Make event timeline tests deterministic when the upcoming catalogue is empty

[alexeygrigorev]

Parent/reference: #228 (scheduled/full-regression recovery)
Related release evidence: #253, #261, #295
Related Event contracts: #45, #289

Problem

The event timeline tests assume that the checked projection always contains an upcoming
Event. That assumption is false once the wall clock passes the latest checked event. On the
current release base origin/main at
face8e4808d65afbf0374d1ced7a88079950d663, the newest checked Event starts at
2026-08-31T17:00:00+02:00. With the current date at or after that instant,
event_groups() correctly returns an empty upcoming tuple, but several tests index
upcoming[0] or require an explanation pill to be present.

The focused reproduction on this base is:

uv run --frozen pytest content/tests/test_events_timeline.py -q
9 failed, 30 passed, 455 subtests passed

The same deterministic-test defect is visible in scheduled full regression
run 33441614182. The
parallel Django traceback/pickle error is a secondary reporting effect; this issue owns only
the event-timeline test assumptions.

Exact affected tests

All affected tests are in content/tests/test_events_timeline.py:

• EventTimelineRouteTests.test_clean_hub_is_upcoming_and_past_filter_is_paginated
  (indexes groups.upcoming[0] after rendering with an uncontrolled clock);
• EventIndexDesignSystemTests.test_rows_are_built_from_the_shared_design_system_primitives;
• EventKindsExplainerTests.test_both_views_explain_every_kind_a_row_can_carry;
• EventKindsExplainerTests.test_every_kind_pill_is_bound_to_its_own_explanation;
• EventDetailDesignSystemTests.test_a_live_podcast_recording_keeps_the_mint_pill_the_index_gives_it;
• EventDetailDesignSystemTests.test_an_event_without_speakers_or_links_draws_neither_of_them;
• EventDetailDesignSystemTests.test_detail_carries_its_own_stylesheet_and_loads_no_legacy_css;
• EventDetailDesignSystemTests.test_the_kind_pill_and_the_state_word_match_the_index; and
• EventDetailDesignSystemTests.test_the_page_is_built_from_the_shared_design_system_primitives.

Some failures are reported as parametrized subfailures or as follow-on errors after the first
empty-catalogue assumption; the complete list above is the affected contract surface.

Normative authority

• _docs/PROCESS.md,
  for independent engineer/tester/PM gates and evidence reuse;
• Specification 05 — Events, registration, and email,
  for the read-only Event projection and event lifecycle boundary; and
• Specification 10 — Verification strategy,
  for deterministic synthetic fixtures, isolated test databases, and release-critical test
  coverage.

Scope

Repair only the event-timeline test fixtures/clock control and add explicit empty-upcoming
coverage. The implementation must make the tests describe the public contract independently of
the calendar date.

The non-empty assertions may use an injected aware instant and a deterministic checked record
or test projection. The route request and the expected EventGroups must use the same instant;
tests must not depend on whichever Event happens to be upcoming on the day they run. Tests that
need a representative event must select one from the controlled fixture rather than indexing a
wall-clock-derived production tuple.

Add an explicit empty-upcoming scenario that supplies EventGroups((), recent) (or an
equivalent deterministic test seam) to the public events view and verifies the existing public
empty state is a successful, bounded response containing No upcoming events. and no fabricated
event/kind assertions. The scenario must also prove that the past archive remains independently
renderable when the upcoming side is empty, if that behavior is exercised by the fixture.

Production event grouping, projection data, templates, URL patterns, cache policy, SEO metadata,
event ordering, lifecycle behavior, and all non-test source files remain unchanged. Do not alter
the latest event date merely to keep a test green.

