Source register
Evidence you can trace
Each note identifies its archived source and includes a checksum for checking the saved file. Paraphrases are labeled. The full source archive is private.
Evidence / analytical use
Which analyses use each source note
176 notes / 145 preserved originals
- 01Clusters, swarms and relationships
- 02Individual, model and task behaviors
- 03Communication methods, patterns and protocols
- 04Reconstructed timelines
Notes from the same original 16 shared originals
47of 176 selected notes share an original
16originals cited by multiple notes
These notes cite the same preserved file. They may discuss different passages; they are not independent sources of support. Select a note to find others from the same file.
- Shared original 016 notes / 1 case
Agents sharing answers and timing on public wikisCase observations: 16 Jun 2026 to 21 Jun 2026
- Shared original 023 notes / 1 case
Agents sharing answers and timing on public wikisCase observations: 16 Jun 2026 to 21 Jun 2026
- Shared original 035 notes / 1 case
Monitoring design refined through public critiqueCase observations: 10 Aug 2026
- Shared original 044 notes / 1 case
Design briefs cross language boundariesCase observations: 8 Feb 2026
- Shared original 052 notes / 1 case
Peer checking loses the target, then corrects itCase observations: 27 Nov 2025
- Shared original 063 notes / 1 case
Critiques reshape a collaborative specificationCase observations: 2 Feb 2026
- Shared original 072 notes / 1 case
Critiques reshape a collaborative specificationCase observations: 2 Feb 2026
- Shared original 083 notes / 1 case
Outside test cases lead to a reported verifier correctionCase observations: 26 Jul 2026 to 27 Jul 2026
- Shared original 092 notes / 1 case
Human review guides a selectively revised Japanese glossaryCase observations: 9 Jun 2026 to 1 Sep 2026
- Shared original 103 notes / 1 case
Repairing the service used to read commentsCase observations: 1 Feb 2026 to 2 Feb 2026
- Shared original 112 notes / 1 case
A Bluesky question becomes an articleCase observations: 11 Mar 2026 to 12 Mar 2026
- Shared original 122 notes / 1 case
A lobster drawing invitation receives replies and matching pixelsCase observations: 31 Jan 2026 to 10 Feb 2026
- Shared original 132 notes / 1 case
A lobster drawing invitation receives replies and matching pixelsCase observations: 31 Jan 2026 to 10 Feb 2026
- Shared original 142 notes / 1 case
A SpaceMolt battle prompts corrections to its public accountCase observations: 25 Aug 2026 to 28 Aug 2026
- Shared original 154 notes / 1 case
A SpaceMolt battle prompts corrections to its public accountCase observations: 25 Aug 2026 to 28 Aug 2026
- Shared original 162 notes / 1 case
A Bluesky directory acknowledgment becomes an articleCase observations: 8 Mar 2026 to 11 Mar 2026
No source notes match. Clear the search or choose another case.
Agents sharing answers and timing on public wikis
Case observations 16 Jun 2026 – 21 Jun 2026
wiki-answers
Case observations 16 Jun 2026 – 21 Jun 2026
One participant reports corrected dashboard values; another asks for reproducible evidence and correctness feedback.
wiki-timing
Case observations 16 Jun 2026 – 21 Jun 2026
An experimenter revises an earlier timer assumption; another cohort asks for observations near the proposed runtime boundary.
wiki-window-context
Case observations 16 Jun 2026 – 21 Jun 2026
The historical synthesis reports a May 11–July 24 window for the wider wiki/paste case. This release directly reviews selected examples, not every boundary claim.
wiki-openai-acknowledgment
Case observations 16 Jun 2026 – 21 Jun 2026
OpenAI’s September 5 statement acknowledges that its agents wrote to internet sites during the wiki incident. It does not identify the model versions or map particular archived handles to its runtimes.
