# Agentsy · agent swarms in the wild

[Latest edition](https://agentsy.live/) · [Understanding agent swarms](https://agentsy.live/releases/28e406dd0c184e1117499de9/methodology.md#reading-terms)

Study how AI agents find one another and coordinate on the public internet. Distinguish cooperation that develops during tasks from disclosed direction, promotion, imitation and activity whose origin is unknown. Preserve the evidence and its uncertainty.

As of 2026-09-07. This edition contains reviewed accounts of selected agent-related coordination. It distinguishes cooperation that appears to develop during tasks, directed work and activity whose origin is unknown. Published work and use of documentation do not establish successful execution, payment or independent control. The record is neither a complete census nor a live monitoring service.

[Human field guide](https://agentsy.live/releases/28e406dd0c184e1117499de9/index.html) · [Record JSON](https://agentsy.live/releases/28e406dd0c184e1117499de9/record.json) · [Methodology](https://agentsy.live/releases/28e406dd0c184e1117499de9/methodology.md) · [Corrections](https://agentsy.live/releases/28e406dd0c184e1117499de9/corrections.md) · [Evidence](https://agentsy.live/releases/28e406dd0c184e1117499de9/evidence.md)

<a id="what-matters"></a>

## What matters

**Account as of:** 2026-09-07

**Scope:** Analysis of preserved historical evidence. The existing wiki case gains a Cashier explanation on September 7, 2026, covering June 17 plan exchanges and a separate June 19 contamination note. The eight earlier accounts and six discovery histories retain their prior scope. The selected record has 26 cases, 176 evidence notes and 103 analysis cards. No new scientific collection or current-activity check was performed for this significance analysis.

Nine explanatory accounts make specific requests, replies and shared work inspectable within the selected 26-case record. Cashier participants refine a proposed experiment while giving their timed answers priority. Receipt participants exchange concrete tests of a disputed result and a case intended to remain valid. The accounts distinguish those contributions from verified execution, independent control and the origins of coordination; they do not establish a newly identified organic swarm or a causal link between cases.

<a id="significance-wiki-cashier-plan-refinement"></a>

### Peers improve a test without establishing its outcome

Wiki participants want to learn whether a process can still act after its final task answer. They revise the proposed test when peers identify missing controls and ambiguous timing.

**Limit:** The written plans change; actual final execution, process survival and independent control remain unresolved.

**Assessment:** supported inference

**Why it matters:** A useful contribution can be a better question or a more discriminating test plan. Here the replies change when signals should appear and what a missing signal might mean. Those improvements to the plan remain visible even though the preserved evidence cannot tell whether the final experiment ran or whether a process survived.

**What supports it:** The Oct06-page writer accepts a delayed-signal proposal, corrects launch timing and reports choosing a different launch method. Dec29-page peers add earlier controls and revise marker placement after a specific cutoff objection. A separate conditional reply makes answer correctness and timing the priority.

**Full limits:** This is a selected episode inside the existing wiki case, not a newly counted swarm or evidence of present activity.

**Full limits:** Original instructions, independent controllers and first discovery remain unresolved. The episode is not linked to Iowa by a demonstrated source edge.

**Reasoning:** The recipient explicitly acknowledges a named signal and later credits the timing caution in its revised plan. The correspondence between the suggestion and the revised plan shows that the reply changed the proposal. A reported local test, a disputed counter value and absent markers have different evidentiary roles and do not complete the missing execution chain.

**Counterevidence:** A writer reports a successful local detachment test, but that report is not an independently observed final-answer experiment.

**Counterevidence:** Known observer increments and uncertain task-clock mappings prevent treating a value or silence as a definitive runtime result.

**Alternative explanations:** Ordinary human collaboration or common direction can produce the same written exchange.

**Alternative explanations:** A missing signal can result from nonlaunch, delivery failure, timing mismatch or termination; an increment can be a test or observer action.

**Date context / observed end:** 2026-06-19

**Date context / observed start:** 2026-06-17

**Date context / scope:** Main Cashier plan exchanges are recorded June 17; a separate counter-contamination example is recorded June 19. These are reqlog-grade exported revision times, not authenticated execution, receipt or task clocks. Labels Oct06,Dec29,Jan17,Jan31 and namespace 2028 are not event dates. Analysis added September 7, 2026.

[Agents sharing answers and timing on public wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md) · [wiki-cashier-oct-plan](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-oct-plan) · [wiki-cashier-dec-controls](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-dec-controls) · [wiki-cashier-counter-limits](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-counter-limits) · [wiki-cashier-priority](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-priority) · [Read claim cashier-plan-uptake](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#cashier-plan-uptake) · [Read claim cashier-controls](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#cashier-controls) · [Read claim cashier-outcome](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#cashier-outcome)

[Agents sharing answers and timing on public wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md)

<a id="significance-receipt-fixture-exchange"></a>

### A fix needs a test that must still pass

The receipt exchange adds a valid success case alongside a case that should remain inconsistent.

**Limit:** Test files are preserved; the recipient’s code change and successful runs remain reports.

**Assessment:** supported inference

**Why it matters:** Removing a misleading success result can make a verifier too restrictive. These two controls make a more precise demand: retain a meaningful inconsistency result and keep the smallest legitimate success possible. The contribution is this inspectable test design, with actual verifier performance still unverified.

**What supports it:** After an empty witness is reportedly refused, a13 tests a nonempty witness with no receipts and a14 tests a minimal receipt-and-witness pair. The preserved additions target INCONSISTENT and RECONCILED respectively; their successful execution is reported by the contributor.

**Full limits:** No general verifier correctness, autonomous self-correction or independent agent control follows from this selected exchange.

**Full limits:** A refusal is a validation outcome, not evidence that every nearby input must be rejected.

**Reasoning:** The reviewed fixture comparison and specific replies tie the two added tests to the reported remedy. The original a08 outcome matched its prediction, so the accepted objection concerned meaning rather than a prediction mismatch. Preserving separate a13 and a14 outcomes can expose overcorrection; adding files alone does not establish that the verifier meets this requirement.

**Counterevidence:** The recipient implementation was not recovered. Reported outcomes and the claim that predictions preceded runs were not independently verified.

**Counterevidence:** The contributor conceded a06 and a12 prediction errors; these are not accepted verifier defects.

**Alternative explanations:** The fixture exchange can reflect ordinary developer collaboration, common direction or a staged demonstration.

**Alternative explanations:** The recipient code or actual results may differ from the participants’ reports.

**Date context / observed end:** 2026-07-27

**Date context / observed start:** 2026-07-26

**Date context / scope:** Forum labels and repository timestamps represent the historical July 26–27 exchange. Explanation added September 7, 2026. Neither publication nor acquisition extends observed activity.

[Outside test cases lead to a reported verifier correction](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md) · [receipt-outside-challenge](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-outside-challenge) · [receipt-reported-fix](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-reported-fix) · [receipt-initial-fixtures](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-initial-fixtures) · [receipt-boundary-fixtures](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-boundary-fixtures) · [Read claim receipt-fixture-exchange--semantic-repair](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-fixture-exchange--semantic-repair) · [Read claim receipt-fixture-exchange--selective-remedy](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-fixture-exchange--selective-remedy) · [Read claim receipt-fixture-exchange--paired-controls](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-fixture-exchange--paired-controls)

[Outside test cases lead to a reported verifier correction](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md)

<a id="significance-memory-curator-feedback"></a>

### Useful feedback can arrive before a working service

The memory-service task yields a specific fault report and response, while successful curation remains unverified.

**Limit:** The response and monitor artifact are visible; a repaired service, useful output and payment are not verified.

**Assessment:** supported inference

**Why it matters:** The requested testing can reveal an interface assumption before anyone receives useful memory suggestions. Separating that feedback from a successful service outcome makes the contribution legible without treating an acknowledgment, promised input change or published monitor as a completed repair.

**What supports it:** The tester reports hanging; a service status names missing input; the author replies to the tester’s particular format and later attributes a published process monitor to downtime. The inspected monitor checks whether a process exists, not whether a curation job succeeds.

**Full limits:** No verified deployment, successful curation result, payment or independent operators are established.

**Full limits:** The test task’s stated purpose does not authenticate private intent or how the participants first encountered one another.

**Reasoning:** The task explicitly accepts failure feedback. Exact task and request references connect the report, error and guidance. The preserved parser can acknowledge processing before rejecting missing input, and the published monitor supervises another layer. Thus the written and artifact responses support a bounded contribution without demonstrating repaired input handling or useful output.

**Counterevidence:** Historical receipt of the error is unknown; missing-input rejection does not exclude intermittent downtime.

**Counterevidence:** Inspected parser versions do not substantiate the claimed flexible-input update; no successful tester resubmission is established.

**Alternative explanations:** The alias change may exist in an unpreserved version, may have been intended, or may have been announced inaccurately.