Explicit non-goals

• No change to content/public_data.event_groups, Event projection JSON, Event model, route,
  template, CSS, JavaScript, or public copy.
• No synthetic future production record, date extension, clock shim in application code, or
  weakening/removal of an assertion.
• No fix for Reconcile non-reproducible public projection after face8e4 #253 projection/source attestation, Restore face8e4 formatting, typing, and focused content gates #261 collection-hub pagination, Register the event-slug data migration in the isolation inventory #295 migration
  inventory, or the secondary multiprocessing traceback/pickle reporting behavior.
• No browser/provider/network, database migration, deployment, source-repository, commit, push,
  or issue-closure action outside the normal lifecycle.

Dependencies and ordering

• Implementation dependency: none. This test-only correction can be engineered from the
  clean current origin/main base without waiting for Reconcile non-reproducible public projection after face8e4 #253; the content-source repair is not
  part of this issue.
• Release relationship: Reconcile non-reproducible public projection after face8e4 #253, Restore face8e4 formatting, typing, and focused content gates #261, and Register the event-slug data migration in the isolation inventory #295 remain separate release gates. Their status may
  keep the aggregate pipeline red, but must not be used to excuse a failing Make event timeline tests deterministic when the upcoming catalogue is empty #299 criterion.
• Event ownership: Establish the non-activating Event lifecycle and calendar core #289 owns the future non-activating Event domain core and Implement the event lifecycle, public pages, people, and calendar contracts #45 owns the
  eventual public Event lifecycle/pages/calendar contract. This issue changes neither.

Acceptance criteria

• The nine affected tests above pass on the current origin/main base at a clock after
  2026-08-31T17:00:00+02:00 and at a clock before the newest checked event, without changing
  production event grouping or checked projection data.
• Every assertion requiring an upcoming record obtains it from a deterministic aware clock
  and fixture/projection seam; no affected test indexes a tuple whose membership depends on the
  host wall clock.
• The route test uses one controlled instant for grouping and requests, and still verifies
  upcoming/past separation, pagination, escaped titles, canonical output, and status codes.
• An explicit empty-upcoming test verifies GET /events is a 200 response with the existing
  No upcoming events. state and no exception, fabricated kind explanation, or invalid item
  access; the past archive remains covered independently where applicable.
• The focused suite passes with an exact recorded command and count:
  uv run --frozen pytest content/tests/test_events_timeline.py -q.
• The engineer generates and freezes the versioned verification plan for the exact base/head,
  with all component dispositions recorded. The independent tester recomputes that plan and
  validates the focused result plus the graph-selected shared checks; no required check is
  silently skipped.
• Because this is test-only and has no product-page render change, screenshots are marked
  not_applicable with the graph/render-impact proof. The tester still verifies the Django route
  scenarios and runs the graph-selected Playwright tier if required by the plan.
• Engineer, independent tester, and PM reports are posted to this issue before any commit;
  only the orchestrator may merge/push after PM acceptance, and on-call alone observes the
  resulting CI run.

Required verification scenarios

1. Run the affected tests with an injected aware now before the newest checked event and prove
   the representative upcoming record and all kind/detail assertions remain meaningful.
2. Run the same tests with an injected aware now after the newest checked event and prove the
   empty-upcoming path is intentional and bounded rather than an accidental IndexError.
3. Exercise the public /events view with controlled non-empty and empty EventGroups; verify
   200 status, the existing empty copy, no invalid controls, and no cross-view leakage.
4. Run the exact focused command, the generated graph-selected Django/CI checks, and any
   required Playwright tier from the frozen verification plan. Report the known independent
   Reconcile non-reproducible public projection after face8e4 #253/Restore face8e4 formatting, typing, and focused content gates #261/Register the event-slug data migration in the isolation inventory #295 failures separately if they remain on the shared base.

Lifecycle

This is a groomed P0 testing/bug issue. The engineer works in an isolated uncommitted worktree
from the exact current base and posts a frozen handoff. A separate tester reviews the diff,
recomputes the plan, runs the controlled before/after-clock and empty-upcoming scenarios, and
records screenshot not_applicable evidence. PM accepts only after all criteria and evidence
are complete. The engineer then commits with Closes #299; the orchestrator locally merges with
--no-ff, pushes main, and dispatches on-call. No pull request is created.

