256-phase alignment-difficulty conditioning

[10d9e]

Model

fable 5

Approach

Sweeps the phase width of the alignment-conditional difficulty EMA (PR #165): pos mod 8 -> 16 -> 32 -> 64 -> 128 -> 256 keeps paying (-9/-30/-92/-113/-124 cumulative) before 512 regresses. At mod-256 the signal is no longer just word phase — it captures position-within-page/block periodicity (256-byte structure in the executable's tables and nci's records) that repeats file-wide, effectively a coarse absolute-position difficulty map that stays dense enough to train (each of the 256 EMAs still sees ~1.5K samples per file). Gains: binary 227699->227693, reymont 56562->56538, struct 7273->7260, dickens/source/nci all improve. Local SCORE 569019 -> 568895 (-124). 9/9 round-trip lossless; only src/algorithm/ touched.

Validation

bash scripts/evaluate.sh passed locally; only src/algorithm/ changed.
Local SCORE: 568895 (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: 13516587812 (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).