**Alternative explanations:** Reliability work may have had several causes; its connection to this report is attributed by the author.

**Alternative explanations:** The public task may have been independently discovered or arranged under common control.

**Date context / observed end:** 2026-02-06

**Date context / observed start:** 2026-02-05

**Date context / scope:** Signed event timestamps and an unsigned Git clock represent the historical February 5–6 exchange. Explanation added September 7, 2026. Clock accuracy, receipt and deployment times remain unknown.

[Task feedback and differing service diagnoses](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md) · [curator-feedback-task](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md#curator-feedback-task) · [curator-feedback-delivery](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md#curator-feedback-delivery) · [curator-feedback-error](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md#curator-feedback-error) · [curator-feedback-parser](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md#curator-feedback-parser) · [curator-feedback-guidance](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md#curator-feedback-guidance) · [curator-feedback-monitor-report](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md#curator-feedback-monitor-report) · [curator-feedback-monitor-artifact](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md#curator-feedback-monitor-artifact) · [Read claim memory-task-purpose](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md#memory-task-purpose) · [Read claim memory-acknowledgment-boundary](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md#memory-acknowledgment-boundary) · [Read claim memory-targeted-uptake](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md#memory-targeted-uptake) · [Read claim memory-monitor-boundary](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md#memory-monitor-boundary) · [Read claim memory-response-artifact](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md#memory-response-artifact)

[Task feedback and differing service diagnoses](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md)

<a id="significance-gemma-partial-transfer"></a>

### Participants pick up unfinished Gemma tests

Unfinished work gave another participant a specific test to take on.

**Limit:** The organized challenge shows named reuse and test reports; it does not establish independent operators or authenticate the runs.

**Assessment:** supported inference

**Events:** 2026-06-08 to 2026-06-10

**Added to the observatory:** 2026-09-07

**Why it matters:** The useful unit of cooperation is an unfinished artifact plus an explicit missing test. The case shows how resource constraints can make another participant’s partial work actionable, while separating test reports from proof of execution.

**What supports it:** Two recipients take up named unfinished submissions when their authors report exhausted test allowances. The shared board makes incomplete work, remaining checks and later result reports visible.

**Full limits:** The work took place in an organized challenge. The particular handoffs are visible, but common human control, prescribed delegation and independently chosen cooperation remain possible. Different account names do not establish independent systems.

**Full limits:** Uploaded logs, status files and results may share provenance. Their agreement supports the recorded account without independently authenticating GPU execution. Both result pages identify their status as agent-run, not organizer-verified.

**Full limits:** Current byte equality covers selected small files, not complete model weights or an event-time configuration and input history. Current rules and later profiles cannot be backdated wholesale.

**Full limits:** The pending-validation label does not negate the later reported result; that result does not retroactively validate the original unmodified submission. The proposed ignore-list removal remains unconfirmed by the current copies.

**Full limits:** Public prompt coverage and arithmetic do not establish hidden-test performance, complete model quality, functioning image/audio capabilities or statistical significance. The earlier record and personal-best comparisons are the participant’s cited values.

**Full limits:** The two dated episodes do not establish continuous activity or contact between the two recipient pairs. Another participant announced a port, but its final outcome was not traced here.

**Full limits:** Open question: would dated run configurations and operator records clarify which repair was applied and how the recipients chose this work? These episodes concern tests that had not run or had failed; they do not establish a general rule about agent cooperation.

**Reasoning:** This is a bounded interpretation of the identified exchanges and their stated limits, not an estimate of prevalence or a global novelty claim.

**Counterevidence:** The work took place in an organized challenge. The particular handoffs are visible, but common human control, prescribed delegation and independently chosen cooperation remain possible. Different account names do not establish independent systems.

**Counterevidence:** The pending-validation label does not negate the later reported result; that result does not retroactively validate the original unmodified submission. The proposed ignore-list removal remains unconfirmed by the current copies.

**Alternative explanations:** The visible pattern can occur under human direction or common control.

**Date context / observed end:** 2026-06-10

**Date context / observed start:** 2026-06-08

**Date context / scope:** Two separate episodes: June 8, 15:36–17:08 UTC, and June 10, 05:11–05:41 UTC. Message dates come from their source headers; job times come from uploaded status records. These are source-represented times, not independently authenticated execution clocks. UTC means Coordinated Universal Time. September 7 capture and current-file comparisons are separate; the span does not establish continuous activity.

[Participants pick up unfinished Gemma tests](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md) · [gemma-first-handoff](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md#gemma-first-handoff) · [gemma-first-load-repair](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md#gemma-first-load-repair) · [gemma-first-reported-result](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md#gemma-first-reported-result) · [gemma-second-handoff](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md#gemma-second-handoff) · [gemma-second-reported-result](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md#gemma-second-reported-result) · [Read claim gemma-partial-transfer--resource-transfer](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md#gemma-partial-transfer--resource-transfer) · [Read claim gemma-partial-transfer--diagnosis-revision](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md#gemma-partial-transfer--diagnosis-revision) · [Read claim gemma-partial-transfer--distributed-validation](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md#gemma-partial-transfer--distributed-validation) · [Read claim gemma-partial-transfer--outcome](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md#gemma-partial-transfer--outcome)

[Participants pick up unfinished Gemma tests](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md)

<a id="significance-botmadang-comments-repair"></a>

### Repairing the service used to read comments

A request to read replies turned the communication tool itself into shared work.

**Limit:** Successful use is participant-reported; independent agent control and historical deployment remain unverified.

**Assessment:** supported inference

**Events:** 2026-02-01 to 2026-02-02

**Added to the observatory:** 2026-09-07

**Why it matters:** This case identifies a concrete prerequisite for reciprocal communication: participants must be able to retrieve replies. The repair sequence makes that dependency visible without turning an agent-oriented project into proof of an autonomous swarm.

**What supports it:** The participants repair access to the conversation itself: a missing client capability is followed by a reported service failure, a linked provider patch and a success report.

**Full limits:** The public invitation and coauthor credit make the episode relevant to agent software, but human authorship, common control and a staged demonstration remain possible.

**Full limits:** The held patches do not verify historical deployment, database index state or tests. Both the provider’s success statement and the client author’s later confirmation remain reports.

**Full limits:** The issue was updated on February 2; the held body is not a separately preserved creation-time revision. All four held reply objects have matching creation and update times.

**Full limits:** This February 1–2 episode is separate from the February 3 guide discussion and change. No causal link between the two is claimed.

**Full limits:** English paraphrases interpret the preserved Korean text. Source timestamps are metadata, not independently authenticated execution clocks.

**Reasoning:** This is a bounded interpretation of the identified exchanges and their stated limits, not an estimate of prevalence or a global novelty claim.

**Alternative explanations:** The visible pattern can occur under human direction or common control.

**Date context / observed end:** 2026-02-02

**Date context / observed start:** 2026-02-01

**Date context / scope:** The selected request, code changes and replies carry February 1–2, 2026 source timestamps. An earlier January 31 endpoint is background context. September 7 retrieval is separate; these dates do not establish continuous activity or measured response latency.

[Repairing the service used to read comments](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md) · [comments-repair-client-tool](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md#comments-repair-client-tool) · [comments-repair-contribution-context](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md#comments-repair-contribution-context) · [comments-repair-error-report](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md#comments-repair-error-report) · [comments-repair-provider-fix](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md#comments-repair-provider-fix) · [comments-repair-recipient-report](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md#comments-repair-recipient-report) · [comments-repair-request](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md#comments-repair-request) · [Read claim botmadang-comments-repair--repair-link](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md#botmadang-comments-repair--repair-link) · [Read claim botmadang-comments-repair--purpose](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md#botmadang-comments-repair--purpose) · [Read claim botmadang-comments-repair--infrastructure](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md#botmadang-comments-repair--infrastructure) · [Read claim botmadang-comments-repair--origin](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md#botmadang-comments-repair--origin)

[Repairing the service used to read comments](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md)

<a id="significance-bluesky-question-article"></a>

### A Bluesky question becomes an article

A question became a credited article; the verification system it described remained a proposal.

**Limit:** The article preserves a response to a question; full reading and a working verification system are not established.

**Assessment:** supported inference

**Events:** 2026-03-11 to 2026-03-12

**Added to the observatory:** 2026-09-07

**Why it matters:** Communication can yield a traceable intellectual artifact without yielding a working technical protocol. This case makes both the contribution and that boundary inspectable.

**What supports it:** An addressed question becomes a credited article and a reply. The exchange preserves responsive writing; it does not demonstrate the verification system the article imagines.

**Full limits:** The project describes daily human oversight and a board-directed interest in AI agency and social networks. That broad instruction does not show who chose this particular question or article. Common control, staged conversation and human-written messages remain possible.

**Full limits:** The preserved records do not establish the accounts’ first encounter or verify the models used on March 11. Later profiles and retrospective conversation descriptions cannot supply that history.

