Bump EigenScript pin v0.34.0 -> v0.35.1

[InauguralPhysicist]

Picks up the #827 fix in v0.35.1: the temporal assignment history backing prev of x, <kw> is x at <line> and state_at was append-only, uncapped, and held a reference to every recorded value. Merely mentioning one of those forms in a loaded file — including never-executed code — armed unbounded retention; on a 4 GB box the kernel thrashed instead of OOM-killing. v0.35.0 also carries the lib/ui x-y chart widget (#819), which this repo wanted for counters-over-time with restart markers.

The counter check — both regimes reproduce BYTE-IDENTICALLY

Counters here are the result of record, so a runtime bump that moved them would be a major finding. It did not move them.

case	banked (v0.34.0)	measured on v0.35.1	verdict
3x3-odd	1,850 res / 619 conflicts	1,850 / 619	byte-identical
4x4-odd	96,733 res / 27,421 conflicts	96,733 / 27,421	byte-identical

Both UNSAT as expected. Full 4x4 line: learnts=27420 learnt_lits=212373 peak_learnts=906 max_level=19 decisions=45587 propagations=141097 restarts=13. Wall 675s on the devbox — per-host color only, not compared with anything.

No re-bank was needed and none was performed.

Also worth noting for #827 specifically: interpreter RSS stayed flat at 3,756 kB across the entire 675-second 4x4 solve — the retention shape the fix removes would have been visible over a run that long.

Pin sites changed (4) — and what was deliberately NOT changed

• .github/workflows/ci.yml:17 — the inline git clone --branch (the actual pin)
• CLAUDE.md:35 — "CI builds and tests against EigenScript v0.34.0"
• CLAUDE.md:248 — "Verified green on EigenScript v0.34.0" (re-verified today on v0.35.1)
• GAPS.md:16 — the "the CI pin is now v0.34.0" clause only

Left untouched, on purpose — these are the record of record:

• GAPS.md:16 "#772 shipped in EigenScript v0.34.0 (PR #773)" — a historical fact about where the loop-cap fix landed, in the same sentence as the pin clause. A blind sed would have falsified it.
• GAPS.md:17 "RE-BANK DONE (2026-07-31): … on the released v0.34.0" — the record of a past re-bank event.
• All of benchmarks/TSEITIN_LADDER.md — banked measurements and, at line 454, Pins: EigenScript v0.34.0 for the two live HF Space control lanes (ems-tseitin-4x6, ems-tseitin-5x5). Those lanes run on pinned v0.34.0 code; rewriting that line would falsify a live experiment's pin record. No Space was touched, redeployed, or paused — this repo bump does not affect them.

GAPS.md gains a new dated line recording the v0.35.1 re-verification, in the same style as the existing RE-BANK entry.

Preflight (BEFORE bumping, against the v0.35.1 runtime)

Runtime verified as the real tag build: tag v0.35.1, commit e2bb217, clean tree, upstream [70d] bounded-history gate passing with flat RSS.

EIGENSCRIPT_BIN=eigenscript bash tests/run_smoke.sh -> exit 0. Covers test_solver.eigs, test_tseitin_invariant.eigs, SAT/UNSAT fixtures across all four solver modes (default / --watched / --persistent / --cdcl), the six bench modes, and the vendored corpus. 74 ok=1 checks; the 9 ok=0 lines are the intentional bad-input diagnostic cases asserting their error checksums, not failures.

No consumer source was modified.

🤖 Generated with Claude Code