Conciseness

The body is ~430 lines, of which roughly 390 are an inlined Lean 4 formal block whose axiom doc comments repeatedly restate judgments that also appear in FLOW, MORPHISM, TOOL GROUNDING, and the Rules section (e.g. "an acknowledgment of your explanation is acceptance, not the person's own reading" appears at least four times). Anchor 2 fits: several padded and redundant sections; it is not anchor 1 because it does not explain concepts Claude already knows — the redundancy is bespoke rather than filler — and not anchor 3 because the duplication is pervasive, not occasional.

Actionability

There is genuinely specific guidance — "cite concrete locations in the target — file and line where it is code", checks write to "a scratch space outside the target's tree", contradictions must show "their words quoted; the narrowest material... never your own explanation" — but no worked example or template of an actual round or map, and the central instructions are compressed into a single ~500-word run-on sentence in the `.present` entry. Anchor 3 fits: some concrete guidance but incomplete, with key details (what output looks like) missing.

Workflow Clarity

The sequence (gather → check → map → present → fuse → close) is present in the FLOW and MORPHISM comments and encoded in `grasp`/`start`, and a feedback loop exists (checks run before rounds, results return through `consult`), but it is implicit: the reader must reverse-engineer the Lean formalism, and no explicit ordered steps or checkpoints are laid out for the executor. Anchor 3 fits; not 2 (the sequence is complete, not gappy) and not 4 (checkpoints are implicit rather than "most present").

Progressive Disclosure

The body has clear section headers and one well-signaled, one-level-deep reference (`references/round-composition.md`, a real file with explicit when-to-read conditions), but roughly 90% of the content — the Lean formal contract — is inlined in SKILL.md where it belongs in a separate reference file. Anchor 3's "content that should be separate is inline" matches exactly; not 2 because the structure is not minimal and the reference is not buried, and not 4 because the inlining is wholesale rather than a minor gap.

Specificity

The description names the domain ("code, a document, a result") but the actions are minimal and abstract — "a map checked with them" is metaphorical rather than a concrete verb like extract, verify, or render. It is not entirely vague (anchor 1) because target types and an action sketch are present, but it lacks the 1-2 concrete actions needed for anchor 3.

Completeness

The "what" ("they can then say their understanding serves their purpose, on a map checked with them") is vague, and the "when" ("can't follow yet or nods along to unsure") is only weakly implied with no explicit 'Use when...' clause. Anchor 2 fits best: a vague what with a weakly implied when; anchor 3 does not fit because the what is not clear.

Trigger Term Quality

"can't follow", "code", "document", and "a result" are natural terms a user might say when needing this skill, but common variations like "explain", "walk me through", or "I don't understand this" are missing. Coverage is moderate, matching anchor 3 rather than anchor 2 (which has only one or two generic keywords) or anchor 4 (good coverage with few missing).

Distinctiveness Conflict Risk

The comprehension-verification framing ("the user can't follow yet... their understanding serves their purpose") carves a niche distinct from file-manipulation skills, but the abstract wording could still overlap with generic explanation/teaching behavior. Anchor 3 fits; not 2 (the comprehension framing is not broad like "helps with document files") and not 4 (no distinct concrete triggers are given).