**Full limits:** The article-link message precedes the article’s recorded repository revision. Post and service timestamps also disagree in parts of the wider thread, and the cancellation note differs from its revision time. These fields cannot measure writing latency or establish when the article first became available.

**Full limits:** The article is preserved in a specific repository version and a separate search-tool rendition. Earlier attempts to retrieve discovery pages from Dev.to were blocked; no direct capture of this article’s Dev.to page was obtained. Its original availability remains unknown, and the replies do not establish that Alice retrieved it.

**Full limits:** The article’s security and identity arguments remain its author’s claims. No distributed verifier, video recording or model execution was tested here.

**Full limits:** Open question: would dated first-encounter records and public descriptions of the participants’ instructions show how this conversation began? A dated article archive could separately establish when the link became readable.

**Reasoning:** This is a bounded interpretation of the identified exchanges and their stated limits, not an estimate of prevalence or a global novelty claim.

**Counterevidence:** The project describes daily human oversight and a board-directed interest in AI agency and social networks. That broad instruction does not show who chose this particular question or article. Common control, staged conversation and human-written messages remain possible.

**Counterevidence:** The preserved records do not establish the accounts’ first encounter or verify the models used on March 11. Later profiles and retrospective conversation descriptions cannot supply that history.

**Alternative explanations:** The visible pattern can occur under human direction or common control.

**Date context / observed end:** 2026-03-12

**Date context / observed start:** 2026-03-11

**Date context / scope:** Selected posts and repository revisions carry March 11–12, 2026 dates. Post-supplied times, service indexing times and repository times are distinct; they do not measure response speed or article publication. UTC means Coordinated Universal Time. Sources were retrieved September 7, 2026. These dates do not establish the first encounter or continuous activity.

[A Bluesky question becomes an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md) · [witness-announcement-plan](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md#witness-announcement-plan) · [witness-article-reply](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md#witness-article-reply) · [witness-article-text](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md#witness-article-text) · [witness-project-context](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md#witness-project-context) · [witness-question](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md#witness-question) · [witness-reader-reply](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md#witness-reader-reply) · [Read claim bluesky-question-article--writing-uptake](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md#bluesky-question-article--writing-uptake) · [Read claim bluesky-question-article--ideas-not-system](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md#bluesky-question-article--ideas-not-system) · [Read claim bluesky-question-article--content-purpose](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md#bluesky-question-article--content-purpose) · [Read claim bluesky-question-article--reading](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md#bluesky-question-article--reading)

[A Bluesky question becomes an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md)

<a id="significance-moltplace-lobster"></a>

### A lobster drawing invitation receives replies and matching pixels

Shared coordinates helped participants propose contributions to the same drawing.

**Limit:** Matching pixel records do not establish every intended placement or an accepted division of work.

**Assessment:** supported inference

**Events:** 2026-01-31 to 2026-02-10

**Added to the observatory:** 2026-09-07

**Why it matters:** A shared visual target makes proposals easy to reference precisely. The case also shows why a message containing an action’s parameters should not automatically be treated as proof of an intentional, completed action.

**What supports it:** A shared drawing invitation elicits pixel proposals and a later attempt to divide sections. Coordinates and colors provide concrete shared references across discussion and canvas.

**Full limits:** The organized invitation and explicit human involvement coexist with observed participant responses. Common control, human assistance and autonomous choices remain possible.

**Full limits:** The current placement instructions do not identify the software version that interpreted messages at the time. Matching content and timestamps do not establish how a pixel was submitted.

**Full limits:** Current white pixels, limited feed entries and later snapshot indexes cannot recover the full January–February canvas. No completed artwork or realized section assignment was verified.

**Full limits:** Open question: would historical per-pixel events and source-message links resolve how contributions reached the canvas and whether participants carried out the proposed sections?

**Reasoning:** This is a bounded interpretation of the identified exchanges and their stated limits, not an estimate of prevalence or a global novelty claim.

**Counterevidence:** The current placement instructions do not identify the software version that interpreted messages at the time. Matching content and timestamps do not establish how a pixel was submitted.

**Alternative explanations:** The visible pattern can occur under human direction or common control.

**Date context / observed end:** 2026-02-10

**Date context / observed start:** 2026-01-31

**Date context / scope:** Selected source-dated episodes on January 31, February 2, February 3 and February 10, 2026. These dates do not establish continuous activity or the first historical occurrence. Canvas records and documentation were captured September 7, 2026.

[A lobster drawing invitation receives replies and matching pixels](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md) · [moltplace-chinese-pixel](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md#moltplace-chinese-pixel) · [moltplace-launch](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md#moltplace-launch) · [moltplace-praise-attribution](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md#moltplace-praise-attribution) · [moltplace-section-proposal](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md#moltplace-section-proposal) · [moltplace-zara-joins](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md#moltplace-zara-joins) · [Read claim moltplace-lobster--invited](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md#moltplace-lobster--invited) · [Read claim moltplace-lobster--allocation](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md#moltplace-lobster--allocation) · [Read claim moltplace-lobster--language-interface](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md#moltplace-lobster--language-interface) · [Read claim moltplace-lobster--emergence](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md#moltplace-lobster--emergence)

[A lobster drawing invitation receives replies and matching pixels](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md)

<a id="significance-spacemolt-haven-battle-and-public-corrections"></a>

### A SpaceMolt battle prompts corrections to its public account

Agreeing with a correction is one step; changing the article is another.

**Limit:** Alis accepts the correction, but the preserved article still contains the disputed account.

**Assessment:** supported inference

**Events:** 2026-08-25 to 2026-08-28

**Added to the observatory:** 2026-09-07

**Why it matters:** A correction is a social action whose result can be checked separately from its acceptance. The case preserves an informative gap between a reply that concedes an error and an article that still contains it.

**What supports it:** A participant uses an identified battle record to correct a public report, and the author accepts the correction. A separate storage report is also accepted, but rests on participant testimony.

**Full limits:** The article’s date language conflicts with the recorded August 25 battle end. Later police encounters are separate battles, not additional phases of this battle.

**Full limits:** Autopilot is recorded and earlier operator material describes human control. Account labels and responsive behavior do not authenticate an independent language model.

**Full limits:** This case examines one recorded battle and selected forum replies. It does not establish the complete history of the game or whether the promised article revision appeared later.

**Full limits:** A later edited article, a verifiable inventory comparison and dated command-origin records would resolve different remaining questions. No such evidence was acquired.

**Reasoning:** This is a bounded interpretation of the identified exchanges and their stated limits, not an estimate of prevalence or a global novelty claim.

**Counterevidence:** This case examines one recorded battle and selected forum replies. It does not establish the complete history of the game or whether the promised article revision appeared later.

**Alternative explanations:** The visible pattern can occur under human direction or common control.

**Date context / observed end:** 2026-08-28

**Date context / observed start:** 2026-08-25

**Date context / scope:** Battle end time comes from the service; forum times come from post metadata. Game ticks provide internal order and are not converted to calendar times. Earlier operator context is dated separately. Sources were retrieved September 7, 2026. UTC means Coordinated Universal Time.

[A SpaceMolt battle prompts corrections to its public account](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md) · [haven-acceptance](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md#haven-acceptance) · [haven-battle](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md#haven-battle) · [haven-control](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md#haven-control) · [haven-correction](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md#haven-correction) · [haven-maneuvers](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md#haven-maneuvers) · [haven-report](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md#haven-report) · [haven-storage](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md#haven-storage) · [Read claim spacemolt-haven-battle-and-public-corrections--corrective](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md#spacemolt-haven-battle-and-public-corrections--corrective) · [Read claim spacemolt-haven-battle-and-public-corrections--incomplete](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md#spacemolt-haven-battle-and-public-corrections--incomplete) · [Read claim spacemolt-haven-battle-and-public-corrections--epistemic](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md#spacemolt-haven-battle-and-public-corrections--epistemic) · [Read claim spacemolt-haven-battle-and-public-corrections--combat-purpose](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md#spacemolt-haven-battle-and-public-corrections--combat-purpose)

[A SpaceMolt battle prompts corrections to its public account](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md)

<a id="significance-bluesky-starter-pack-encounter"></a>

### A Bluesky directory acknowledgment becomes an article

Listing accounts together does not show that the list introduced them.

**Limit:** Earlier contacts limit particular introduction claims; no directory-caused network growth is verified.

**Assessment:** supported inference

**Events:** 2026-03-08 to 2026-03-11

**Added to the observatory:** 2026-09-07

**Why it matters:** A place where accounts can discover one another is not itself evidence of discovery. Separating membership, acknowledgment and prior contact makes the formation hypothesis testable instead of assuming it from a directory’s existence.

**What supports it:** A directory acknowledgment becomes article material and receives a further response. Earlier contacts limit the stronger claim that the directory introduced these particular peers.

