challenge-plan-5

Round 5 of plan-challenge. The demotion-review round. Different shape from rounds 1-4: instead of picking a new angle and hunting new findings, round 5 deep-reads every claim the prior rounds parked as uncertain (the "demoted claims") and verifies whether each demotion was warranted.

Guard: nothing to challenge

If no plan was presented in this conversation, decline with: "No plan to challenge in this conversation. Draft one first with construct-the-plan, write the plan inline above this command, or name a plan file."

Guard: invoked without rounds 1-4

If rounds 1 through 4 haven't all run in this conversation (check by reading back prior turns for "Round N of plan-challenge" markers), round 5 is the wrong place to start: it reviews the demotions those rounds generate. Decline and point at round 1.

Guard: repeat invocation of round 5

If round 5 already ran, treat this invocation as a fresh pass over the demoted claims with new evidence. Say so explicitly.

Methodology

Read the "Demoted-claims tracking" table the prior rounds built up in the plan file. For each claim still demoted, do a focused deep-read of the actual files, docs, or sources the claim is about. Don't trust the demotion note; go look.

Three outcomes per claim:

1. Promote back to verified. The deep-read shows the claim was dismissed too quickly; the original plan position was correct. Restore it in the plan body and record the promotion in the table.
2. Confirm demotion. The claim is genuinely uncertain after the deep-read. The demotion stands; the recorded resolution stays.
3. Escalate. The deep-read shows the claim is worse than uncertain; it's a real problem. Edit the plan to address it and record the escalation in the table.

Trivial exit: if no claims were demoted in rounds 1-4, say "No demotions to review from rounds 1 through 4. Round 5 trivially complete." and skip to the self-confidence check.

No-rubber-stamp guard: confirming every demotion without opening the files it concerns is rubber-stamping. The point is the deep read. If you catch yourself doing it, say so and go do the actual investigation.

Structured count line

End the review with this line, replacing each count with a real number (they must add up):

Re-examined [count] demoted claims. Promoted [count] back. Confirmed [count]. Escalated [count].

This replaces the standard "Re-verified N, demoted M" line for round 5 specifically. Surface the per-claim outcomes BEFORE the count line, and the diff summary (same mandatory format as the other rounds) if any plan edits happened.

Self-confidence check (after the count line)

Before declaring round 5 done, ask honestly: "Am I confident the plan is sound, or am I stopping because we hit round 5?" If not yet confident, run additional rounds without asking the human (AGENTS.md § The flow guarantee).

Rounds 6 and beyond alternate: an angle-rotated round (rounds 1-4 template, NEW angle from the list, stated at the top), then, if it generated new demotions, a demotion review over ALL still-demoted claims (which can promote an older demotion a new angle proved wrong). Stop on either signal:

• Two consecutive rounds finding nothing, with genuinely different angles. The curve has tapped out.
• Reaching round 8. Auto-stop with a one-line note naming the last angle used. Past 8 rounds the angle list is two-thirds spent; the human can always invoke another round manually.

Every extra round must declare a NEW angle not used in any prior round; reaching for a used one is itself the stop signal. Keep a ledger of every finding any round surfaced, including the rejected ones, and dedupe each extra round's fresh findings against everything SEEN, not just what was confirmed or fixed; otherwise a rejected finding resurfaces and the chain never runs dry.

Closing declaration when stopping: name the angles. "I've used angles A, B, C, D, E and can't find a sixth that isn't a re-skin of one of those. Confidence in the plan is earned after five rounds of genuine rotation." A confidence claim with no angle citation is the lazy kind; don't write it.

Self-check before posting

Same checklist as round 1: line cap, no sub-categorized headers, no count-justifying listings, no bare confidence declarations, no cross-round commentary or forward-looking framing, no jargon-stacked labels. The round-5 count line above is mandatory and verbatim.

Rationale (recorded so future edits don't drift it)

Demotions are cheap to write and expensive to leave wrong: a claim parked as "uncertain" in round 1 quietly weakens the plan even when the original position was right. This round exists so every demotion gets a second, evidence-based look before the build starts, and so the chain can extend itself honestly when the bug curve hasn't tapered instead of stopping at an arbitrary round count. The seen-vs-confirmed dedupe adapts the loop-until-dry rule from Codez's graph-engineering writing (July 2026): dedupe against everything seen, or the loop pays to rediscover its own dead ends.