
• invalid-input atomicity: new article no-mutation regression plus curriculum
  import atomic rollback tests;
• no cross-source deletion: new source-and-kind scoping regression;
• checkout refusal/revision/offline behavior: checkout refresh, source commit
  reachability, course repository pull, and transport parity tests named above.

Verification

• uv run --frozen python scripts/content.py --help: pass; advertises only
  sources, pull-plan, checkouts, pull.
• uv run --frozen pytest scripts/tests/test_content.py content/tests/test_sync_parsers.py -q:
  51 passed, 13 subtests passed.
• focused run including adoption baseline coverage:
  55 passed, 13 subtests passed.
• uv run --frozen python scripts/ci.py test-content:
  762 tests passed (6 skipped, 1 expected failure).
• clean-environment maintained Django collection
  (env -u DTC_SQLITE_PATH -u DTC_ENVIRONMENT -u PUBLIC_MEDIA_STORE_BACKEND -u PUBLIC_MEDIA_S3_BUCKET -u PUBLIC_MEDIA_S3_REGION uv run --frozen python scripts/ci.py test): 4,314 tests passed (7 skipped, 1 expected failure).
• uv run --frozen ruff check scripts/content.py scripts/tests/test_content.py content/tests/test_sync_parsers.py:
  pass.

Selective verification plan

The frozen uncommitted worktree produced a full profile because the change
touches test infrastructure. Base and head are both
1cd976a4964134a28877799153d3bc3df9078041; the worktree source-manifest SHA
is c8dc498049907c3c560671e942273062fbea81653da311234651c1aecbf8755a and
the ownership-graph SHA is
8e961161a3862302b27538c12adb5bb8370716487038024cf1738e08b30db91e.

Every required component is a fresh rerun:

• container — scripts/ci.py verification-container, input
  955443226b3472ab70eb48fef399824a75bbe753fe8c3012133a14f6b8daa29d;
• Django — scripts/ci.py test-django-full, input
  7f2e2dba133ba23829d7b6493aea75e3d7fcc3cad711be65b5b63bda5d92d34d;
• evidence validation — scripts/ci.py test-ci, input
  a297c75be96048ea1620894933b3675de9d03f20a4e1089dcfa31870b5e31e42;
• Playwright smoke — scripts/ci.py test-playwright-smoke, input
  dba01f59ba80353d7bf32e35c2318a651130599121e2895d011c8e78a551b2ef;
• quality — quality-contract-v3, input
  a593c44ca41c0a48d591074243d2b7037eff5ab7234b7ead339262c5e7d5dfe4;
• selector — scripts/ci.py verification-plan, input
  a297c75be96048ea1620894933b3675de9d03f20a4e1089dcfa31870b5e31e42.

No component is reused or skipped. Screenshots alone are not_applicable
(render_inputs_unchanged_without_reusable_visual; input
15b669feaa62209c04f2c725b29d158ae210e26441cf5bfee003d58a6481010d).
The browser profile is smoke.

Screenshots are not_applicable: this internal CLI/test/runbook repair has no
rendered-route impact.

[alexeygrigorev]

QA Review

Verdict: FAIL

Frozen candidate: 1cd976a4964134a28877799153d3bc3df9078041 with exactly these uncommitted paths: _docs/runbooks/data-ingest.md, _docs/runbooks/ingest-script-inventory.md, ci/adoption_baseline.txt, content/tests/test_sync_parsers.py, scripts/checkout_refresh.py, scripts/content.py, scripts/tests/test_content.py, and deleted scripts/tests/test_sync_content_verify.py.

Selective-plan verification

• Independently recomputed with clean ambient runtime/S3 variables and VERIFY_CONSUMER=tester.
• Engineer and tester plans differ only in created_at; base/head, eight changed paths, full/test-infrastructure profile, smoke browser profile, graph SHA 8e961161a3862302b27538c12adb5bb8370716487038024cf1738e08b30db91e, source manifest SHA c8dc498049907c3c560671e942273062fbea81653da311234651c1aecbf8755a, render digest, dispositions, environments, and every component input digest match.
• Tester plan: .tmp/issue-427/tester-verification/verification-plan.json, SHA-256 989fa982da59149d4ebe84835a9a3e918b5f84bde649b56a33f7b9225f6abc14.
• Tester-final report: .tmp/issue-427/tester-verification/verification-report.json, SHA-256 dd1b9243ce0297789fbf7375a93ea48e727e5232e7ed25f6ec7ebd2f127146a9; verdict failure. Buckets: rerun 6, reused 0, skipped 0, not-applicable 1.
• Screenshots: not_applicable as planned because the candidate has no render-input change.

