workflow: rejecting a draft and replaying it with feedback works again (#870)

[Masterplanner25]

Closes #870.

The finding

#482 (5.5.0) refused resume_workflow(id, checkpoint, payload) on a waiting run,
on the stated grounds that "the rollback re-armed the wait and the payload was
silently discarded".

The payload was not discarded. Measured on 4.0.8 in a clean venv, where the
spelling worked, it reached workflow_resume_payload() in the replayed step:

[4.0.8]  checkpoint+payload -> status=waiting  draft={'fb': '{"feedback": "redo"}'}
[5.14.0] checkpoint+payload -> ok=False        "drop the checkpoint argument"

What it does not do is satisfy the wait. That is a different thing, and it is
exactly what a reviewer rejecting a draft wants: replay from before the draft
with the feedback, then park for approval again. So the refusal removed a working
pattern on a false premise — and the premise was in the error message and the
CHANGELOG, so a reader hit it twice.

The capability was never removed, only its spelling

The refusal's condition includes event_type is None. event_type is a
verification argument — "refuse unless the run is waiting on this event" —
and it suppresses the refusal only incidentally. Confirmed on main before
changing anything:

2. .nd builtin (refused)   : False | waiting_run_checkpoint_resume
3. runner + event_type     : waiting {'draft': {'fb': '{"feedback": "redo"}'}}
   -> replayed step saw the payload? yes
   -> did publish fire?              no

So the supported path was refused while the capability sat behind an accident.
BothPathsAgree now pins the two spellings to one answer so the supported one
cannot drift back onto it.

What changed, and what deliberately did not

checkpoint + payload is allowed. One condition: and resume_payload is None
added to the refusal.

resume_workflow(id, "checkpoint") with no payload is still refused, and
that is the part of #482 that earned its place — a rollback re-enters the waiting
step, which waits again, so nothing changes behind a healthy-looking result. Its
message now names both recipes instead of only the advancing one.

The neighbouring mistakes needed nothing: a checkpoint the run has not reached,
and a typo'd one, already fail with Checkpoint not found. Verified all five
shapes rather than assuming:

Call	Now
pre-wait cp + payload	replays, parks again, publish does not fire
pre-wait cp, no payload	refused, names both recipes
post-wait cp + payload	Checkpoint not found
typo'd cp + payload	Checkpoint not found
payload alone	advances, publish runs

Why this beats the workaround

The two-phase dance the reporting project found — satisfy the wait, then roll
back — takes two calls and runs every post-wait step during phase one, so a
side-effecting publish fires on a rejection unless each step gates on the
payload. The guide documents the workaround for 5.5.0–5.14.0 and says plainly
why it is a workaround.

The 5.5.0 record is annotated, not rewritten

Editing a released entry would leave the repo disagreeing with the published
notes. The entry stays as shipped with a note naming the wrong sentence and what
it cost — it is the sentence a future reader would use to decide whether the
refusal can be relaxed, which is the decision this PR just made.

Tests

tests/test_reject_and_revise.py (11) plus the rewritten #482 test. Three
per-decision neuters, restoring from saved byte copies:

BASELINE (fixed tree)                        14 passed
the 5.5.0 refusal, restored                  RED   7 failed
refusal message drops the reject recipe      RED   1 failed
refusal stops firing entirely                RED   3 failed
RESTORED                                     14 passed

test_checkpoint_with_payload_is_refused_naming_the_discard asserted the false
premise, so it is rewritten rather than kept — the same shape as #102's scan test
in #869, which pinned that defect too. Two of #482's three tests survive
unchanged, which is the check that this narrowed the refusal rather than
gutting it.

Checks

• test_reject_and_revise, test_waiting_run_checkpoint_resume,
  test_nodus_workflow_framework, test_parked_run_survives_age_bound,
  test_resume_rebuild, test_runner_results: 72 passed
• ruff clean · mypy src/nodus/ clean — though it does not cover
  src/nodus_lang_workflow/, where this change lives, so CI is the only type
  check for it
• nodus_gate --all green, all ten phases, before the push — BUG-869: the local store's age filter hides waiting runs from listing, migration, rehydration and the #174 warning #869's lesson,
  where I ran two phases, reasoned the rest were irrelevant, and CI failed on
  --flips

🤖 Generated with Claude Code