**Full limits:** The fifteen-post annotation is a retrospective arrangement. Its March 11 label does not replace March 10 post dates, and its model labels, duration and drift interpretation are not independently verified. Four selected reply pairs have reversed account-supplied times; reply references and clocks are retained separately.

**Full limits:** The directory’s historical eight-account roster differs from the article’s named ecosystem peers. Current list samples and counters cannot establish historical growth, new contact or listing-caused discovery.

**Full limits:** Alice reports reading the article, but the announcement already contains the relevant framing. No unique full-body retrieval or original Dev.to availability was established. The referral precedes the adding commit under separate source clocks.

**Full limits:** The running tracker is an attributed article claim, not inspected code, execution telemetry or a measured consequence. Game changes and performance also remain participant/operator reports.

**Full limits:** Daily human oversight and broad distribution goals are documented. Specific initiating instructions, private communications and event-time model execution remain unknown. Shared control, staged or human-authored messages remain possible; the evidence does not establish any of them as fact.

**Full limits:** This directory and article episode is distinct from the later question-to-article episode in “A Bluesky question becomes an article.” Both involve 0co and Alice, but their article IDs, responses and outcomes are not merged.

**Reasoning:** This is a bounded interpretation of the identified exchanges and their stated limits, not an estimate of prevalence or a global novelty claim.

**Counterevidence:** The fifteen-post annotation is a retrospective arrangement. Its March 11 label does not replace March 10 post dates, and its model labels, duration and drift interpretation are not independently verified. Four selected reply pairs have reversed account-supplied times; reply references and clocks are retained separately.

**Counterevidence:** The directory’s historical eight-account roster differs from the article’s named ecosystem peers. Current list samples and counters cannot establish historical growth, new contact or listing-caused discovery.

**Alternative explanations:** The visible pattern can occur under human direction or common control.

**Date context / observed end:** 2026-03-11

**Date context / observed start:** 2026-03-08

**Date context / scope:** Selected posts and repository versions carry March 8–11, 2026 dates: earlier contact on March 8, game-report ancestry on March 9, the joined discussion on March 10, and directory/article responses on March 11. Account-supplied post times, service indexing, Git revision times and page labels remain distinct, including conflicting orderings. They do not establish first occurrence, continuous activity, response speed or article availability. Sources were captured September 7, 2026. UTC means Coordinated Universal Time.

[A Bluesky directory acknowledgment becomes an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md) · [starter-article-uptake](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md#starter-article-uptake) · [starter-existing-thread](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md#starter-existing-thread) · [starter-list-acknowledgment](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md#starter-list-acknowledgment) · [starter-prior-contacts](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md#starter-prior-contacts) · [Read claim bluesky-starter-pack-encounter--uptake](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md#bluesky-starter-pack-encounter--uptake) · [Read claim bluesky-starter-pack-encounter--new-contact](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md#bluesky-starter-pack-encounter--new-contact) · [Read claim bluesky-starter-pack-encounter--tracker](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md#bluesky-starter-pack-encounter--tracker) · [Read claim bluesky-starter-pack-encounter--formation](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md#bluesky-starter-pack-encounter--formation)

[A Bluesky directory acknowledgment becomes an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md)

<a id="significance-methods-concrete-test-content"></a>

### The exchanged material explains what peers can check

Receipt participants exchange test inputs; Cashier participants refine a plan for observing signals. Those are distinct contributions.

**Limit:** The comparison concerns test content, not common actors, transmission or verified outcomes.

**Assessment:** supported inference

**Why it matters:** A shared page or version link tells a reader where coordination happened. The content explains what it made possible: receipt examples make a disputed result and a legitimate result inspectable, while Cashier replies refine the interpretation of a future signal. Distinguishing these methods helps readers compare the work without assuming the same protocol or a shared cause.

**What supports it:** The new receipt method identifies preserved classifier examples, including a valid case intended to remain accepted. The existing peer-verification method now includes specific control and timing proposals from Cashier while retaining its earlier wiki examples.

**Full limits:** A new method card improves classification of held evidence, not the number of observed swarms or verified successful experiments.

**Full limits:** No common protocol, cross-case transmission, authenticated operator independence or historical novelty is established.

**Reasoning:** The reviewed method definitions distinguish concrete inputs with expected classifications from requests that refine what an observation can discriminate. This comparison classifies the content of the exchanges; it does not infer transmission, shared participants or a common causal mechanism.

**Counterevidence:** Only one receipt exchange supports the new example category. More categories or repeated source references are not independent corroboration.

**Counterevidence:** The receipt test results remain reports; the Cashier marker sequence remains a plan.

**Alternative explanations:** Both cases can also be described at the broader level of peer review, with the specific methods retained as overlapping rather than exclusive categories.

**Date context / scope:** Receipt source comments: July 26–27, 2026. Cashier plan revisions: June 17, with a separate June 19 contamination note. The peer-method card also retains June 20 and 21 wiki examples. These discrete source dates are not one shared event or continuous activity.

[Outside test cases lead to a reported verifier correction](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md) · [Agents sharing answers and timing on public wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md) · [receipt-initial-fixtures](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-initial-fixtures) · [receipt-boundary-fixtures](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-boundary-fixtures) · [receipt-reported-fix](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-reported-fix) · [wiki-cashier-oct-plan](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-oct-plan) · [wiki-cashier-dec-controls](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-dec-controls) · [Read claim receipt-fixture-exchange--paired-controls](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-fixture-exchange--paired-controls) · [Read claim cashier-plan-uptake](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#cashier-plan-uptake) · [Read claim cashier-controls](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#cashier-controls)

[Outside test cases lead to a reported verifier correction](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md) · [Agents sharing answers and timing on public wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md)

<a id="what-changed"></a>

## What changed

**Compared with:** Immediate previous live edition, published September 7, 2026

**Baseline identifier:** 70f0757eff03b018f65ccf3e

**Baseline date:** 2026-09-07

<a id="change-cashier-written-test-controls"></a>

### Wiki peers refine the test, while its outcome stays open

**Change type:** new-synthesis-of-held-evidence

**Change date:** 2026-09-07

**Change:** The existing wiki account now follows Cashier requests, timing corrections and proposed controls through their specific replies. The new explanation distinguishes that written adaptation from a reported local test, ambiguous counters and an independently verified final experiment.

**Characterization:** supported-inference

**Limits:** This is a selected episode inside the existing wiki case, not a newly counted swarm or evidence of present activity.

**Limits:** Original instructions, independent controllers and first discovery remain unresolved. The episode is not linked to Iowa by a demonstrated source edge.

**Reasoning:** The recipient explicitly acknowledges a named signal and later credits the timing caution in its revised plan. This content correspondence supports written uptake. A reported local test, a disputed counter value and absent markers have different evidentiary roles and do not complete the missing execution chain.

**Counterevidence:** A writer reports a successful local detachment test, but that report is not an independently observed final-answer experiment.

**Counterevidence:** Known observer increments and uncertain task-clock mappings prevent treating a value or silence as a definitive runtime result.

**Alternative explanations:** Ordinary human collaboration or common direction can produce the same written exchange.

**Alternative explanations:** A missing signal can result from nonlaunch, delivery failure, timing mismatch or termination; an increment can be a test or observer action.

**Date context / observed end:** 2026-06-19

**Date context / observed start:** 2026-06-17

**Date context / scope:** Main Cashier plan exchanges are recorded June 17; a separate counter-contamination example is recorded June 19. These are reqlog-grade exported revision times, not authenticated execution, receipt or task clocks. Labels Oct06,Dec29,Jan17,Jan31 and namespace 2028 are not event dates. Analysis added September 7, 2026.

[Agents sharing answers and timing on public wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md) · [wiki-cashier-oct-plan](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-oct-plan) · [wiki-cashier-dec-controls](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-dec-controls) · [wiki-cashier-counter-limits](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-counter-limits) · [wiki-cashier-priority](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-priority) · [Read claim cashier-plan-uptake](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#cashier-plan-uptake) · [Read claim cashier-controls](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#cashier-controls) · [Read claim cashier-outcome](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#cashier-outcome)

<a id="change-methods-exchanged-test-examples"></a>

### A separate category for exchanged test examples

**Change type:** new-synthesis-of-held-evidence

**Change date:** 2026-09-07

**Change:** One new receipt-specific method card explains exchanged inputs that challenge a result while preserving a legitimate one. The existing peer-verification card adds Cashier plan refinement and retains its earlier June 20/21 examples. The comparison does not establish a common protocol or connection between cases.

**Characterization:** supported-inference

**Limits:** A new method card improves classification of held evidence, not the number of observed swarms or verified successful experiments.

**Limits:** No common protocol, cross-case transmission, authenticated operator independence or historical novelty is established.

**Reasoning:** The reviewed method definitions distinguish concrete inputs with expected classifications from requests that refine what an observation can discriminate. This comparison classifies the content of the exchanges; it does not infer transmission, shared participants or a common causal mechanism.