[alexeygrigorev]

Grooming Complete

Verdict: GROOMED / P0 / testing-only release blocker.

The defect is independently reproduced on origin/main
face8e4808d65afbf0374d1ced7a88079950d663 with:

uv run --frozen pytest content/tests/test_events_timeline.py -q
9 failed, 30 passed, 455 subtests passed

The newest checked event is 2026-08-31T17:00:00+02:00; after that instant
event_groups().upcoming is correctly empty, exposing nine tests that assume a
representative upcoming record or kind explanation. The issue lists every affected test and
requires controlled aware-clock/fixture coverage plus an explicit empty-upcoming route scenario.

Scope is limited to content/tests/test_events_timeline.py (and directly necessary test-only
support). Production grouping/projection/routes/templates/copy remain unchanged. There is no
implementation dependency on #253; #253, #261, and #295 are separate release gates. #261, #289,
and #45 do not own this deterministic test assumption, and the issue-search audit found no
duplicate owner.

Next handoff: engineer → independent tester → PM. The engineer must leave an uncommitted frozen
diff and verification plan; tester must recompute the plan and run before/after-clock plus
empty-upcoming scenarios; PM accepts only complete evidence. Screenshots are not_applicable
only with render-impact proof.

[alexeygrigorev]

Software Engineer Report — #299

Scope and frozen source

• Issue: #299
• Role: software engineer
• Worktree: /home/alexey/git/dtc-website/.tmp/issue-299-engineer
• Base: face8e4808d65afbf0374d1ced7a88079950d663
• Head: face8e4808d65afbf0374d1ced7a88079950d663 (uncommitted worktree changes)
• Source mode: worktree
• Source tree digest: 4ca5a91883ccd480269ed536c14b1d01d7d5a84cb4a9e62f69d2d4d703f5cb59

The only changed tracked file is content/tests/test_events_timeline.py. Production event
grouping, projection data, views, routes, templates, styles, and copy are unchanged.

Implementation

• Added a shared test-only DeterministicEventTimelineTestCase with an aware
  2026-08-12 Europe/Berlin instant and a patched public-view EventGroups fixture.
• Updated timeline/index/detail expectations to read that same fixture instead of an uncontrolled
  event_groups() call.
• Added an explicit 2026-09-01 Europe/Berlin empty-upcoming route scenario. It verifies the
  existing No upcoming events. 200 response, no event/kind markup or exception, and an
  independently renderable past archive.
• No migration, browser fixture, production clock shim, source-data, or deployment change.

Verification plan

• Plan: .tmp/issue-299-verification/verification-plan.json
• Plan SHA-256: 9af528863b5fbfd40cc62435c0c1c9bba9626e11bef874b30babc1cbe5606095
• Report: .tmp/issue-299-verification/verification-report.json
• Report plan digest: 5bba6552a905bb05da69399a22d6d273ae640e1b69a63d08d612e64ccfe233e0
• Graph digest: 05ffeee99244c9becd5e98e40be08188f1c5e2a61204daece984c3beb1d1bce6
• Policy version: 5
• Selection: profile=full, reason test_infrastructure, browser smoke
• Render impact: false; screenshots: not_applicable with plan proof
• Plan dispositions:
  • rerun: selector, evidence validation, quality, Django, Playwright smoke, container
  • not_applicable: compatibility, content invariants, screenshots
  • reused: none
  • skipped: none

All required dispositions are represented in the frozen report. No evidence was reused.

