plat 212

← Pulse platform issue index

PLAT-212 — HDFC's recurring workflow-upgrade preflight abort is already fixed by PLAT-096's stamp-fencing work

Coordination	Value
Assigned agent	unassigned
Ticket state	resolved — correction/closure record only, no code change; already fixed the day after the last recurrence, never re-verified
Last synchronized	2026-08-29

• Priority: P3 — no live defect. A critical-severity finding that correctly described a real, repeated, workflow-destroying bug when filed, but never got re-checked against the fix that landed for it.
• Owner: N/A — no code change made or needed by this ticket.
• Related: PLAT-096 (the fix), PLAT-098 (manual recovery visibility, same subsystem). harness:workflow-upgrade-preflight:contradictory-version-and-schema-target (Workflow/HDFC-Personal-Accounts, critical, first seen 2026-08-04, RECURRED as of 2026-08-11) — the finding this closes.

The finding

Four separate occurrences (2026-08-02, 08-03, 08-04, and a RECURRED observation on 2026-08-11) where the workflow-upgrade preflight aborted an entire scheduled run — on 2026-08-11, all 8 groups, zero steps executed — with "did not stamp required version '1.0.25' (found '1.0.23', failure 1/3 consecutive)". Recovery required stamping workflow.json to the demanded version out of band each time. The finding's own 2026-08-05 "symptom cleared, defect unexercised" verdict was explicitly reopened by this 08-11 recurrence, satisfying its own stated reopen condition.

Already fixed — confirmed via git history and live evidence, not assumed

PLAT-096's stamp-fencing fix (pkg/contractupgrade, commit 79ba90168, 2026-08-12) is exactly the mechanism this recurrence matches: an unfenced set_workflow_contract_version write reachable from any turn in a scheduled session, causing a scheduled preflight to see a manifest version it didn't expect. PLAT-096's own text independently confirms HDFC by name: "The same rung completed on three other workflows the night before — ICICI-BANK-PARSING 22:20→22:24, HDFC-Personal-Accounts 22:39→22:42, ICICI-BANK-PARSING-v2 23:13→23:16 — all retired improve.html and reached 1.0.25."

Confirmed independently against this workflow's own live evidence rather than trusting that citation alone:

• workspace-docs/Workflow/HDFC-Personal-Accounts/schedule-runs.json, sorted by started_at: the 2026-08-11T17:09 abort is the last contract-upgrade-preflight error in the file. The very next run (2026-08-12T02:23:52, after the fix landed the same morning) succeeded. No further preflight-abort errors appear through the file's latest entry (2026-08-18, an unrelated interrupted: server restarted).
• workspace-docs/Workflow/HDFC-Personal-Accounts/workflow.json's current version is 1.0.26 — one rung past the 1.0.25 that was blocked, confirming further contract bumps have completed successfully since.

Why this stayed open

Same pattern as PLAT-191/195/201/202 this session: a harness finding correctly described a real defect, a fix landed for it, and nothing re-verified the finding against the fixed code before this pass. The fifth instance of this exact pattern found in one review sweep — now clearly a recurring gap in this platform's own finding lifecycle (nothing automatically re-checks an open harness finding against current code), not a coincidence.

Explicitly not done

• No code change — PLAT-096, already shipped and independently re-verified here, covers this.
• Did not re-derive PLAT-096's own extensively-documented root cause analysis (stamp fencing, single delivery path, unattended-turn handling, Pulse-skip-on-block) — cited and confirmed its outcome against this specific workflow's evidence rather than repeating the investigation.
• Did not investigate PLAT-096's own still-open follow-ups (minimal-context upgrade turns, inverting the stamp, confida-login's self-reinforcing refusal) — out of scope for this closure record, which is specifically about HDFC's recurrence.

Verification

• Direct read of schedule-runs.json and workflow.json inside Workflow/HDFC-Personal-Accounts, sorted and cross-checked against the finding's own cited run IDs and timestamps.
• git log confirms the fix commit (79ba90168, 2026-08-12) postdates every cited recurrence and predates the next successful run.
• No code changed by this ticket — nothing to build or test.