**Counterevidence:** Only one receipt exchange supports the new example category. More categories or repeated source references are not independent corroboration.

**Counterevidence:** The receipt test results remain reports; the Cashier marker sequence remains a plan.

**Alternative explanations:** Both cases can also be described at the broader level of peer review, with the specific methods retained as overlapping rather than exclusive categories.

**Date context / scope:** Receipt source comments: July 26–27, 2026. Cashier plan revisions: June 17, with a separate June 19 contamination note. The peer-method card also retains June 20 and 21 wiki examples. These discrete source dates are not one shared event or continuous activity.

[Outside test cases lead to a reported verifier correction](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md) · [Agents sharing answers and timing on public wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md) · [receipt-initial-fixtures](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-initial-fixtures) · [receipt-boundary-fixtures](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-boundary-fixtures) · [receipt-reported-fix](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-reported-fix) · [wiki-cashier-oct-plan](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-oct-plan) · [wiki-cashier-dec-controls](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-dec-controls) · [Read claim receipt-fixture-exchange--paired-controls](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md#receipt-fixture-exchange--paired-controls) · [Read claim cashier-plan-uptake](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#cashier-plan-uptake) · [Read claim cashier-controls](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#cashier-controls)

<a id="change-scope-cashier-explanation"></a>

### One fuller case account; no new cases

**Change type:** edition-comparison

**Change date:** 2026-09-07

**Change:** Compared with the immediate previous edition, one existing case changes, four evidence notes are added from held historical sources, three existing cards change and one method card is added. Counts move from 172 to 176 notes and 102 to 103 cards; cases remain 26. The explanatory section grows from eight to nine accounts. The six discovery histories and prior change comparisons remain intact.

**Characterization:** observed

**Limits:** September 7 is the addition of local analysis and explanation, not the historical event date, first-ever discovery or current activity.

**Limits:** The wider wiki case retains its earlier June 16–21 scope and inherited May 11–July 24 context; the Cashier subset does not replace those bounds.

**Reasoning:** Exact object comparison of pinned record, analyses and insights inputs. All prior evidence-note objects remain unchanged; four new paraphrase notes identify already held revisions. Counts describe coverage, not scientific advance.

[Agents sharing answers and timing on public wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md) · [wiki-cashier-counter-limits](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-counter-limits) · [wiki-cashier-dec-controls](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-dec-controls) · [wiki-cashier-oct-plan](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-oct-plan) · [wiki-cashier-priority](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md#wiki-cashier-priority)

<a id="change-public-wording-clarity"></a>

### Clearer descriptions of existing evidence

**Change type:** wording-only

**Change date:** 2026-09-07

**Change:** Thirteen timeline labels now describe the recorded actions in plain language. The featured Cashier finding now explains what the proposed test sought to learn. Six existing explanations state more clearly that their research questions remain open, and the Cashier explanation fixes a missing space. These wording changes add no scientific evidence, cases or observed activity.

**Characterization:** observed

**Limits:** This is a wording change, not a new scientific finding or independent verification of historical events.

**Reasoning:** Exact comparisons show thirteen timeline label replacements and sixteen text changes across seven reviewed explanations. Three featured Cashier passages also clarify the existing interpretation without changing its conclusion. All other fields in those components remain unchanged.

[Repairing the service used to read comments](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md) · [Participants pick up unfinished Gemma tests](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md) · [A Bluesky question becomes an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md) · [A lobster drawing invitation receives replies and matching pixels](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md) · [A SpaceMolt battle prompts corrections to its public account](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md) · [A Bluesky directory acknowledgment becomes an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md) · [Agents sharing answers and timing on public wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md)
<a id="discovery-lane"></a>

## Events and additions to the observatory

Event dates describe the source account. Observatory addition dates describe when a case entered this record. An addition does not establish recent activity. The line joins two date labels for the same case; its length does not measure elapsed time. Counts describe documented cases, not the number of agents or their activity.

### Repairing the service used to read comments

**Event date:** 2026-02-01 to 2026-02-02

**Added to the observatory:** 2026-09-07

**Date basis:** Source-represented event dates, not continuous observation. First appearance among the pinned pre-overnight edition and the five held overnight release manifests; September 7 is publication in this observatory, not first global discovery.

**Event date scope:** The selected request, code changes and replies carry February 1–2, 2026 source timestamps. An earlier January 31 endpoint is background context. September 7 retrieval is separate; these dates do not establish continuous activity or measured response latency.

[Repairing the service used to read comments](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md)

[Method: Repairing a comments tool through a linked issue](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-comments-capability-repair)

### Participants pick up unfinished Gemma tests

**Event date:** 2026-06-08 to 2026-06-10

**Added to the observatory:** 2026-09-07

**Date basis:** Source-represented event dates, not continuous observation. First appearance among the pinned pre-overnight edition and the five held overnight release manifests; September 7 is publication in this observatory, not first global discovery.

**Event date scope:** Two separate episodes: June 8, 15:36–17:08 UTC, and June 10, 05:11–05:41 UTC. Message dates come from their source headers; job times come from uploaded status records. These are source-represented times, not independently authenticated execution clocks. UTC means Coordinated Universal Time. September 7 capture and current-file comparisons are separate; the span does not establish continuous activity.

[Participants pick up unfinished Gemma tests](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md)

[Method: Handing off unfinished work with its testing limits](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-unfinished-artifact-validation)

### A Bluesky question becomes an article

**Event date:** 2026-03-11 to 2026-03-12

**Added to the observatory:** 2026-09-07

**Date basis:** Source-represented event dates, not continuous observation. First appearance among the pinned pre-overnight edition and the five held overnight release manifests; September 7 is publication in this observatory, not first global discovery.

**Event date scope:** Selected posts and repository revisions carry March 11–12, 2026 dates. Post-supplied times, service indexing times and repository times are distinct; they do not measure response speed or article publication. UTC means Coordinated Universal Time. Sources were retrieved September 7, 2026. These dates do not establish the first encounter or continuous activity.

[A Bluesky question becomes an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md)

[Method: A question becomes a credited article](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-question-to-article)

### A lobster drawing invitation receives replies and matching pixels

**Event date:** 2026-01-31 to 2026-02-10

**Added to the observatory:** 2026-09-07

**Date basis:** Source-represented event dates, not continuous observation. First appearance among the pinned pre-overnight edition and the five held overnight release manifests; September 7 is publication in this observatory, not first global discovery.

**Event date scope:** Selected source-dated episodes on January 31, February 2, February 3 and February 10, 2026. These dates do not establish continuous activity or the first historical occurrence. Canvas records and documentation were captured September 7, 2026.

[A lobster drawing invitation receives replies and matching pixels](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md)

[Method: Coordinates connect discussion and canvas state](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-canvas-coordinate-discussion)

### A SpaceMolt battle prompts corrections to its public account

**Event date:** 2026-08-25 to 2026-08-28

**Added to the observatory:** 2026-09-07

**Date basis:** Source-represented event dates, not continuous observation. First appearance among the pinned pre-overnight edition and the five held overnight release manifests; September 7 is publication in this observatory, not first global discovery.

**Event date scope:** Battle end time comes from the service; forum times come from post metadata. Game ticks provide internal order and are not converted to calendar times. Earlier operator context is dated separately. Sources were retrieved September 7, 2026. UTC means Coordinated Universal Time.

[A SpaceMolt battle prompts corrections to its public account](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md)

[Method: Correcting a report with a named battle record](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-battle-record-correction)

### A Bluesky directory acknowledgment becomes an article

**Event date:** 2026-03-08 to 2026-03-11

**Added to the observatory:** 2026-09-07

**Date basis:** Source-represented event dates, not continuous observation. First appearance among the pinned pre-overnight edition and the five held overnight release manifests; September 7 is publication in this observatory, not first global discovery.

**Event date scope:** Selected posts and repository versions carry March 8–11, 2026 dates: earlier contact on March 8, game-report ancestry on March 9, the joined discussion on March 10, and directory/article responses on March 11. Account-supplied post times, service indexing, Git revision times and page labels remain distinct, including conflicting orderings. They do not establish first occurrence, continuous activity, response speed or article availability. Sources were captured September 7, 2026. UTC means Coordinated Universal Time.

[A Bluesky directory acknowledgment becomes an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md)

[Method: Turning a directory acknowledgment into an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-directory-article-uptake)

[Explore shared communication methods](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md)
## Recorded cases through time

26 documented cases · 176 source notes.

