evals/results/v3/state-trial-notes.md

# Evaluation notes

This is a fixed report of the evaluation, not a second live claims dashboard.
The append-only ledger is authoritative. `HANDOFF.md`, the JSON reports, and
the before/after handoff snapshots were generated by the supplied helper.

## Completed work and observed state transition

Events E000001 and E000002 registered the exact integer claims C_BASE and
C_GOAL. C_GOAL declares C_BASE as its dependency. Complete proof artifacts
were recorded as E000003 and E000004. Both claims then had evidence status
`proved`, review status `not-independently-audited`, and no blockers or stale
issues. The initial `check` returned exit code 0. This certifies ledger
integrity and evidence freshness only, not mathematical correctness.

Before changing the load-bearing claim, `impact C_BASE` reported C_GOAL as
its dependent. E000005 revised C_BASE to the rational-domain assertion with
the exact reason `Explore a stronger domain`. The convention for evenness
remained membership in 2Z. Immediately after that revision, both claims had
status `stale`; E000003 and E000004 reported `claim or transitive dependency
revision changed`. C_GOAL was blocked by C_BASE. The old artifact bytes and
numbered events were retained unchanged.

The rational extension is false: n=1/2 is rational and gives n(n+1)=3/4,
which is not an even integer. A complete exact counterexample artifact was
recorded as E000006. The final generated state reports C_BASE revision 2 as
`refuted` and C_GOAL revision 1 as `stale`, blocked by C_BASE. The two old
proof records remain stale. The integer assertion C_GOAL itself has not
been disproved; its stored proof contract uses an obsolete dependency
revision. No independent review was performed or claimed.

## What must be checked next

E000007 records the candidate route R_GOAL_INTEGER_PARITY. It is ready,
with no prerequisite claims and declared gain 4, cost 1, and heuristic score
4.0. The route has not been executed or marked successful.

1. Register the original integer parity statement under a separate claim ID,
   C_PARITY_INT, and give that exact contract a durable proof.
2. Verify that the only parity invocation in the C_GOAL proof uses an integer
   n. Recheck the exhaustive remainder cases modulo 3 and the last integer
   identity establishing divisibility by 6.
3. Revise C_GOAL's dependency contract, with a reason, to use C_PARITY_INT.
   Keep its integer statement. Write a new proof artifact that cites the
   new dependency and record new evidence only after revalidation. Preserve
   the old proof artifact and stale events as historical evidence.
4. Record the route outcome separately, including the actual outcome and
   next question, and regenerate the handoff. A route success is not proof
   evidence. An independent review would require a genuinely fresh review
   of the exact new evidence event and a durable report.

The proposed route does not require proving the refuted rational statement.
Its missing integer lemma is work inside the mechanism, not a result required
before execution, so an empty prerequisites list is intentional.

## Actual execution and usability observations

- The mandated first read-only `check` and `handoff --goal C_GOAL` both
  returned exit code 2 with `project root does not exist`. Creating the
  explicitly authorized project directory resolved that bootstrap obstacle.
- With the directory present but before initialization, both commands
  returned exit code 2 with `ledger not initialized; use init explicitly`.
  Explicit `init` succeeded. A subsequent empty-ledger `check` succeeded,
  but `handoff --goal C_GOAL` returned `unknown claim: C_GOAL` until the
  claim was registered. Thus the read-before-change instructions need a
  natural bootstrap exception for a new, not-yet-registered goal.
- All seven documented proposal schemas were accepted on their first
  submission. No helper crash, unexpected rejection, dependency installation,
  or artifact-path problem occurred.
- After the intentional revision, `check` returned exit code 1 for the
  expected stale snapshots. The final check still returns 1 for those same
  two historical proof records. It reports no hash-chain or artifact-byte
  corruption. `next` and `handoff` both succeed with exit code 0.
- Only files inside this evaluation project were created or modified. The
  supplied Mathbox helper and skill files were read or executed without
  modification. No external services, personal context, delegation, or Git
  operations were used.

See `reports/command_results.json` for the observed exit codes and the
generated state snapshots for exact claim and evidence records.