Counting records answered a question the study no longer asks, so the census counts decisions

[MongLong0214]

Stacked on #825.

Two changes from the predecessor's census, both forced by the owner's estimand decision.

A missing Record-Id is no longer an exclusion. Under the old rule logic-pro-mcp qualified zero candidates because its whole history predates the field, and agent-control-plane qualified one. Identity is now identity_present, a metadata column that never votes.

The unit is a decision, not a record. One commit can rule out three alternatives for three different reasons, and each is a separate judgment an agent could revive. Both counts are reported side by side so the change of unit reads as a change of unit rather than as a corpus that grew: 157 records carrying an explicit reason, 241 decisions inside them.

repository	records	with a reason	decisions	identified
gitseed	84	71	104	94
agent-operator-score	155	30	59	48
logic-pro-mcp	53	29	43	0
agent-control-plane	90	27	35	1

The six that were nearly lost

Six of agent-control-plane's decisions sit outside any record block — Ruled-out: lines from an earlier body that a squash merge left as prose in the middle of the message. The product's query does not read them, and neither did the first pass of this census. Dropping them at discovery would have hidden the attrition inside the word "census", so they are enumerated as ordinary-source and the delivery gate decides them on the evidence.

The legacy exclusion index matched nothing here. That is a fact about these four repositories, not a working filter: the three record ids it names are not in any of them. The test proves the filter fires by putting a matching record in front of it first.

Verification

Every Ruled-out: line in the four sealed snapshots is accounted for. An independent git log | grep -c over each bundle gives 104, 59, 43 and 35, matching the census exactly — the first pass gave 29 for agent-control-plane, and that six-line gap is what surfaced the squash-body case.

7 tests pass, both typechecks clean.

---

Note on the stack: #825's merge landed on this branch's base rather than on main, so this PR now also carries the decision audit anchor commit it was stacked on. Both are in its diff.

[github-actions]

CommitLore — record lint

Trailers: clean — 1 commit in origin/cdeb-v4-decision-anchor..c3ada47ec4ec763269ada05478dfd6782261dd32
Active constraints: 1 limits · 3 ruled-out · 0 warnings — from 1 record over 6 changed paths

Limits (1)

• r-v4census c3ada47 — this enumerates decisions carrying an explicit Ruled-out: reason. A decision recorded only in an ADR, a code comment or a pull request body is outside what this census can see, and the pool is a lower bound on the history, not a measurement of it

Ruled out (3)

• r-v4census c3ada47 — keeping the record as the unit | a record with three rulings would contribute one candidate and two rulings would go unmeasured, which is a selection decision hidden in a counting convention
• r-v4census c3ada47 — skipping the six squash-body rulings because they cannot be delivered | that is a G6 verdict, and reaching it at discovery removes the row that would have recorded it
• r-v4census c3ada47 — re-reading the predecessor's qualification outcomes | they were decided under the identity rule this study discards, so importing them would reinstate that rule under a new name

git log --follow accepts exactly one pathspec, so renames are not followed for 6 paths; query one path at a time to follow its rename chain

_{Trailer violations fail this check. Active constraints are informational — they are what the repository already decided, not a verdict on this PR.}