Daily Evidence Briefing

Full coverage changes the pipeline's epistemological status. Prior to tonight, the 1,081 unassessed entities were unknown — possibly quiet, possibly volatile. As of tonight, 1,134 entities are documented as quiet within the 14-day window, and 21 are documented as material. The benchmark can now distinguish between "no news" and "we did not look." This distinction matters for the published methodology: entities confirmed quiet are more defensible at their published scores than entities that were simply never rescanned.

The Anthropic reversal is a methodological case study, not an error. Two runs on the same entity, same night, produced opposite proposals (+1.5 vs. -6.6). The difference is the evidence-weighting rule: the old spec treated the most salient recent event (Mythos) as dominant; the new spec applies a 14-day window with recency weighting across all evidence, which surfaces the RSP rollback, the IL SB 3261 filing, and the Pentagon litigation as an aggregate negative cluster. Both outcomes are internally consistent with their respective specs. The new spec is more rigorous and should be treated as authoritative going forward.

The private prison sector has been structurally absent from the benchmark. GEO Group and CoreCivic are two of the most legally exposed companies in the Fortune 500 on human rights grounds — both running immigration detention facilities with documented forced labor, medical neglect, and inhumane conditions allegations. Neither had a published score baseline before tonight. The sector was invisible not because it lacked evidence, but because prior scanning did not surface it as a priority. Full coverage made it visible. The Adelanto trial and SPLC settlement are not new events — they are events the old scanner never retrieved.

Published scores for floor-adjacent entities are systematically understating negative evidence. Haiti and South Sudan both show published composites of 0.0 that appear to be display artifacts, not true assessments — raw subdimension averages produce 4.7 and 3.8 respectively. This is the same issue identified in prior digests for Russia (2.5 assessed vs. 0.0 published) and North Korea (1.3 assessed vs. 0.0 published). There is a methodology issue at the index floor: the display layer appears to clamp small non-zero scores to zero. This should be fixed before the next index build.

Sector alerts as a new intelligence product. The new scanner's sector sweep (T3) generated 6 sector alerts tonight — intelligence that would be invisible if the pipeline only ran entity-level individual searches. The DEI disclosure collapse (65% drop, Fortune 500) and the private prison cluster are both sector-level signals that do not attach to a single entity; they are background priors that affect scoring across dozens of entities. The pipeline should consider whether sector alerts warrant a separate published product distinct from entity scores.