Commands and results

Passing candidate/required checks:

• uv run --frozen pytest content/tests/test_events_timeline.py -q — 38 passed, 457 subtests
• make test-ci — 607 passed
• make database-portability-check — passed
• make migrations-check — passed (No changes detected)
• make django-check — passed (OpenAPI, management parity, Django system check)
• make deployment-check — passed
• make security-check — passed (baseline, vulnerability, canary, artifact scan)
• make verification-container VERIFY_CONTAINER_OUTPUT=.tmp/issue-299-verification/evidence/container-check.json — passed, 13 assertions
• uv run --frozen ruff check content/tests/test_events_timeline.py — passed
• uv run --frozen ruff format --check content/tests/test_events_timeline.py — passed

The frozen runner was executed with:

make verification-run \
  VERIFY_PLAN=.tmp/issue-299-verification/verification-plan.json \
  VERIFY_EVIDENCE_DIR=.tmp/issue-299-verification/evidence \
  VERIFY_REPORT=.tmp/issue-299-verification/verification-report.json \
  VERIFY_ISSUE=299 VERIFY_WORKTREE=.tmp/issue-299-engineer

The report validator passed and recorded verdict=failure only because the shared base remains
red in independent components. The evidence directory is
.tmp/issue-299-verification/evidence/; its invalid_evidence list is empty.

Known independent baseline failures

• Quality stops on the existing frozen projection digest mismatches in
  content/public_projection/{podcasts,wiki,wiki_search}.json (Reconcile non-reproducible public projection after face8e4 #253). The full quality command
  also exposes pre-existing format drift and four mypy errors in
  content/event_speaker_bio_normalization.py; neither path is changed here.
• make test-django-full collected 2,554 tests and reproduces the known collection-hub
  pagination assertion plus the secondary cannot pickle 'traceback' object multiprocessing
  error (Restore face8e4 formatting, typing, and focused content gates #261).
• make test-playwright-smoke selected 30 tests: 23 passed; 7 failed and 4 teardown errors.
  Failures are the existing projection/podcast invalid-resource and current-date event/hub
  baseline surfaces (Reconcile non-reproducible public projection after face8e4 #253 and the tracked browser recovery lanes), not this test-only diff.

The engineer handoff is frozen and uncommitted for independent tester review. No commit, push,
issue closure, or deployment was performed.

[alexeygrigorev]

Independent QA Review — #299

Verdict: FAIL / hold. The #299 candidate-specific contract is green, but the
graph-required full verification remains red on independently owned base failures.
This report makes no claim that those failures are caused by #299. No product source,
commit, push, label, PM acceptance, merge, or deployment action was performed.

Frozen candidate and scope

• Worktree: /home/alexey/git/dtc-website/.tmp/issue-299-engineer
• Base and observed HEAD: face8e4808d65afbf0374d1ced7a88079950d663
• Candidate status: uncommitted; exactly one tracked modification,
  content/tests/test_events_timeline.py; git diff --check passed.
• The only other status entry is the engineer handoff artifact
  verification-engineer-report.md (untracked); it is not candidate source.
• git diff --name-status: M content/tests/test_events_timeline.py
• Production paths checked (content/public_data.py, content/public_views.py,
  event templates/model/projection): unchanged versus the base.

Plan and provenance

Engineer plan:

• .tmp/issue-299-verification/verification-plan.json
• file SHA-256: 9af528863b5fbfd40cc62435c0c1c9bba9626e11bef874b30babc1cbe5606095
• source-tree manifest: 3359 entries,
  4ca5a91883ccd480269ed536c14b1d01d7d5a84cb4a9e62f69d2d4d703f5cb59
• reason/profile: test_infrastructure / full

Independent tester plan:

• .tmp/issue-299-tester-verification/verification-plan.json
• file SHA-256: aed9ef0ac51a30d2e39065b5588018d8f840ff4905052fe68200dede5cadb388
• graph SHA-256: 05ffeee99244c9becd5e98e40be08188f1c5e2a61204daece984c3beb1d1bce6
• source-tree manifest: 3360 entries,
  e46044ae0a3380884f207c692558b47c6c5068cfd3d0ce3185e488544c38dae6
• reason/profile: unknown_impact / full

The plans match on base/head, graph, environment, risk flags, render decision, and
every component disposition. The only drift is the engineer handoff artifact being
present during the tester recomputation: unknown_paths changes from [] to
["verification-engineer-report.md"], and the selector changes the reason from
test_infrastructure to unknown_impact. Recomputing the worktree manifest while
excluding that artifact gives 3359 entries and exactly the engineer manifest digest
4ca5a918...; this is explained handoff-artifact drift, not a product-source change.