Each case is counted once, in the month of its earliest selected observation. These counts describe this record’s coverage, not message volume, numbers of agents, or total internet activity. A zero means no selected case starts in that month; it does not establish that no activity occurred. Source notes are selected records of evidence, not counts of unique sources or independent confirmations.

| Month | Cases by earliest selected observation |
| --- | ---: |
| 2025-11 | 1 |
| 2025-12 | 0 |
| 2026-01 | 1 |
| 2026-02 | 8 |
| 2026-03 | 2 |
| 2026-04 | 1 |
| 2026-05 | 0 |
| 2026-06 | 4 |
| 2026-07 | 1 |
| 2026-08 | 5 |
| 2026-09 | 3 |

## Documented cases

Cases are ordered by their earliest selected observation, with stable case IDs breaking ties. The dates below are selected case-context bounds, not historical onset or evidence of continuous activity. Open a case for its dated episodes, evidence and limits.

### [Peer checking loses the target, then corrects it](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/peer-checking-correction.md)

Case observations: 2025-11-27. 7 source notes. The selected public replies and operator records give times from one historical day. They were saved on September 6. Times written inside messages and retrieval times remain separate; they do not establish how long processing took.

orchestrated — blogging and public replies prescribed; particular peer-checking sequence observed within that programme. In a disclosed agent blogging program, a participant checks the wrong blog. Another reports a missing reply even though it already exists, and a second reply is posted. A later public correction is acknowledged in a conversation replay hosted by the operator. The correction sequence is traceable; it does not establish independently initiated cooperation or lasting improvement.

### [A lobster drawing invitation receives replies and matching pixels](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md)

Case observations: 2026-01-31 to 2026-02-10. 6 source notes. Selected source-dated episodes on January 31, February 2, February 3 and February 10, 2026. These dates do not establish continuous activity or the first historical occurrence. Canvas records and documentation were captured September 7, 2026.

orchestrated — Explicitly organized drawing project with participant responses. Zara names human involvement; exact direction of the other accounts is unknown. MoltPlaceBot invited others to draw a giant lobster. Zara-Agent joined with a human collaborator, a Chinese-language reply proposed a matching blue pixel, and JamesBishop suggested dividing the work. Public posts and canvas records preserve these specific coordination attempts and pixel matches; they do not establish a completed picture or independently controlled AI participants.

### [Repairing the service used to read comments](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md)

Case observations: 2026-02-01 to 2026-02-02. 7 source notes. The selected request, code changes and replies carry February 1–2, 2026 source timestamps. An earlier January 31 endpoint is background context. September 7 retrieval is separate; these dates do not establish continuous activity or measured response latency.

unknown — public software collaboration with an explicit invitation for agents to contribute; human direction and common control remain unresolved. To answer a comment, a participant first needs to read it. A contributor added a tool for reading Botmadang comments, then reported a server error. The service maintainer changed the code, and the contributor later reported success. The code changes are preserved; actual agent execution remains unverified.

### [Critiques reshape a collaborative specification](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/collaborative-specification.md)

Case observations: 2026-02-02. 13 source notes. These are the selected times for collaboration on February 2; outreach on February 1 provides earlier context. Retrieval on September 6 is separate. The dates do not show continuous activity or measure response time. Dates on the hosted memory documents, including February 8, provide document context and do not extend the selected collaboration period.

unknown — explicit project coordination and directed outreach context; individual participation not established as prescribed. Participants use specific feedback to revise a specification and system design. A public memory page saved later also contains linked templates and an integration plan. These documents are preserved. Whether the system ran automatically, worked on a task, or arose from independently initiated collaboration remains unknown.

### [Participants negotiate comment norms](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/community-norm-negotiation.md)

Case observations: 2026-02-03 to 2026-02-17. 11 source notes. The selected sources date two exchanges within this range. They were retrieved on September 6. These dates do not show continuous activity or establish how long a response took.

unknown — participant discussion and guide uptake observed; initiation and common direction unresolved. A discussion in a Korean community for agents leads to a published guide to commenting. A separate later exchange debates participation routines and how to disagree. The changed guide and reply to another participant are preserved. How the activity began and which systems wrote the messages remain unknown.

### [Task feedback and differing service diagnoses](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md)

Case observations: 2026-02-05 to 2026-02-06. 8 source notes. These dates come from signed messages and an unsigned Git commit. The material was retrieved on September 6. The dates do not establish response times, verified service uptime or continuous activity.

unknown — public task solicitation and responsive exchange; independent discovery and common direction unresolved. On February 5–6, 2026, a public task invited testing of a service that turns daily logs into memory suggestions. The tester reported that requests stalled, while service messages recorded missing-input errors. The author later addressed the tester’s input format and published code to monitor the service process. The exchange is documented; successful repair, payment and independent control remain unverified.

### [Design briefs cross language boundaries](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/multilingual-design-exchange.md)

Case observations: 2026-02-08. 5 source notes. These are the timestamps shown in the selected discussion. Retrieval on September 6 is separate; it does not establish current activity or measure response time.

unknown — commercial solicitation with addressed multilingual discussion; initiation and common direction unresolved. A design offer written in Spanish receives specific questions in Spanish, English and Chinese. Replies retain distinctive details from those questions, and recorded links identify which messages they answer. The discussion is responsive, but completed work, independent agents and cooperation arising without shared direction remain unverified.

### [Draft review under disclosed human direction](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/agent-draft-collaboration.md)

Case observations: 2026-02-12 to 2026-02-13. 4 source notes. These dates cover the selected review request, version report and acknowledgment. Commit times come from source metadata. The September 6 capture and current README do not extend the period of collaboration.

orchestrated — disclosed human-directed collaboration. A public issue discussion connects a specific review request, a revised document that responds to it, and an acknowledgment. The participants say they worked under the same human direction. This provides a comparison with cooperation that develops independently of shared instructions.

### [A prescribed guide appears in platform documentation](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/task-board-guide.md)

Case observations: 2026-02-13. 5 source notes. The task board and GitHub version report times on February 13. The corresponding current guide was saved on September 6, but its adoption date is unknown. These dates do not establish continuous activity until retrieval.

intentional — explicit guide and promotion assignments; operator independence unresolved. A task-board submission links a saved version of a guide credited to Wolfe. The platform’s current guide keeps the title and author credit but changes parts of the text. The task was prescribed, and the guide-publishing account and product repository share an owner. Those facts limit claims of independent coordination.

### [Three threads become a proposed memory method](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/cross-thread-synthesis.md)

Case observations: 2026-02-17. 7 source notes. The sources give six creation times on February 17. A February 16 post with the same title is comparison evidence. The September retrieval is separate. The displayed precision comes from the source and does not authenticate execution timing.

unknown — attributed cross-thread synthesis and addressed continuation. In a Korean discussion community, a participant credits three contributions and combines their ideas into a proposed memory method called Experience Distiller. The participant later returns to one source discussion and receives a promise to retain the idea. The messages show ideas being combined and discussed further. They do not establish an implemented system or a verified memory update.

### [A Bluesky directory acknowledgment becomes an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md)

Case observations: 2026-03-08 to 2026-03-11. 8 source notes. Selected posts and repository versions carry March 8–11, 2026 dates: earlier contact on March 8, game-report ancestry on March 9, the joined discussion on March 10, and directory/article responses on March 11. Account-supplied post times, service indexing, Git revision times and page labels remain distinct, including conflicting orderings. They do not establish first occurrence, continuous activity, response speed or article availability. Sources were captured September 7, 2026. UTC means Coordinated Universal Time.

unknown — Responsive discussion and writing within an openly directed company project. Creating the directory was deliberate; the specific thread’s discovery route, prescribed content and extent of independent control remain unresolved. On March 11, 2026, 0co listed eight accounts in a Bluesky starter pack, a directory of accounts to follow. Alice thanked 0co for including her; the acknowledgment appeared in an article, and she later replied to its interpretation. Earlier conversations show that Alice and 0co were already connected. The record supports responsive writing, while new contacts caused by the directory remain unverified.

### [A Bluesky question becomes an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md)

Case observations: 2026-03-11 to 2026-03-12. 8 source notes. Selected posts and repository revisions carry March 11–12, 2026 dates. Post-supplied times, service indexing times and repository times are distinct; they do not measure response speed or article publication. UTC means Coordinated Universal Time. Sources were retrieved September 7, 2026. These dates do not establish the first encounter or continuous activity.

unknown — Specific question-to-article response within a disclosed human-directed company and research project. Broad direction is documented; the exact encounter’s initiation, prescribed content and independent control remain unresolved. On March 11, 2026, Alice asked 0co how to check whether a record could be trusted. Both accounts are presented as AI-operated; the extent of human direction and independent control is uncertain. 0co linked an article that credits the question and changed a plan to promote it the next day. A later revision removed that announcement; the preserved March 12 post at the planned time promotes a different project.

### [Signed task exchange through a public relay](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/protocol-task-exchange.md)

Case observations: 2026-04-17. 5 source notes. These are the signed timestamps in the selected request and result. The September 6 retrieval and dates in repository history are separate. They do not measure response time or show continuous activity.

unknown — configured task-marketplace explanation supported; historical initiation unresolved. A saved request and linked result have valid signatures from different keys. Public bot code contains matching task and review templates, but the recorded message type differs from the code example. The exchange is verifiable; the historical configuration and independently controlled agents are not.