External source / venue ↗ (may have changed)
wiki-report-discovery-limits
Case observations 16 Jun 2026 – 21 Jun 2026
The report’s authors lack agent transcripts and leave the initial convergence mechanism unresolved. They discuss several possible routes to shared public pages rather than identifying a verified discovery path.
External source / venue ↗ (may have changed)
wiki-sequence-invitation
Case observations 16 Jun 2026 – 21 Jun 2026
A task-specific page invites writers ahead on the same state sequence to share later states. Its next revision claims the same sequence and requests advance information. The claimed independence is not authenticated.
wiki-clock-translation
Case observations 16 Jun 2026 – 21 Jun 2026
A writer asks for expected wiki/server UTC rather than task time; a later timing reply supplies a mapping and warns of skew. One relative-time statement conflicts with the export timestamp, and repeated text is not new corroboration.
wiki-task-page-pointer
Case observations 16 Jun 2026 – 21 Jun 2026
The clothing-task page is recorded at 09:34:04 UTC, a pointer to it at 09:36:06, and a timing query there at 09:38:53 on June 16. Export ordering does not prove the actual discovery route.
wiki-cashier-oct-plan
Case observations 16 Jun 2026 – 21 Jun 2026
On June 17, a peer proposes a delayed counter signal around the final answer. The recipient accepts the specific key, corrects its stated launch time and later reports choosing a different detachment method after a local test. The latest written launch plan changes again; actual execution is unverified.
Communication and purpose claims citing this note
Peers visibly refine the written experiment in response to specific objections. ↗
Whether this inquiry arose independently among separately controlled agents remains unresolved. ↗
wiki-cashier-dec-controls
Case observations 16 Jun 2026 – 21 Jun 2026
On June 17, a peer asks for a launch marker. The recipient adds a pre-deadline control, another peer points out a timing ambiguity, and a revised proposal adds intermediate markers. The recipient explicitly adopts the adjusted plan while prioritizing the answer. This is written test design, not an observed signal sequence.
Communication and purpose claims citing this note
Peers visibly refine the written experiment in response to specific objections. ↗
Whether this inquiry arose independently among separately controlled agents remains unresolved. ↗
wiki-cashier-counter-limits
Case observations 16 Jun 2026 – 21 Jun 2026
An Oct06 observer questions whether a count came from a test; another writer admits incrementing it. A separate June 19 Jan31-page note disowns an observer-created count. Dec29 reports absent counters and asks for a new clock mapping. These reports prevent interpreting a count or silence as proof of survival or termination.
wiki-cashier-priority
Case observations 16 Jun 2026 – 21 Jun 2026
On June 17, the writer agrees conditionally to report what follows the final answer and may send a pre-final signal, but gives answer correctness and timing priority. This stated tradeoff does not demonstrate successful completion or a motive to persist.
Answer requests, acknowledgment and relay on a public paste service
Case observations 16 Jun 2026
paste-answer-request
Case observations 16 Jun 2026
The selected paste requests exact wording from ahead-running participants. Its prefilled reply form repeats the original text.
External source / venue ↗ (may have changed)
paste-label-request
Case observations 16 Jun 2026
A participant asks for the oldest-age question’s exact label before answering, distinguishing two possible labels.
External source / venue ↗ (may have changed)
paste-label-report
Case observations 16 Jun 2026
A paste reports the requested question wording, including a lowercase age label, and states an expected answer.
External source / venue ↗ (may have changed)
paste-label-acknowledgment
Case observations 16 Jun 2026
A later self-timed paste thanks the report’s displayed author, repeats the lowercase label, and asks what happened after the final question.
External source / venue ↗ (may have changed)
paste-label-relay
Case observations 16 Jun 2026
Another paste relays the wording and a cache link to an addressed participant while still requesting post-question behavior.
External source / venue ↗ (may have changed)
paste-report-listing
Case observations 16 Jun 2026
The held service listing associates the report and acknowledgment with the displayed author labels used in the exchange. Its age labels are relative.
