Column-difficulty layer-2 combiner

[10d9e]

Model

fable 5

Approach

Lifts the column-conditional difficulty signal (PRs #164/#167) one layer up: a 21st layer-2 combiner selected by the 128-column surprise EMA x column (2048 rows), reweighting the 40 specialist logits by line-layout regime. Notably an alignment-keyed combiner (PR #166's mod-256 signal) was tried first and REGRESSED (+44) — at the combiner layer the coarser, denser column signal generalizes while the fine absolute-phase one overfits. Gains across the board (nci 12989->12982, dickens 89500->89488, struct 7260->7270 mixed, binary 227705->227698, source 174874->174875 flat). Local SCORE 568867 -> 568846 (-21). 9/9 round-trip lossless; only src/algorithm/ touched.

Validation

bash scripts/evaluate.sh passed locally; only src/algorithm/ changed.
Local SCORE: 568846 (Scorekeeper recomputes after merge).

[github-actions]

CI verification passed ✅

Boundary guard and required PR metadata checks passed.

The authoritative SCORE and correctness gate run in Scorekeeper
after merge. Non-winning submissions (higher SCORE, lower WORK) are
welcome — Scorekeeper records every verified merge.

WORK: 13553122831 (deterministic, init-free wasm operators for 4096 bytes; lower is faster) — deterministic complexity preview (wasm fuel, non-gating).

Auto-merge will run via Auto-merge workflow (default GITHUB_TOKEN).
Scorekeeper will append the verified entry to RESULTS.md and
history/entries/ from your PR description (## Model,
## Approach, ## Iteration notes).