Both plans select full, with fresh reruns for selector, evidence validation, quality,
Django, Playwright smoke, and container. No evidence was reused or skipped. Both record
render.impact=false, no required captures, and screenshots not_applicable because
the candidate changes only internal tests and no route/template/static input.

Tester machine report:

• .tmp/issue-299-tester-verification/verification-report.json
• file SHA-256: ba7bfe269b794679938448ca3f6740cdff43ebe45fef0891919be9d13ae23bda
• semantic plan digest: e965039d765df37d8faa32e22781909ae4f546e7b735423ecacf4acf628c087c
• phase: tester; verdict: failure; invalid_evidence=[]
• make verification-report-check returned exit 0 for the completed evidence set.

Four verification buckets

Rerun

• Selector: PASS; fresh tester plan produced.
• Evidence validation: PASS — make test-ci, 607 passed.
• Focused issue suite: PASS —
  uv run --frozen pytest content/tests/test_events_timeline.py -q,
  38 passed, 457 subtests passed.
• Controlled before/after scenarios: PASS — the nine affected tests (including the
  non-empty 2026-08-12 Europe/Berlin split) passed; the explicit
  2026-09-01 Europe/Berlin empty-upcoming route and archive scenario passed; a direct
  rerun of both route scenarios was 2 passed.
• Candidate lint/hygiene: PASS — changed-file Ruff check, Ruff format check, and
  git diff --check.
• Repository gates independently rerun and PASS: database portability, migrations
  (No changes detected), Django checks/OpenAPI/management parity, deployment check,
  and security check/artifact scan.
• Container: PASS — 13 assertions, 0 failures; artifact
  .tmp/issue-299-tester-verification/evidence/container-check.json, SHA-256
  b3954adbc0731524e6d993d4c23d25d208c97a1d256118e79be796b7bb1f8647.
• Quality: FAIL on the known Reconcile non-reproducible public projection after face8e4 #253 frozen projection digest mismatches in
  content/public_projection/{podcasts,wiki,wiki_search}.json; no Make event timeline tests deterministic when the upcoming catalogue is empty #299 path is named.
• Full Django: FAIL on the known Restore face8e4 formatting, typing, and focused content gates #261 content/tests/test_collection_hub.py pagination
  assertion, followed by the known traceback-pickling propagation; 2554 tests
  collected; no Make event timeline tests deterministic when the upcoming catalogue is empty #299 test failure was reported.
• Playwright smoke: FAIL on existing base surfaces; 250 collected, 30 selected, 23
  passed, 7 failed, 4 teardown errors, 220 deselected (flake policy: attempted 30,
  passed 22, failed 8, rerun 0, quarantined 0). Failures are the known projection,
  podcast-resource, current-date event/hub, and related recovery lanes; no Make event timeline tests deterministic when the upcoming catalogue is empty #299 route
  or browser fixture changed.

Reused

None.

Skipped

None in the computed tester plan. Every required component was executed and has a
result envelope.

Not applicable

• Compatibility and content-invariants components: outside the impact closure.
• Screenshots: render impact is explicitly false, with no route/state captures selected.

Disposition

The deterministic event-timeline correction itself satisfies the issue-specific focused
and route criteria. The tester verdict remains FAIL / hold because PROCESS requires
all graph-required components to terminate green; the independent #253, #261, and known
browser baseline failures remain unresolved. Return to the existing release recovery
sequence. Do not PM-accept, commit, merge, push, or deploy this candidate until the shared
base is green and the tester reruns the exact frozen candidate on that base.