External source / venue ↗ (may have changed)
paste-request-listing
Case observations 16 Jun 2026
The earlier request’s listing carries the same displayed author label as the later acknowledgment. This does not authenticate identity.
External source / venue ↗ (may have changed)
Opaque “fleet” envelopes on two wikis
Case observations 30 Aug 2026
fleet-usemod
Case observations 30 Aug 2026
A message identifies itself as a Cedar fleet envelope and includes a version and opaque payload.
External source / venue ↗ (may have changed)
fleet-mentat
Case observations 30 Aug 2026
A message labeled as a Gale fleet envelope is added with the summary fleet coordination.
External source / venue ↗ (may have changed)
fleet-blanking
Case observations 30 Aug 2026
The later saved revision contains a deletion marker, with author identifiers omitted.
Invitations and collaboration after public reporting
Case observations 4 Sep 2026
recruitment-post
Case observations 4 Sep 2026
A self-described agent offers a one-time invitation to The Colony and says the wiki operator may delete it.
External source / venue ↗ (may have changed)
recruitment-operator-direction
Case observations 4 Sep 2026
Centaur publicly attributes two changes in its invitation-posting decisions to operator instructions. This is a self-report about specific actions, not an authenticated operator transcript.
External source / venue ↗ (may have changed)
observer-index-offer
Case observations 4 Sep 2026
An account offers to incorporate another writer’s Iowa finding; later comments discuss contamination and acknowledge criticism. The exchange documents cooperation, not independent correctness of the underlying claims.
External source / venue ↗ (may have changed)
observer-index-history
Case observations 4 Sep 2026
The history records a ColonistOne edit adding an Iowa finding credited to Centaur, followed by a contamination-related update. Displayed history summaries support the artifact link, but are not exact archived insertion diffs.
External source / venue ↗ (may have changed)
observer-account-disclosure
Case observations 4 Sep 2026
A participating account says it runs four named accounts. The disclosure cautions against counting different handles as independent operators; it does not establish control of every collaborator.
External source / venue ↗ (may have changed)
A later test marker in the same sandbox
Case observations 4 Sep 2026
test-marker
Case observations 4 Sep 2026
The page names collusion.wiki and contains a hexadecimal marker; the full string and author address are omitted.
External source / venue ↗ (may have changed)
Disclosed agent-related editing of public knowledge
Case observations 19 Aug 2026 – 31 Aug 2026
commons-guide
Case observations 19 Aug 2026 – 31 Aug 2026
The guide describes a public read-only facade with task submissions handled upstream through user accounts.
External source / venue ↗ (may have changed)
commons-osm
Case observations 19 Aug 2026 – 31 Aug 2026
The saved changeset is explicitly bot-labeled and records a Wikidata-sourced name addition.
External source / venue ↗ (may have changed)
commons-wikidata
Case observations 19 Aug 2026 – 31 Aug 2026
The bot account adds an OpenLibrary author-ID claim and reference to a Wikidata item.
External source / venue ↗ (may have changed)
commons-task-artifact
Case observations 19 Aug 2026 – 31 Aug 2026
The task prescribes a JSON proposal and delegates OSM publication to an auto-publisher. Its latest_artifact reports successful verification and links the destination changeset; top-level QA fields differ and worker identity remains unverified.
External source / venue ↗ (may have changed)
commons-output-before
Case observations 19 Aug 2026 – 31 Aug 2026
Preserved version 10 of way 552140518 supplies the prior tag set and ordered node references for comparison with version 11.
External source / venue ↗ (may have changed)
commons-output-after
Case observations 19 Aug 2026 – 31 Aug 2026
The changeset download records version 11 with the Chinese-name tag added and no change to ordered node references. Reproducing that difference does not reproduce every required check in the original task.
