[case+refine] rubric refinements: scope/ambiguity split + auto+human mode (v2.17.1)

[codxse]

Two authoring-rubric sharpenings that make stories easier for a budget-tier solver to follow, plus the doc note they surfaced. Rubric wording only — no behavioral/tooling change.

1. Budget-Solver Fit — split into two labeled axes

The checklist was headed "Too-large signals", but two of its bullets — an AC forcing an unsettled design decision, and a bugfix whose root cause isn't reproduced — are not size signals. They're ambiguity signals: a hole in the middle of an otherwise-small story. Under a "too-large" heading, an author sizing a small-but-ambiguous story could correctly conclude "not too large" and skip exactly the two checks that most reliably make a budget solver drift.

Now two groups — Too big (scope) (→ decompose or split) and Unsettled middle (ambiguity) (→ settle in the contract or split out) — under an opening line naming both axes as co-equal. INVEST mapping corrected: Estimable → ambiguity, Small → scope.

Caught by a Fable-tier review, which flagged the small-but-ambiguous story as the exact case the old single-heading list let slip.

2. Verification mode both → auto+human

both is relational — meaningless until you've memorized the other two modes — and the token travels into a story body, where the budget solver reading it never sees the rubric that defines it. auto+human names its own constituents, so it reads clearly wherever it lands. Its definition also moves from stating composition to stating the action: the solver auto-verifies the assertable part and spells out the experiential part for a person at /evaluate.

/solve and /evaluate cite the mode by name outside the shared rubric block, so their `human`/`both` call sites were updated by hand to `human`/`auto+human`.

3. Docs

CLAUDE.md gains the caveat those hand-edits taught: rubrics-sync.sh --write propagates only the shared block, so a token rename that other skills cite in prose needs its call sites fixed manually — the sync still reports green while the skills drift.

Verification

• rubrics-sync.sh green; no stray both token anywhere in the skills; four manifests valid and uniform at 2.17.1.
• Single commit; diff is 11 files, wording + version + changelog only.