Graph evidence 3/8: card section — change-set completeness (change-impact) #10

Part of the graph-evidence integration. This is the wedge feature.

What

For each signature-change trigger (issue #8), run prism change-impact '<query>' on the head index and compare the returned site set against the files/lines the PR actually touches.

Card section:

### Change-set completeness
✔ TypeSerializer.writeTypePrefix — 38/38 required sites touched (completeness: closed)
⚠ JsonSerializer.serialize — 3 of 104 required sites NOT in this diff:
    src/.../ObjectNode.java:494
    src/.../JsonValueSerializer.java:565
    ...

Why this lands

Measured basis (paper §5.5, 162-run compile experiment): every answer that missed a change-site failed to compile; every complete one compiled. A missed site on this card is a pre-merge build-break catch, deterministically. LLM review bots guess; this proves.

Rules (non-negotiable, from the study)

• Relay discipline: render engine output as-is. No LLM re-processing between engine and card (measured: re-derivation drops real sites and adds spurious ones).
• Completeness field is part of the answer: project-local (external contract) must render as advisory ℹ, never as ✔/⚠ — the set is not authoritative there.
• Site matching is file+line-range against the diff hunks; a site inside a touched hunk counts as touched.

Acceptance

• shale-test-bed PR that changes a signature and updates all callers → green ✔ with counts.
• Same PR with one caller edit reverted → ⚠ listing exactly that site.
• A project-local query renders the advisory framing with the external contract named.