docs(adr): decline ADR 0089 Phase D (helper descent) by owner ruling

[wshallwshall]

Owner ruled 2026-07-30: helper descent is too risky. ADR 0076 Amendment B goes PROPOSED → ⛔ DECLINED, and ADR 0089 Phase D is not to be built.

Recorded in four places, deliberately

A decline that lives only where it was argued is a decline nobody finds. The load-bearing one is ADR 0089 itself — a session planning the next lens phase reads 0089's phase list, not an amendment to a different ADR. And 0089's own text sells Phase D as "the largest structural lever" against "265 delegating call-sites".

File	Change
adr/0089 §Phase D	⛔ ruling banner above the original paragraph, which is retained unchanged
adr/0076 Amendment B	status block → DECLINED, with the case and the reopening bar
adr/0076 status line	Amendment B marked declined
adr/README.md	index row updated (edited in place — the ledger gate rejects a second [0076] row)

Also corrects, at 0089 §Phase D: "265 delegating call-sites" is a statement-scan count, not a yield estimate. Descent makes nothing editable by itself — it swaps opaque delegating rows for helper-body rows whose editability is whatever Phases A–C already achieve inside helpers.

Decline-by-design, not deferral

The specification stays. Not as a plan — so the reasoning is auditable, and so the next reader who concludes "helper delegation is 41.8%, that's the biggest remaining win" finds the work already done rather than re-deriving it.

Reopening requires a new amendment, and must first:

1. Answer aliasing — two call sites of one helper produce identical child row spans; expect_src matches both, so an edit to one silently rewrites the other. No ADR addresses this.
2. Produce a real per-helper editable-share measurement, not the 218/522 heuristic superset, which never checks whether a callee resolves to a same-module def.

The yield may be negative: msg["X"] = v is not recognized (only msg.set(...) is), and the shipped samples/config/_demo_oru_transforms.py writes exclusively in that form — descending it returns six opaque rows and zero editable ones.

Promoted from precondition to recommendation

Teaching Phase A the ast.Assign-to-ast.Subscript form (msg["X"] = v) converts writes into editable rows without touching the row shape at all, and is on no phase list. It is explicitly not declined.

Incidental fix — a stale cross-reference this session introduced

BACKLOG #239's banner cited #240 (comment-only rows) from before that item was renumbered to #248. #240 is now a permanent hole, so the citation resolved to nothing — the exact failure mode the #88 erratum describes, committed by the session that wrote the erratum. Corrected, and the same banner's "~70% of the opaque mass" argument is amended: with Phase D declined it rests on comment-only rows alone, so re-measure before assuming the number moves.

Verification

python scripts/docs/backlog_status_check.py   → OK — 242 items, each declaring exactly one status
python scripts/security/scan_forbidden.py     → clean
python scripts/hooks/ledger_check.py          → no blocks
grep "#240 (comment"                          → no matches remaining

Docs-only, +44/−8 across four files. No code, no behaviour change, no number allocated.

🤖 Generated with Claude Code