docs: correct false upstream-filing commitments to honest declining status

[Brian Krabach (bkrabach)]

What

specs/EXTENSIONS.md §25, §29, and §33 each carried upstream action: deferred, reason: ..., review-by: 2026-09-05 — a commitment to file spec-correction
PRs against strongdm/attractor. That commitment was written into a shipped
public ledger even though the principal had already explicitly declined this
action weeks earlier ("skip #2"). This PR corrects it.

Why the previous value was false

Re-verified independently rather than taking the prior claim on faith:

• strongdm/attractor has had no commits since 2026-03-17 (~5 months
  quiet as of this PR).
• The repo has issues disabled and no CONTRIBUTING guide.
• Two open community spec-correction PRs there have sat unmerged for 4+
  months: #9 "Correct Attractor spec inconsistencies"
  (opened 2026-03-26) and #10 "8 spec improvements from building a
  PHP/Laravel implementation"
  (opened 2026-03-29).
• Our own maintainer's earlier attempt there, #6,
  was closed, not merged.
• The one PR of this shape that did merge, #8
  (2026-03-17), is the same day the repo went quiet.

Filing another spec-correction PR there would not land. declining is the
honest value; deferred, ..., review-by: <date> was not.

What changed

1. §25, §29, §33: upstream action: changed from deferred, reason: ..., review-by: 2026-09-05 to declining, reason: <one-line, evidence-based>,
   stating plainly that the divergence stays tracked in this ledger instead —
   which is what actually serves consumers of this engine, filing or no
   filing.
2. Entry Format section: the format block only documented two legal
   upstream action: values (a real link, or deferred-with-date). Extended
   it to explicitly permit declining, reason: <one-line reason> as a third
   legal form — otherwise the three corrected entries above would violate
   the very format they sit under. The existing "no non-date review-by
   values" rule is unchanged.
3. Sweep: checked SPEC_CONFORMANCE.md, README.md, PRINCIPLES.md,
   docs/, and the rest of specs/EXTENSIONS.md for any other undated
   promise of an upstream spec-correction PR. Found none carrying the same
   false commitment. PRINCIPLES.md's two walk-upstream notes (§25, §33) and
   feedback.py's docstring (§29) describe the recommended upstream change
   and point readers to the ledger entry for current disposition, but make
   no dated or otherwise falsifiable promise of their own — left unchanged.
4. Guard test: extended
   modules/loop-pipeline/tests/test_extensions_ledger_integrity.py with a
   structural check that every upstream action: value in the live ledger
   opens with one of the legal forms (link / deferred-with-date / declining
   / not applicable), and that every deferred value carries a real
   review-by: YYYY-MM-DD date. Proven RED against a synthetic vague-promise
   value and GREEN against one example of each legal form plus the live
   file. This cannot (and does not try to) judge whether a given reason is
   true — only that the value is shaped like a legal form at all.

Gates

• test_extensions_ledger_integrity.py: 8/8 passed (4 pre-existing + 4 new).
• test_engine_semantics_doc_guard.py + test_topological_lint.py +
  test_command_content_lint.py + test_examples_lint_clean.py: 135/135
  passed.
• Full loop-pipeline suite: 1794 passed, 1 skipped (pre-existing skip,
  unrelated).
• git diff --check: clean.
• Ledger section-heading contiguity (§1–§35, no gaps/duplicates): verified
  by the existing guard test, still green after this change (no headings
  touched, only body/value text).

Not in scope

This PR does not re-litigate whether §25/§29/§33 are the right
engineering decisions — only that their upstream action: field states the
true plan rather than a plan already known to be false.

🤖 Generated with Amplifier

Co-Authored-By: Amplifier 240397093+microsoft-amplifier@users.noreply.github.com