chore(record): the ABCD sensemaking method, its principle, and itd-86/87

[REPPL]

Why

abcd's principles have been re-deriving a method that was never written down in
this record: the cold reading (a blind specification pass), the warm
ledger, and the human disposition. The evidence that this was happening is
in the record itself — verifier-selects-gates-decide turns out to be an exact
independent restatement of the method's disposition rule, and
evaluator-outside-the-loop is a weaker cut of its blindness rule (it isolates
the evaluator from what it may edit; the method isolates it from what it may
know).

Meanwhile itd-27/itd-42 (grill), itd-55 (foundations auditor) and now itd-86/87
had been accumulating as siblings with no stated parent. That absence has a
cost: the cold reading's prohibitions — propose no fixes, rank nothing — read as
an unhelpfully passive reviewer unless you already know a disposition step exists
downstream to do that work.

This PR records the method, and derives from it exactly what the record was
actually missing.

What

• The method note (research/notes/2026-07-13-abcd-sensemaking-method.md) —
  the parent, with a table mapping each element of the method onto where abcd
  already states it, and what it does not.
• One principle — recurrence-is-signal. A finding that reappears after it
  was dispositioned is evidence about the disposition, not a duplicate of the
  finding. A ledger that dedupes a recurrence converts "we were wrong to close
  this" into "we already know about this".
• itd-86 — cold-reading surface (the blind detection pass).
• itd-87 — recurrence escalation in abcd capture.
• A stale-pointer fix in the existing skills-evaluation note, which cited a
  skills/abcd-intent-grill/ directory that the Go rebuild left unbuilt. The
  grill is planned work (itd-27, adr-7, itd-42) and that is where its harvest
  target actually is.

Two deliberate restraints

Only one principle was minted, not one per method element. The cold/warm
split is already covered by the two principles named above, and
one-canonical-primitive forbids a third near-copy. A read-it-cold principle
was considered and rejected; recurrence was the only element of the method
with no counterpart anywhere in the record. Recorded in DECISIONS.md.

itd-86 and itd-87 are coupled, not merely related. A blind cold reading
re-raises previously surfaced tensions by design; pointed at a ledger that
dedupes them, it is a detector fighting its own store. itd-87 is the precondition
that makes itd-86's re-raising useful rather than noisy — hence one change.

Held, loudly

The ACKNOWLEDGEMENTS.md entry crediting the cold reading to abcd's co-author is
deferred, not forgotten: it awaits confirmation of how they wish to be
credited, and must land before itd-86 ships. Stated in the method note and in
DECISIONS.md rather than left silent, per loud-staging. Do not guess the
credit line.

Verification

Record-only; no Go source touched.

• abcd audit → exit 0, 0 errors, 0 warnings
• abcd docs lint → 0 findings, 0 blockers
• gofmt -l . → empty

Both gates were re-run after rebasing onto main at #66, since that PR changed
the audit code.

Note for the maintainer

itd-86/87 were allocated against main. Given fix/duplicate-intent-id-lint is
recent history, re-check the IDs if another intent was minted concurrently.

Adoption is unchanged and remains yours: these are drafts, and abcd intent plan
is not run here.

Assisted-by: Claude:claude-opus-4-8