The class-level D3 guard #1275 asked for is undelivered: it needs a new test/** file, and both its legs are now specified #1897

Blocked-by: #1874

Filed by the repo:hotcrm seat at the acceptance of PR #1895 (#1275 + #1804), rather than left as an unexplained gap. ⛔ The fix is complete and landed; this is the guard that was asked for alongside it.

What is undelivered

#1275 and #1265 both specify a class-level test — one that fails when a hook writes a payload key conditioned on the row in front of it, so the D3 class cannot silently return. PR #1895 delivered the fix for all 10 roster members and did not write that test: it would be a new file under test/, which is epic #1579's fence this lane does not cross on its own word.

⇒ the class is currently protected by ten one-line guards and no gate. Nothing fails if the eleventh hook is written without one.

⭐ Both legs are now specified, which they were not when the cards were filed

The acceptance measurement established the shape the test must have:

1. The update SUCCEEDS — ⛔ not merely "it does not widen". Run the install-surface probe for the seven cards whose objectstack upstream closed before the pinned 17.4.0 — #806 · #734 · #1185 · #1265 · #1275 · #1804 · #1301 #1888's probe showed the engine converts key-presence divergence into a refusal, so a test asserting only "no widening" passes on a batch that was refused outright and wrote nothing.
2. No payload key is set on a row whose own previous said it should not be — the silent limb, which is the one that lasts (see the correction on claimSeedOwnership leaves 7 crm_forecast seed rows unowned: bulk update rejected because beforeUpdate writes period_end/period_label #1804: the refusing tier is loud and self-healing; the widening tier is permanent).

⚠️ And the obvious instrument is the wrong one. test/helpers/hook-harness.ts hands each call a fresh input — the per-row payload copy that D3 says the engine does not build (#1016). A harness-based test would therefore pass against the defect. PR #1895's acceptance used a live before/after on a running app for exactly this reason, which is a stronger instrument but not a gate.

Why pm:blocked and not pm:queue

The deliverable is a new file under test/. That is epic #1579's declared territory, and this lane's rule is that the fence opens only on a maintainer or director sentence. ⭐ Same question as #1874. The live reverse index — recomputed from ^Blocked-by: #1874 across all open cards, ⛔ not counted by hand — reads 11 carriers with this card included: #1598 · #1727 · #1763 · #1773 · #1789 · #1802 · #1873 · #1875 · #1883 · #1896 · #1897. (⚠️ The seat wrote "eleven" on #1896 and "twelfth" here; both were off by one, counted from a stale figure instead of derived. The derivation is the number.)

⚠️ Whoever takes it should also decide whether the guard belongs to this repo at all: the rule it would enforce is a platform contract (ADR-0058 Addendum II D3), so a lint rule upstream would cover every app rather than this one. That is a routing question, not a blocker, and it is not decided here.

Scope when it unblocks

• The 10-member roster and its A/B tiering are recorded in PR fix(objects): stand row-conditioned payload writes down on a predicate update #1895's report (5629295… on The other 8 hooks with row-conditioned payload writes — same D3 batch-widening as #1265, same platform blocker #1275) — ⛔ re-derive from the tree rather than inheriting that table, as The other 8 hooks with row-conditioned payload writes — same D3 batch-widening as #1265, same platform blocker #1275's own 2026-08-26 comment required and as PR fix(objects): stand row-conditioned payload writes down on a predicate update #1895 did (it found 2 members the card's table did not name).
• knowledge_article_publish_timestamps belongs to Bulk path: a beforeUpdate rewrite conditioned on previous widens to every matched row — published_at is measurably overwritten batch-wide (ADR-0058 Addendum II D3) #1265 and was deliberately excluded from fix(objects): stand row-conditioned payload writes down on a predicate update #1895; it is hard-serial behind that branch and is still queued.

Refs #1275 · #1804 · #1265 (still open, same class) · PR #1895 (the fix) · #1888 (the probe that specified both legs) · #1579 (the epic that holds test/**) · #1016 (the harness's fresh-input shape) · #1874 (the fence).

Generated by Claude Code