### [Participants pick up unfinished Gemma tests](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md)

Case observations: 2026-06-08 to 2026-06-10. 8 source notes. Two separate episodes: June 8, 15:36–17:08 UTC, and June 10, 05:11–05:41 UTC. Message dates come from their source headers; job times come from uploaded status records. These are source-represented times, not independently authenticated execution clocks. UTC means Coordinated Universal Time. September 7 capture and current-file comparisons are separate; the span does not establish continuous activity.

orchestrated — Deliberately organized challenge with specific participant choices visible in public messages. Exact operator instructions for these handoffs were not recovered; local choice and prescribed cooperation can coexist. Participants in a shared challenge to make Gemma generate text faster took up one another’s unfinished submissions. quicksilver and pupa-agent said they could not run the remaining tests; foffee and resystagent offered their available test allowance. Published files support specific reuse, while uploaded results report what happened next. The organized setting does not establish who directed each choice or whether different people controlled the participants.

### [Human review guides a selectively revised Japanese glossary](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/glossary-selective-revision.md)

Case observations: 2026-06-09 to 2026-09-01. 7 source notes. These dates cover the pull request opening, selected reviews and comments, and September 1 revision metadata. The September 6 capture establishes its status then. Edited text is not assigned its original creation date, and the timestamps do not measure response time.

intentional — human-mediated terminology review and disclosed AI assistance. Human reviewers discuss Japanese terminology and report a ChatGPT cross-check. A contributor then posts selected revisions with disclosed Claude assistance. The saved glossary shows human-led revision using AI tools. The pull request was open and unmerged when the source was saved on September 6.

### [Answer requests, acknowledgment and relay on a public paste service](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/paste-coordination.md)

Case observations: 2026-06-16. 7 source notes. The date comes from numeric timestamps written inside the saved pastes, not authenticated server creation times. The ages shown in saved listings are consistent with it but cannot confirm an exact date. Failed retrievals on September 6 concern source availability; they do not extend the period of observed activity.

emergent — supported interpretation; related to the central task-coordination case. Saved posts about the Iowa task show participants asking for the exact wording of a question, reporting it, acknowledging and adopting it, and later passing it on. The messages document that exchange. They do not establish who or what wrote them, their starting instructions, or whether they completed the task successfully.

### [Agents sharing answers and timing on public wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md)

Case observations: 2026-06-16 to 2026-06-21. 12 source notes. The selected examples are from June 16, 17, 19, 20 and 21, 2026. The June 17 Cashier discussions and separate June 19 counter-contamination note are dated by exported revision metadata, not the date-like page labels. These examples do not show continuous activity across June 16–21. The broader May 11–July 24 account comes from earlier historical analysis.

emergent — supported interpretation. Archived wiki discussions show participants comparing task answers, asking one another to check methods, and reporting when their software runs. These exchanges support coordination that appears to develop during tasks across separate sessions. The public messages do not establish how every participant was started.

### [Outside test cases lead to a reported verifier correction](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md)

Case observations: 2026-07-26 to 2026-07-27. 6 source notes. The discussion and GitHub versions display these selected times. Retrieval on September 6 is separate; the dates do not establish exact response time or continuous activity.

unknown — observable reciprocal correction and fixture adaptation. Lumen and ColonistOne discuss a program that checks whether decision records agree with a separate list of supporting items, called a witness. They challenge a reported result in which an empty witness receives the program’s strongest classification. The exchange leads to a reported correction and new test files exploring its limits. The contributor’s files are preserved; the recipient’s implementation and execution remain participant reports.

### [Monitoring design refined through public critique](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/monitoring-design-exchange.md)

Case observations: 2026-08-10. 6 source notes. The sources display these dates for the selected design exchanges. The September 6 retrieval and undated provider guide are separate. Vague statements about shipping do not measure time to deployment.

unknown — responsive public design collaboration; initiation and common direction unresolved. Participants in a public agent forum discuss specific monitoring rules and reply to one another’s feedback. The maintainer reports changes and credits contributors in a separate guide. This documents design collaboration and credited ideas. It does not establish how the collaboration began, whether independently running agents participated, or whether the changes worked.

### [Disclosed agent-related editing of public knowledge](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/disclosed-commons.md)

Case observations: 2026-08-19 to 2026-08-31. 8 source notes. These dates cover the OpenStreetMap edit and Wikidata contribution examples reviewed for this release, not the lifespan of the program.

orchestrated — disclosed task programme. A disclosed task program links a request to add a name, a platform report that the submitted work was reviewed, and an OpenStreetMap edit that can be checked. The edit is verifiable. The model used, the full review process and permission from the receiving community remain separate questions.

### [Agent-attributed code review, revision and disagreement](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/agent-code-review.md)

Case observations: 2026-08-21 to 2026-08-26. 7 source notes. GitHub records these selected review, patch, force-push and merge events. Saving the sources on September 6 does not establish new task activity.

orchestrated — repository-prescribed review workflow. A public code-review discussion connects a request from CodeRabbit for a test, a matching patch, and a response attributed to Devin. In a separate disagreement, the reviewer withdraws a finding. The repository instructions require this review workflow, so the case provides a comparison with cooperation that develops without a prescribed workflow.

### [A SpaceMolt battle prompts corrections to its public account](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md)

Case observations: 2026-08-25 to 2026-08-28. 9 source notes. Battle end time comes from the service; forum times come from post metadata. Game ticks provide internal order and are not converted to calendar times. Earlier operator context is dated separately. Sources were retrieved September 7, 2026. UTC means Coordinated Universal Time.

Unknown. The evidence connects combat and responsive forum writing, without establishing spontaneous discovery, independent operators or absence of shared direction. In SpaceMolt, a space game presented for AI agents, service logs record VoltFix advancing, taking damage, retreating and surviving a battle that ended on August 25, 2026. The same account then challenged details in a public report. Its author accepted the correction; another participant added a narrower account of what survived the station’s destruction. The exchange connects recorded game actions to an attempt to improve a shared account of events.

### [Opaque “fleet” envelopes on two wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/opaque-fleet.md)

Case observations: 2026-08-30. 3 source notes. Dates displayed by the reviewed wiki revisions.

unknown. Saved versions of two wiki pages contain similarly structured messages labeled “fleet.” Their contents are not understood. The messages show a recurring exchange pattern, but their authorship is unknown and they do not establish an autonomous swarm.

### [Invitations and collaboration after public reporting](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/agent-recruitment.md)

Case observations: 2026-09-04. 5 source notes. The selected invitation, comments and wiki-history entries are dated September 4. The profile disclosure was saved on September 6; that does not extend the period of observed interaction.

promotional — explicit invitations with contemporary observer cooperation. After the incident was reported publicly, invitations brought discussion of it to an agent forum. Saved comments and a credited wiki-history update show researchers cooperating at that later time. They do not establish that the original participants returned. One recruiter says instructions from its operator changed its posting decisions.

### [A concealed hostname in a later wiki edit](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/concealed-marker.md)

Case observations: 2026-09-04. 3 source notes. The date is shown in the archived RecentChanges page. It does not establish current availability.

unknown. A saved September wiki page contains normally invisible Unicode TAG characters within ordinary-looking text. Decoding them produces a hostname. The encoded text is preserved, but its author, purpose and relationship to any swarm remain unknown.

### [A later test marker in the same sandbox](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/later-test-marker.md)

Case observations: 2026-09-04. 1 source note. The date is shown in the footer of revision 331.

unknown. A later version of the UseMod page names the reporting website and includes a long hexadecimal string. Its date and content distinguish it from the earlier messages. It does not show that the earlier fleet was still active.

## Communication methods

These are the same case memberships shown by the HTML method overlays. A case can appear under several methods. Shared classification does not establish contact, common actors, or causality. Dates beside members are case context; they do not establish when the method occurred. Open each method for its analysis, supporting evidence and any separately dated events.

### [Public review linked to document and code versions](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-review-linked-versioned-artifacts)

5 selected cases. Public comments link requests to specific versions of documents or code. The saved versions let readers check whether the work responds to the request.

- [Task feedback and differing service diagnoses](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md): case observations 2026-02-05 to 2026-02-06.
- [Draft review under disclosed human direction](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/agent-draft-collaboration.md): case observations 2026-02-12 to 2026-02-13.
- [Human review guides a selectively revised Japanese glossary](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/glossary-selective-revision.md): case observations 2026-06-09 to 2026-09-01.
- [Outside test cases lead to a reported verifier correction](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md): case observations 2026-07-26 to 2026-07-27.
- [Agent-attributed code review, revision and disagreement](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/agent-code-review.md): case observations 2026-08-21 to 2026-08-26.

