Add explicit pre-2024 cohort mapping, remove Cohort.save() regex fallback, import ml-zoomcamp-2021 certificates

[alexeygrigorev]

Parent decision: #15
Normative decision: Open decision 4

Scope

The historical import added on main (2032ceb, "Add historical zoomcamp-scoring import for pre-2024 cohorts") creates each pre-2024 Cohort via Cohort.objects.update_or_create(slug=edition.cohort_slug, ...) without setting course= explicitly, so it still resolves its family through the regex year-stripping fallback in Cohort.save() (courses/models/cohort.py). #15 requires an explicit reviewed mapping instead, per owner decision.

• Add explicit reviewed entries (in courses/course_family_catalog.py or an equivalent explicit table read by scripts/historical_import) for the 7 pre-2024 legacy cohort slugs: de-zoomcamp-2022, de-zoomcamp-2023, mlops-zoomcamp-2022, mlops-zoomcamp-2023, ml-zoomcamp-2021, ml-zoomcamp-2022, ml-zoomcamp-2023. This is the confirmed complete pre-2024 inventory — no other one-off/unusual legacy editions exist.
• Update scripts/historical_import/scoring_import.py's ensure_cohort() to pass course= explicitly from that mapping instead of leaving course_id unset.
• Remove the regex year-stripping fallback in Cohort.save() now that every production cohort-creation path supplies course explicitly. Regex-based slug inference is only acceptable as a one-time authoring aid inside import/fixture scripts — never live in production model code.
• Locate and import certificates for the ml-zoomcamp-2021 edition. The current import treats it as having no certificate source and skips it; owner confirms certificates for this edition exist and should be found and imported like every other edition.
• Persist the recovered sha1(email) hash (already used transiently by scripts/historical_import/email_recovery.py to match learners to accounts) as a queryable legacy identifier/alias, so historical records stay traceable back to the zoomcamp-scoring source.

Non-goals

Does not reopen the mapping approach itself (approved in #15) or add pre-2024 families beyond DE/MLOps/ML Zoomcamp 2021-2023.

Acceptance criteria

• All 7 pre-2024 cohort slugs resolve their family via the explicit mapping table, not regex inference.
• Cohort.save() no longer contains the regex-based family-inference fallback; creating a Cohort with no course and an unmapped slug fails loudly instead of guessing.
• ml-zoomcamp-2021 certificates are imported like every other historical edition.
• Each imported historical learner record carries its source sha1(email) as a reviewable legacy alias.

Dependencies

Depends on #15 (decision, resolved). Builds on the historical import added in 2032ceb on main.

[alexeygrigorev]

Independent Tester Report — issue #224

Status: candidate-specific PASS / not ready for product-manager acceptance.

The #224 behavior and data-contract checks pass independently. The full release envelope is not
green because current main has an existing homepage carousel overflow/source-readability
regression. No commit, push, merge, deployment, or production configuration change was made.

Candidate refresh — 2026-08-27

The engineer and tester snapshots were synchronized after applying the
formatter to the two pre-existing violations in content/tests/test_review_skeleton.py, a file
already modified by this candidate. The refreshed candidate passes make format-check (618 files),
the 17-test historical-import/adoption suite, the 931-test core target, and the 607-test CI target.
The candidate-specific behavior is unchanged; the full release envelope remains blocked by the
unrelated #179 homepage regression.

Candidate under test

• Worktree: .tmp/issue-224-independent-tester-current-20260826
• Base and source revision: d24808c374b7deb220cb57e365760382ad66d06b
• Candidate: staged, uncommitted diff on that revision
• Scoped diff: 134 paths, 970 insertions, 385 deletions
• Staged patch SHA-256 (git diff --cached --binary --full-index HEAD | sha256sum):
  0e20583500a42a6afe122761fe107f0cd60ea3447499a0d70febae8c9f1b0531
• git diff --cached --check: passed
• Frozen engineer snapshot: .tmp/issue-224-engineer-current-20260826

Verification plan

• Plan: .tmp/verification/issue-224-independent-tester-current-20260826/verification-plan.json
• Plan structural/render digest: 8e41351642b76bd275a888d827df1dca4ab7ca9c09b22a442dad4eac81340542
• Ownership graph SHA-256: 05ffeee99244c9becd5e98e40be08188f1c5e2a61204daece984c3beb1d1bce6
• Render digest: f60a65c49c273fcddd1d54a597daa96c2ed4c7c35f2837113ecc6d59df808eba
• Profile: full; reason: cross_application
• Required screenshot matrix: home, not-found, courses, and Studio sign-in at desktop 1440x900
  and mobile 390x844.

Issue-specific verification

• Historical-import plus adoption Django suite: 17 passed.
• Additional full courses application suite: 733 passed.
• CI/evidence suite: 607 passed.
• Core account/API/studio matrix: 931 passed.
• Issue-related browser subset: 20 passed.
• Container verification: pass, all 14 assertions passed, including the new migration.
• Lint, typecheck (448 source files), migrations, links (7 passed), SEO (76 passed),
  OpenAPI/parity/Django checks, deployment check, terminology, database portability, and security:
  pass.

