plat 129

← Pulse platform issue index

PLAT-129 — 13 more guidance test assertions traced to specific renames and removals from 0174b6aff/aad50dfb0/f67ccc832

Coordination	Value
Assigned agent	unassigned
Ticket state	implemented (13 of PLAT-128's 19-failure baseline); 6 remain, itemized below, not attempted here
Last synchronized	2026-08-17

• Priority: P3 — same class as PLAT-128: test/content drift, nothing product-facing was broken.
• Owner: cmd/server/guidance/render_all_test.go

Method

Same discipline as PLAT-128: for every missing phrase, git log -S"<phrase>" -- <template file> to find the commit that last changed its count, then read that commit's actual diff to confirm removal vs. rename, then confirm the current state by rendering the template (renderFromRegistry), never by grepping the raw file (a rendered doc can differ from its source file via shared includes). Every claim below was rendered, not assumed.

Findings, grouped by cause

1. pulse-archive doesn't exist (1 assertion). Deleted in 0174b6aff along with its Go implementation (pulse_archive.go, 174 lines) and its own test file. Zero live callers. Removed from TestFocusedScheduledPulseReferencesStayComplete's map.

2. goal-advisor.md's entire pedagogical/mechanics apparatus was stripped by 0174b6aff (9 test functions, ~50 assertion lines). The file went 538 lines → a fraction of that in one commit. Confirmed by rendering the current template: every one of the ~50 phrases these tests checked — the active-experiment lifecycle, the analysis-first-over-HTML contract, the strategy-first-pass apparatus (PHASE 1A/1B, data-experiment-kind, NON-NEGOTIABLE STRATEGY-FIRST PASS), the metrics-subsystem-abstention guidance, assumption-audit linking, and its shared "coherent agentic steps"/"deterministic fetcher"/"shared-context span" phrasing — is absent.

Two whole functions removed outright (TestGoalAdvisorPrioritizesStrategyOverHTMLFormatting, TestGoalAdvisorTreatsCleanAbstentionAsStrategyEvidence: 5/5 and 32/32 phrases confirmed absent). Everywhere else, goal-advisor was one entry in a multi-kind map (design-plan, plan-design, regular, message-sequence, step-config, improve-report, etc. alongside it) — only the goal-advisor entry was dropped; every sibling kind in the same map was individually confirmed still correct by rendering it, not assumed correct by association.

3. builder/improve.html dashboard markup is gone from every template on disk (3 test functions, ~25 assertion lines). grep -rc "data-pulse-section\|data-module=" cmd/server/guidance/templates returns nothing anywhere. This is the same root cause as PLAT-128's review-improve-log finding, surfacing through a structurally different set of tests (per-kind HTML-attribute checks rather than direct review-improve-log kind rendering). TestPulseRelatedGuidanceUsesFourPartSectionOwnership removed outright (8/8 kinds' phrases confirmed absent or markup-only); pulse-setup's want-block in TestPulseRunsEveryDueReviewerAndWritesAttributedResults removed (4/4 absent, including its own literal "builder/improve.html" check); the trailing data-pulse-section/data-module pair dropped from that same test's ops-review/strategy-auditor loop, keeping "Do not truncate", which is real.

4. Three genuine renames, not removals (4 test functions, updated rather than deleted):

• call_generic_agent → run_in_background for strategy-auditor and review-artifact-drift (aad50dfb0 "stabilize pulse orchestration and scheduled sessions")
• "READ-ONLY REVIEW" → "READ-ONLY STRATEGY AUDIT" for strategy-auditor's dispatch instruction, same commit. (A second, unrelated "READ-ONLY REVIEW" occurrence in TestPulseGuidanceRequiresRuntimeAuthorityAndVisibleFreshness checks the raw docs/design/pulse-post-run-monitor-spec.md file directly — confirmed that file still says it verbatim — left untouched.)
• finding_id/target_key → "no invented identifier" for ops-review, strategy-auditor, review-artifact-drift specifically (f67ccc832 "Pulse platform hardening"), while 6 other specialists still use finding_id. TestPulseSpecialistsReturnStructuredPacketsAndParentOwnsHTML restructured to a per-kind want-list instead of one shared list, so this distinction — a real, deliberate design split, not drift — is expressible going forward rather than fought against.
• improve-learnings/improve-knowledge/improve-database/improve-report/ improve-evaluation consolidated into a shared "ENGINEERING REVIEW — <lens> LENS" header, dispatched through prose ("normal Engineering/Ops background executor") rather than each carrying its own standalone "READ-ONLY <X> HEALTH REVIEW" title and call_generic_agent call. TestMaintenanceImproveGuidanceIsReadOnlyForPulseFixerHandoff updated to the current titles and mechanism phrase.

5. Two stray single-phrase removals from 0174b6aff, unrelated to the above categories: design-plan's "never discard findings" and assumption-audit.md's "Do not add an assumptions panel" (a dashboard-specific instruction, moot now that the dashboard it referred to is retired). Each was the sole remaining failure in an otherwise-passing function; both confirmed via git log -S against the same commit.

Verification

Fail-before/pass-after: git stash push -u (all changed + new files), re-run, git stash pop, re-run — 19 failures before, 6 after, across cmd/server, guidance, virtual-tools, step_based_workflow. Zero new failures anywhere. TestEveryPulsePlatformTicketIsLinkedFromTheRegister (the register-integrity check added since PLAT-128) passes against this ticket's own row.

The 6 that remain — itemized, not attempted here

These are not drift; each is either a real product-architecture question or a real scheduler bug, and guessed-in test edits would hide rather than surface them:

1. TestWorkflowScheduleTrackingWindowStartSurvivesEmptySchedulerState and
2. TestRecordWorkflowSchedulePreflightFailureFailsOpenAtThreshold — both exercise the exact mechanism PLAT-080 ("an old cron schedule with no durable fire-decision row restarted from 'now'") claims to have fixed, currently marked implemented — runtime reverify pending. The second fails with a real read/write round-trip bug: RecordWorkflowSchedulePreflightFailure's count goes backward on the second call (schedule_execution_history_test.go:191, "failure count = 1, want 2"). This may mean PLAT-080 regressed or was never fully fixed — worth its own investigation, referenced from PLAT-080 rather than duplicated here.
3. TestStandalonePulseReviewCommandsUsePersistedReviewerPipeline — the call_generic_agent → run_in_background rename (item 4 above) is not a simple word-swap here: probed, and the entire "persisted standalone-review" contract this test checks (pulse_run_id/review_run_id handling, SQLite persistence, get_pulse_state(view="review")) is absent for ops-review/strategy-auditor/review-artifact-drift alongside the old dispatch call. run_in_background's actual persistence/attribution model is unfamiliar territory not investigated here — needs someone who knows what replaced it, not a guessed test edit.
4. TestFocusedScheduledPulseReferencesStayComplete and
5. TestStrategyAuditorGuidanceRequiresLongitudinalEvidenceAndReadOnlyHandoff — both check pulse-review-fixer.md for phrases ("automatic completion notifications", "Do not finish this Review+Fix turn until their evidence has been consolidated") that git log -S across the entire repo history, every file cannot find having ever existed anywhere. Unlike every finding above, this is not "removed" — it looks like content that was planned and asserted for but never actually written. A product decision (write it, or decide the test was aspirational and retire it), not a mechanical fix.
6. TestArtifactDriftAuditsTheSchedule (schedule_drift_scope_test.go) — review-artifact-drift.md is missing "execute_step", "validate_plan", and a drift-case phrase "drives no plan step". Not yet traced to a commit; likely connected to the same 0174b6aff/later restructuring given how much else in this file changed, but not confirmed.