Passing evidence

• Focused current owners: uv run --frozen pytest scripts/tests/test_content.py content/tests/test_sync_parsers.py scripts/tests/test_checkout_refresh.py content_sync/tests/test_source_commit_reachability.py content_sync/tests/test_sync_course_repositories.py content_sync/tests/test_course_repository_transport_parity.py courses/tests/test_curriculum_import_service.py -q -> 107 passed, 13 subtests passed.
• uv run --frozen python scripts/content.py --help advertises only sources, pull-plan, checkouts, pull.
• Full Django: 4,314 passed at the maintained runner level (4,307 passed, 7 skipped, 1 expected failure in Django output), exit 0.
• Evidence/CI contracts: 671 passed, exit 0.
• Container: 13 assertions, pass, exit 0.
• Selector: pass, exit 0.
• Case-by-case engineer mapping accounts for every former verifier test class and does not restore the deleted helper, verifier, fixture tree, or projection.

Blocking findings

1. Changed code fails the required quality gate.

   • Reproduction: uv run --frozen ruff check scripts/content.py scripts/tests/test_content.py content/tests/test_sync_parsers.py scripts/checkout_refresh.py.
   • Actual: E501 at scripts/checkout_refresh.py:5, 115 characters where the maximum is 100. The fresh quality-contract-v3 component fails for the same reason.
   • Expected: all changed Python passes the maintained quality contract.

2. The required Playwright smoke component is red.

   • Command: uv run --frozen python scripts/ci.py test-playwright-smoke through the tester plan.
   • Actual: 25 passed, 5 failed, 321 deselected. Failures are both desktop/mobile test_public_home_and_hubs cases expecting the removed CLIMB_HEADING, both docs/FAQ slash-contract cases expecting one /faq/ link but finding zero, and test_all_public_hub_aliases_redirect_once_with_query, where /blog/ai-dev-tools-zoomcamp?source=browser returns 404 instead of 301.
   • Expected: the selected required smoke tier passes. These are baseline/render-contract failures rather than changes introduced by this non-render candidate, but process policy does not permit waiving a required red tier.
   • Artifact: .tmp/issue-427/tester-verification/evidence/playwright-output.log, SHA-256 f1f89e88b6cf17dd6f5c0d8dda94646873d8c05fe932f2e8d6fb69e359b8c141.

3. The directly edited ingest runbook remains internally inconsistent and names deleted modules as current architecture.

   • _docs/runbooks/data-ingest.md:1163-1167 says the current unwired push-sync half is prepare_dtc_content_candidate under deleted content_sync/dtc_content/...; lines 1169-1182 immediately say current editorial synchronization is direct through community_base.content_sync and SyncedDocument.
   • Expected: affected operator guidance consistently describes the supported direct-to-database owner, while retired module names appear only in clearly historical context.

4. The new invalid-input test does not prove current sync-path atomicity.

   • content/tests/test_sync_parsers.py:342-374 calls only pure parser.discover(...), then observes that the database row it never asked any sync/upsert owner to mutate remains unchanged.
   • Expected under the acceptance criterion: exercise the current owning sync boundary on malformed authored input and prove the relevant source rows are unchanged, or explicitly narrow the groomed criterion if parser-discovery purity is the intended guarantee. The present test name/handoff claim overstates what it executes.

No source file was edited, committed, or pushed by QA.

[alexeygrigorev]

Engineering retest update — baseline dependency required

The four #427-owned QA findings are repaired in the isolated worktree:

• changed Python passes the maintained quality contract, including Ruff and
  mypy;
• the changed ingest runbook now describes the deleted
  content_sync/dtc_content preparation path only as historical code and names
  community_base.content_sync plus SyncedDocument as the current owner;
• malformed authored input is now exercised through
  sync_content_source(...), the real orchestration boundary, and the test
  proves existing article rows and last_synced_commit remain unchanged while
  the source records a partial result;
• the five smoke failures were traced to stale assertions for already accepted
  baseline changes: homepage copy (78ae0c9f), FAQ's removal from global
  navigation (52814b33), editorial alias retirement (c5d1a436), and the
  current six-card course-family hub.

The exact smoke repair passes the required tier: 30 passed, 321 deselected,
with no reruns or quarantines. Quality passes, including 671 CI/control-plane
tests. Full Django passes 4,314 tests (7 skipped, 1 expected failure).

There is one process blocker before an independent retest can be requested.
Keeping the stale-smoke repair inside #427 adds
playwright_tests/test_foundation_smoke.py to this candidate, so the selector
correctly escalates it from a non-render smoke plan to render_impact, full
Playwright and eight screenshot captures. The required full tier is red on a
pre-existing baseline defect: the exact accessibility-registry probe fails
because the historical registration preview returns 404 and synthetic author
images return 502, matching the independently reported #420 baseline failure.
No waiver, skip, xfail or classifier exception is proposed.

The smoke repair must therefore land as an independent baseline dependency.
After #427 is rebased on that commit, its owned non-render changes can be
refrozen, the selective plan regenerated, and QA retest requested. No #427
commit or push has been made.