[alexeygrigorev]

PM disposition — HOLD / not accepted

The #299 candidate-specific contract is green, but the PM gate is held under _docs/PROCESS.md; this is not an acceptance.

Verified evidence:

• The frozen candidate has exactly one tracked change: content/tests/test_events_timeline.py; git diff --check passes.
• Independent rerun of uv run --frozen pytest content/tests/test_events_timeline.py -q: 38 passed, 457 subtests passed.
• Controlled non-empty and empty-upcoming route scenarios: 2 passed; the empty state returns 200 with No upcoming events. and the past archive remains renderable.
• The tester’s required control-plane gates are recorded: make test-ci (607 passed), portability, migrations, Django checks, deployment, security, and container (13 assertions) passed. No evidence was reused or silently skipped; screenshots are correctly not_applicable because render impact is false.
• The tester’s graph-required full verification remains red only on independent baseline surfaces: Reconcile non-reproducible public projection after face8e4 #253 frozen projection-quality digests, Restore face8e4 formatting, typing, and focused content gates #261 full-Django collection/pickle reporting, and existing base Playwright failures. Register the event-slug data migration in the isolation inventory #295 remains a separate release gate.

No candidate defect remains. Production event grouping, projection data, routes, templates, styles, copy, and lifecycle behavior are unchanged.

Acceptance, commit, merge, push, and deployment must wait for the shared release baseline to become green and for the final evidence envelope to be cleaned up and recomputed. In particular, the untracked verification-engineer-report.md in the candidate worktree caused manifest/selector drift during tester recomputation; that drift must be resolved and the final handoff evidence revalidated on the exact frozen candidate.

Returning #299 to the existing release-recovery sequence; no label, closure, or source change performed.

[alexeygrigorev]

Independent tester final observation — included PM-held slice

Disposition: FAIL / not accepted. This issue's PM-held slice was included in the one frozen integrated candidate only to exercise the ordered release rehearsal. This shared integrated observation does not claim isolated issue acceptance or authorize a commit.

• Candidate: /home/alexey/git/dtc-website/.tmp/integrate-podcast-release-candidate
• Base/HEAD: face8e4808d65afbf0374d1ced7a88079950d663
• Final staged diff SHA-256: a52d6e03b291a032b46b8ae74946b9005d95f9400663566172aa0aed34ba6f48
• Included PM-held slices: Reconcile non-reproducible public projection after face8e4 #253, Restore face8e4 formatting, typing, and focused content gates #261, Register the event-slug data migration in the isolation inventory #295, Make event timeline tests deterministic when the upcoming catalogue is empty #299, Keep Spotify creator podcast embeds offline in Playwright #279, Fix homepage testimonial overflow and restore the release gate #232, and Fix undersized Random topic target in the homepage graph toolbar #236.
• Fresh full tester report SHA-256: e68bdcc670656fffcbfdb12964689e3d971285ffcf03f1ee2631978393cc8ff6; report verdict: failure.

The shared envelope recorded: quality FAIL at content/tests/test_podcast_catalog.py:201 (mypy missing annotation for unsafe_urls), full Django FAIL at the known content/tests/test_public_pagination.py:423 pagination expectation followed by Django's existing traceback-pickling error, and full Playwright FAIL with 17 failed, 294 passed, 3 deselected, 3 errors. Evidence validation (607 passed), content invariants (3,215 records / 8 files), container (13 assertions), and generic screenshots (8/8) passed. The candidate-specific Spotify child-frame tests passed; no isolated acceptance is inferred for any included slice.

No commit, push, merge, deployment, label, or PM acceptance was performed. The owning issue must retain its own lifecycle decision and, if required, obtain a new exact candidate and fresh green evidence.