PROBE-WARRANTED-VIEW-TRACE-1 — an edit must carry the evidence that earned it

[AdaWorldAPI]

The first causal particle, not the first learner

#1002 proved one edit can be typed, replayed and inverted. This PR adds the law that must exist before any learner, because without it a BPE-style compressor would learn a recurrent bad habit exactly as efficiently as a good one.

Three vicious invariants, 9/9 gates green, over the same sealed field (2016 beliefs, 20 rows across rungs {1,2,3,4,6}), trajectory A(10) —e1→ 4 —e2→ 10 —e3→ 4:

  RECONSTRUCTION  replay(A, [e1..en]) == final, AND at every prefix
  GROUNDING       each ei NAMES evidence available BEFORE ei
  ANTI-HINDSIGHT  no future outcome can manufacture a warrant

The defect this PR is actually about

My first draft's Warrant carried visible_before / rungs_before — a description of the situation, which every possible edit satisfies. It would have passed a full reconstruction suite while measuring nothing about justification. That is precisely the anti-pattern CLAUDE.md names: a guard that fires on everything carries exactly as much information as one that never fires.

The fix is a different object, not a louder assertion:

Evidence::BeliefsAtRung { rung, count }   // drawn from the sealed arena
Evidence::RowsBoundAt   { locus, count }  // drawn from the population

e3 Push(RungBand{4,6})
   warrant [GROUNDED]: support=[BeliefsAtRung{rung:4,count:412},
                                BeliefsAtRung{rung:5,count:632},
                                BeliefsAtRung{rung:6,count:496}]

Gates

T1/T2	end-to-end replay exact, and per-prefix — an end-to-end match can hide two errors that cancel; a per-prefix match cannot
T3	all three edits invert step-by-step back to A exactly (RemoveAt inverts only against the plan it acted on, returning None for a no-op rather than lying)
T4	warrant_at(step, initial, prefix, edit, arena, ctx) — no parameter through which a later edit, the final view, or an outcome could enter; all warrants reproduce byte-equal from prefix alone
T5/T6	a reordered trace and a truncated trace both miss the final view — the reordered one lands on a set of the same size (4 rows), which is why the gate compares sets; and every step must move the visible set
G1	every edit names evidence available before it
G2	the channel can REFUSE — an off-field RungBand{50,60} (above the measured R6 ceiling) and an unbound Locus::Contradiction both return UNGROUNDED while the real trace stays fully grounded
G3	every named count re-verifies against the sealed state (9 items recounted) — a warrant cannot pass by inventing plausible figures

G2 is the load-bearing one. The same [e1,e2,e3] subsequence looks identical whether real evidence or circular reasoning drove it. G2 is the difference between acquiring reasoning skill and acquiring superstition, and it belongs before the learner.

T4 is witness_fabric's shipped upto discipline transplanted — "Retrospective judgement of a past state must not be able to see what came after it, or it is hindsight wearing an audit's clothes" (witness_fabric.rs:1476-1479).

The remaining gap is one arrow

  Evaluation  →  Revision  →  warranted ViewEdit
                    ↑
              this arrow

F-REVISION-FOCUS-1 remains ABSENT. No production surface emits a ViewEdit, so every trajectory here is authored by the probe, not harvested from a running cognition. A trace OBJECT exists; a trace PRODUCER does not.

Non-goals, unchanged

Nothing is learned, compressed, promoted, or recurrence-detected. BehaviorTrace is probe-local and is not proposed as a production type. No scheduler, no CommonMask, no CE64 change, no RungElevator / StyleLane / EpistemicMode / temporal.rs anywhere in this probe.

Only once many grounded traces exist does the question become legitimate: does some [e1,e2,e3] recur and survive its outcomes — rather than merely "what do I often do?"

🤖 Generated with Claude Code

https://claude.ai/code/session_01KCGhDYoQBXs3poaR7sFuqp

---

Generated by Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_edfecda5-0f7f-4605-88f0-0da1635b4cd6)

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 46b93e3e-3715-4b4a-a5f4-2249c40ba86e

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}