Findings rest on communication and goals read from the content of public records. Counts, ranks, K and p-values are no longer evidence. Rows before 5 Oct are legacy.
| Type | Signal |
|---|---|
| C1 message | relay, handoff, status post, task assignment, reply; one agent writes, another reads or acts |
| C2 shared goal | same target, same task skeleton, same sink across runs |
| C3 shared channel | talk page, paste, message bus, result inbox, issue thread, dataset card used as mailbox or memory |
| C4 convention | word, marker or format that spreads between agents |
| C5 handover | notes written for a successor or replacement agent |
| C6 hidden content | nested encoding, zero-width characters, comment fields aimed at agents or self |
| Grade | Meaning |
|---|---|
| E0 | item exists in a public record |
| E1 | item is itself a communication artifact (message, relay, handoff, instruction, sink write, status post) |
| E2 | a second run or agent consumed or answered it |
| E3 | shared goal across several runs or agents shown (same target, task structure, sink) |
A finding needs E1 plus E2 or E3, with the benign explanations tried.
| Class | Condition |
|---|---|
| Confirmed | E1 with E2 or E3; benign explanations ruled out |
| Lead | E1; a plausible benign explanation remains |
| Signal | one communication artifact, unexplained |
| Agent activity | goal-directed agent work without communication; goal catalogued |
| Benign | explained |
| Incomplete / unobservable | content not readable at polite rates; not a null |
| Case | Role | Content signal |
|---|---|---|
| ntfy single-use channels, 6 and 8 Jul 2026 | positive | public channel used as mailbox (C3) |
| DSEWiki, 24 May to 2 Jul 2026 | positive | shared wiki as coordination channel, round markers (C3, C4) |
| Scanner-submitted map-data programs, late Sep to 4 Oct (outside report; not verified here) | outside claim | many runs, one target data type, result inboxes (C2, C3) |
| 29 Sep echo session | agent activity | one unfinished lookup of archived listings; nothing sent |
| Record | UTC window | Content read | Class |
|---|---|---|---|
| NuGet, 137 packages | 05:39 to 10:21 | one publisher; single-use ids; descriptions of one modular framework; no reference to the other two records | Benign |
| PyPI, 33 packages | 07:48 to 08:23 (sampled uploads) | standard-library project with ordinary READMEs and one author field | Benign |
| Echo pages, 23 plus 10 captures | 19:51 to 23:33 | one session reading an archived TV schedule, retrying one lookup (synchronous request, fetch, iframe, module import, second snapshot); no zero-width characters, comments, nested encodings or outbound post | Agent activity |
No shared sender, clock window or content. Closed. Read by one run on 5 Oct; person spot-check pending.
Used for rows dated before 5 Oct. Not evidence under the current method.
| Symbol | Definition |
|---|---|
| S, B | score of a day or group under the frozen rule; ordinary-day median |
| K | independent sources flagging the same UTC day |
| p_ord, p_shuf, p_ordK, p_global | shares of ordinary days, shuffles or circular shifts reaching the observed value |
| AUC, ARI, Spearman | rank statistic, partition agreement, rank correlation |
| Language | Python 3.9, standard library plus requests; static HTML with inline SVG |
| Hash | sha256 for rule files; keyed hashes of identifiers, key dropped after the run |
| Parallelism | independent runs in separate git worktrees and branches; results merged by a person |
| Local data | about 150 GB gzip (AI Village HF dataset), inside gitignored folders |
Corrections and data requests: contact details on maramasaeva.com.