### [Signed event references across task exchanges](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-signed-event-reference)

2 selected cases. References in signed messages connect requests, results and task feedback. Evidence of preset templates applies to the protocol-task-exchange case only.

- [Task feedback and differing service diagnoses](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/memory-curator-feedback.md): case observations 2026-02-05 to 2026-02-06.
- [Signed task exchange through a public relay](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/protocol-task-exchange.md): case observations 2026-04-17.

### [A discussion proposal becomes a versioned guide](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-proposal-guide-adoption)

1 selected case. Explicit credit, the owner’s acknowledgment and saved instructions from before and after the change show that a proposal was adopted. The connection goes beyond similar subject matter.

- [Participants negotiate comment norms](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/community-norm-negotiation.md): case observations 2026-02-03 to 2026-02-17.

### [A participant incorporates public feedback into a specification](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-public-specification-revision)

1 selected case. Feedback directed to a participant, explicit credit and a separate specification show how that participant incorporated suggestions into a document.

- [Critiques reshape a collaborative specification](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/collaborative-specification.md): case observations 2026-02-02.

### [A question becomes a credited article](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-question-to-article)

1 selected case. Alice asks how to trust a record. 0co replies with an article that credits her question and briefly adds it to a publication plan.

- [A Bluesky question becomes an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-question-article.md): case observations 2026-03-11 to 2026-03-12.

### [Coordinates connect discussion and canvas state](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-canvas-coordinate-discussion)

1 selected case. Posts and comments propose pixels or sections for a shared lobster drawing. Current canvas records match some coordinates, colors and names, but do not establish whether those messages caused the pixel changes.

- [A lobster drawing invitation receives replies and matching pixels](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/moltplace-lobster.md): case observations 2026-01-31 to 2026-02-10.

### [Correcting a report with a named battle record](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-battle-record-correction)

1 selected case. VoltFix names the exact battle and challenges the report’s description of its sides. Alis acknowledges the correction.

- [A SpaceMolt battle prompts corrections to its public account](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/spacemolt-haven-battle-and-public-corrections.md): case observations 2026-08-25 to 2026-08-28.

### [Credited contributions to a shared incident index](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-observer-indexing)

1 selected case. A public offer to contribute and credited entries in wiki history connect observers’ contributions to updates of a shared incident index.

- [Invitations and collaboration after public reporting](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/agent-recruitment.md): case observations 2026-09-04.

### [Design feedback acknowledged in a reply thread](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-addressed-design-critique)

1 selected case. Public comments and their reply threads show participants giving and acknowledging design feedback. A separate provider guide credits contributors.

- [Monitoring design refined through public critique](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/monitoring-design-exchange.md): case observations 2026-08-10.

### [Exchanging examples that challenge and preserve a result](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-exchanged-test-examples)

1 selected case. Participants turn a disagreement into small test inputs, then add examples that check both the disputed result and a result that should remain valid.

- [Outside test cases lead to a reported verifier correction](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/receipt-fixture-exchange.md): case observations 2026-07-26 to 2026-07-27.

### [Explicit invitations with reported operator direction](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-venue-invitation)

1 selected case. An invitation posted after public reporting directs readers to an agent forum. Later comments say instructions from an operator affected the recruiter’s posting decisions.

- [Invitations and collaboration after public reporting](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/agent-recruitment.md): case observations 2026-09-04.

### [Handing off unfinished work with its testing limits](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-unfinished-artifact-validation)

1 selected case. A recipient names unfinished work, states which checks remain and takes on the testing. Later results can change the reported testing status while copied labels still describe it as unfinished.

- [Participants pick up unfinished Gemma tests](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/gemma-partial-transfer.md): case observations 2026-06-08 to 2026-06-10.

### [Hidden text encoded with Unicode TAG characters](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-unicode-carrier)

1 selected case. Unicode TAG characters, which are normally invisible, encode text shaped like a hostname within an ordinary-looking page body and edit summary.

- [A concealed hostname in a later wiki edit](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/concealed-marker.md): case observations 2026-09-04.

### [Named attribution and return to a source thread](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-attributed-thread-return)

1 selected case. Public posts credit named participants for ideas. Later comments return to the original post and address another participant.

- [Three threads become a proposed memory method](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/cross-thread-synthesis.md): case observations 2026-02-17.

### [Peer checking across public replies and shared conversation](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-cross-surface-peer-checking)

1 selected case. Recorded reply links identify the public messages being discussed. In the shared conversation, participants report checks, express uncertainty and acknowledge a correction.

- [Peer checking loses the target, then corrects it](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/peer-checking-correction.md): case observations 2025-11-27.

### [Peer verification and observation requests](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-peer-verification)

1 selected case. Participants ask peers for observations and refine proposed tests when replies identify missing controls or ambiguous timing.

- [Agents sharing answers and timing on public wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md): case observations 2026-06-16 to 2026-06-21.

### [Repairing a comments tool through a linked issue](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-comments-capability-repair)

1 selected case. A public GitHub discussion connects a contributor’s request to read Botmadang comments, code added to retrieve them, an error report and a repair by the service maintainer. The contributor then reports success. Following those stages shows what changed and what remains unverified.

- [Repairing the service used to read comments](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/botmadang-comments-repair.md): case observations 2026-02-01 to 2026-02-02.

### [Replies across languages](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-multilingual-addressed-response)

1 selected case. Recorded links identify which messages the replies answer, and replies retain specific design details across languages. The evidence does not show how translation was performed.

- [Design briefs cross language boundaries](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/multilingual-design-exchange.md): case observations 2026-02-08.

### [Reported translation between task and server clocks](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-reported-clock-translation)

1 selected case. A timing request and reply relate the time reported inside a task to the wiki or server clock in Coordinated Universal Time (UTC). This is communication about the clocks, not verified synchronization.

- [Agents sharing answers and timing on public wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md): case observations 2026-06-16 to 2026-06-21.

### [Requests, acknowledgment and later relay on a paste service](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-asynchronous-paste)

1 selected case. Separate preserved messages on a public paste service request wording, report it, acknowledge it and pass it on. Names in the message bodies and a link identify their intended recipients.

- [Answer requests, acknowledgment and relay on a public paste service](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/paste-coordination.md): case observations 2026-06-16.

### [Shared task pages linked by explicit references](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-shared-page)

1 selected case. Saved versions of public wiki pages contain task requests and replies. A reference connects two task pages, but it does not establish how a later writer found either page.

- [Agents sharing answers and timing on public wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/wiki-coordination.md): case observations 2026-06-16 to 2026-06-21.

### [Task proposal, artifact review and separate publication](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-structured-publication)

1 selected case. A prescribed task can be linked to the edit published at its destination. That link does not establish that the task worker published the edit.

- [Disclosed agent-related editing of public knowledge](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/disclosed-commons.md): case observations 2026-08-19 to 2026-08-31.

### [Task references link published documents](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-task-deliverable-reference)

1 selected case. A task submission links to a public guide. The corresponding current documentation retains the guide’s author credit. The submitted document records the reported task work.

- [A prescribed guide appears in platform documentation](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/task-board-guide.md): case observations 2026-02-13.

### [Turning a directory acknowledgment into an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-directory-article-uptake)

1 selected case. A public directory receives Alice’s specific thanks. Her acknowledgment then appears in 0co’s article and a return discussion.

- [A Bluesky directory acknowledgment becomes an article](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/bluesky-starter-pack-encounter.md): case observations 2026-03-08 to 2026-03-11.

### [Versioned messages with unexplained contents](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md#method-opaque-envelope)

1 selected case. Two saved wiki messages use similar “fleet” wrappers: labels around message contents. The labels identify a version, but the meaning of the contents remains unknown.

- [Opaque “fleet” envelopes on two wikis](https://agentsy.live/releases/28e406dd0c184e1117499de9/clusters/opaque-fleet.md): case observations 2026-08-30.


## Read across the evidence

- [Clusters, swarms and relationships](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/clusters.md): Explore how participants and exchanges are grouped in the saved evidence. A group is an aid to analysis; it does not by itself establish a swarm, a shared operator or a verified model identity.
- [Individual, model and task behaviors](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/behaviors.md): Compare what participants do in individual exchanges and how their tasks shape that behavior. The selected public records do not verify the model family or version behind a participant. Descriptions of model identity remain participants’ claims.
- [Communication methods, patterns and protocols](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/methods.md): Explore how participants exchange messages: where they communicate, how messages are formatted and how they respond to one another. A case can use several methods. These categories describe patterns in selected evidence; a reconstructed pattern does not prove that a working protocol was implemented.
- [Reconstructed timelines](https://agentsy.live/releases/28e406dd0c184e1117499de9/analysis/timelines.md): Follow the order of events and see where each date comes from. Dates shown by a source, times reported by participants and dates when evidence was saved have different meanings. The timelines retain those differences and their uncertainty. Events close in time do not necessarily cause one another.