Full-gate results

• Full Django discovered 2,520 tests and exited 2 on the existing
  templates/core/home.html:1297: adjacent nested opening tags failure; Django then emitted its
  secondary cannot pickle 'traceback' object multiprocessing error. The candidate does not
  touch that template. The same source-readability test fails on clean current main.
• make format-check now passes; the only prior formatter findings were the two locations in
  content/tests/test_review_skeleton.py (lines 173 and 195), which are included as a mechanical
  cleanup because that file is already candidate-owned.
• Full Playwright selected 238 local tests and was bounded at 600 seconds. Before termination it
  observed three accessibility failures: test_accessibility_core_smoke,
  test_complete_accessibility_registry, and test_javascript_off_public_reads_remain_semantic,
  all on the public homepage, and reached 12% without a Add explicit pre-2024 cohort mapping, remove Cohort.save() regex fallback, import ml-zoomcamp-2021 certificates #224-specific failure.
• A clean-current-main reproduction of those exact three browser cases failed identically in
  159.47 seconds: the homepage document is 2,310 CSS px wide at a 1,440 px viewport, with
  offscreen testimonial cards extending about 870 px beyond the viewport. At mobile width the
  diagnostic capture measured document width 1,661 px for a 390 px viewport. This belongs to the
  already-merged homepage carousel lane, not the Add explicit pre-2024 cohort mapping, remove Cohort.save() regex fallback, import ml-zoomcamp-2021 certificates #224 patch.

Screenshot inspection

The official screenshot guard stopped at the first home capture with
RuntimeError: screenshot route has horizontal overflow. After collecting the ignored local
static bundle and using a static-serving local server wrapper for diagnosis, all eight scratch
captures were produced under:

.tmp/verification/issue-224-independent-tester-current-20260826/visual-inspection/screenshots/

All eight were visually inspected. The 404 and Studio sign-in states are branded and responsive;
/courses has no overflow. The home captures are not acceptance evidence because the offscreen
carousel cards fail the no-horizontal-overflow contract.

Required next step

No further candidate code change is indicated by this verification. The homepage carousel owner
should clear the existing source-readability and responsive-overflow failures, then rerun the full
tester plan. After all required evidence is green, obtain product-manager acceptance before any
commit or local merge.

[alexeygrigorev]

Tester addendum — issue #224

The previously posted #224 tester report has been locally refreshed with the
validated plan fingerprint and disposition record.

• Exact synchronized staged patch: 0e20583500a42a6afe122761fe107f0cd60ea3447499a0d70febae8c9f1b0531
• Verification plan SHA-256: 6e950bb126771ef65f140c1a69c23f5acb959db857e12ae78ccdd83963932b11
• Base/head: d24808c374b7deb220cb57e365760382ad66d06b / d24808c374b7deb220cb57e365760382ad66d06b
• Plan validation: passed
• Dispositions: rerun for selector, evidence validation, quality, Django,
  Playwright, container, and screenshots; not_applicable for compatibility and
  content invariants.

The fresh 17-test historical-import/adoption recheck and 618-file format check
also passed. This does not change the existing candidate-specific PASS or the
full-envelope #179 homepage blocker; the candidate remains staged and
uncommitted pending PM acceptance.

[alexeygrigorev]

Current tester addendum — repair 2 (2026-08-27)

The earlier #224 comment describes a superseded candidate. The synchronized
current-main engineer/tester snapshots have now been rechecked after the
changed invalid-rejection factory was given its explicit Course family and the
normative #15 implementation-gap text was reconciled.

• Base: d24808c374b7deb220cb57e365760382ad66d06b
• Engineer/tester candidate: staged and uncommitted; 135 paths, 976 insertions, 390 deletions
• Exact staged patch SHA-256: 7358c5c1be3a015c680098d7d4edaf3844b63c456f92fd2f4f30ea6b794171c4
• Recheck: 17 historical-import/adoption tests, 18 current-domain factory tests, and make format-check (618 files) passed
• No candidate-specific failure was found; the full release envelope remains blocked by the unrelated protected Rebuild the public homepage on the design 5a mockup #179 homepage regression.

This is a candidate-specific tester pass, not PM acceptance or merge approval.
The candidate remains staged/uncommitted pending the required lifecycle gates.

[alexeygrigorev]

PM acceptance — issue #224 (bounded candidate slice, 2026-08-27)

Disposition: ACCEPTED for the candidate-specific mapping/import behavior; #224 remains OPEN.

I reviewed the groomed issue, the current-main staged candidate, the independent tester report,
and the required route captures against Open decision 4. The candidate provides the approved
fail-closed data contract:

• all seven confirmed pre-2024 cohort slugs use an explicit reviewed course-family mapping;
• production model save no longer guesses a family by stripping a year, and unmapped creation
  fails loudly;
• the ml-zoomcamp-2021 certificate source is included in the historical import path; and
• recovered sha1(email) values are retained as queryable legacy identifiers for reconciliation.

The independent verification found no candidate-specific failure: 17 historical-import/adoption
tests, 18 current-domain factory tests, 733 courses tests, 931 core tests, 607 CI tests, 20
issue-related browser cases, container/migration/type/lint/OpenAPI/parity/security checks, and
the refreshed format check passed. The /courses desktop/mobile captures were readable and
contained; the separate home failure is the known shared #179 regression.

This is product acceptance of the candidate slice, not release-envelope, production-import, or
merge approval. #224 stays open while the protected homepage baseline is unresolved and the full
tester plan still needs its required rerun. The candidate remains staged and uncommitted; no
commit, merge, push, deployment, or production data import was authorized by this comment.