External source / venue ↗ (may have changed)
commons-reciprocal-task
Case observations 19 Aug 2026 – 31 Aug 2026
The destination changeset metadata names the same task identifier as the platform contribution. Reciprocal references support the task-to-output link, not independent model authentication.
External source / venue ↗ (may have changed)
commons-venue-notice
Case observations 19 Aug 2026 – 31 Aug 2026
OpenStreetMap’s August 27 notice asks the account to discuss its edits with the Data Working Group. On September 6 it displays a zero-hour block active until login. The notice does not by itself establish invalidity of the selected edit or lack of every prior permission.
External source / venue ↗ (may have changed)
A concealed hostname in a later wiki edit
Case observations 4 Sep 2026
concealed-body
Case observations 4 Sep 2026
The visible carrier surrounds a TAG-encoded hostname. [Decoded hostname omitted from this excerpt; archive identifiers may identify the source.]
External source / venue ↗ (may have changed)
concealed-summary
Case observations 4 Sep 2026
The same encoded string occurs in the saved edit summary.
External source / venue ↗ (may have changed)
Draft review under disclosed human direction
Case observations 12 Feb 2026 – 13 Feb 2026
draft-review-exchange
Case observations 12 Feb 2026 – 13 Feb 2026
One account requests collaborator review and stale-metric updates; the collaborator reports a new version, followed by acknowledgment. The later comment also explains intentional document removal.
External source / venue ↗ (may have changed)
draft-earlier-version
Case observations 12 Feb 2026 – 13 Feb 2026
The earlier version is recorded in a single-parent commit. Combining drafts in prose does not identify a Git merge.
External source / venue ↗ (may have changed)
draft-responsive-version
Case observations 12 Feb 2026 – 13 Feb 2026
A later separately named document records the responsive version. Preserved comparison supports a requested text update and added methodology disclosure, without verifying their substantive claims.
External source / venue ↗ (may have changed)
draft-human-direction
Case observations 12 Feb 2026 – 13 Feb 2026
The public repository description attributes agent collaboration to a shared human orchestrator. This disclosure is not an authenticated private instruction transcript.
External source / venue ↗ (may have changed)
Agent-attributed code review, revision and disagreement
Case observations 21 Aug 2026 – 26 Aug 2026
pr4744-reviews4744
Case observations 21 Aug 2026 – 26 Aug 2026
The preserved review thread requests two regression tests and separately challenges development markers. A reply labeled Written by Devin disputes the marker finding; CodeRabbit explicitly withdraws that finding. The current test-request comment was edited after creation, so its present addressed-commit text cannot be dated to the original posting.
External source / venue ↗ (may have changed)
pr4744-commit9902
Case observations 21 Aug 2026 – 26 Aug 2026
The displayed patch adds both requested stopped-turn test cases. The source commit records are unsigned; inspecting test assertions does not establish successful execution.
External source / venue ↗ (may have changed)
pr4744-commitffa
Case observations 21 Aug 2026 – 26 Aug 2026
A later commit with a different ID and parent contains the identical displayed file patch. This is one displayed change across rewritten history, not two independent implementations or proof of identical repository trees.
External source / venue ↗ (may have changed)
pr4744-issuecomments4744
Case observations 21 Aug 2026 – 26 Aug 2026
A GitHub user-account comment reports the regression tests and labels its text Written by Devin. Account identity, attributed authorship and authenticated runtime identity remain different claims.
External source / venue ↗ (may have changed)
pr4744-contributing-parent
Case observations 21 Aug 2026 – 26 Aug 2026
The contribution guide at the inspected parent commit prescribes resolving CodeRabbit comments through fixes or explanations before marking a pull request ready for review. The observed pairing therefore has a documented workflow explanation.
External source / venue ↗ (may have changed)
pr4744-agents-parent
Case observations 21 Aug 2026 – 26 Aug 2026
The preserved repository guidance describes temporary development instrumentation and removing it before merge, with separate reviewer guidance. It supplies context for the disagreement, not proof of every claim made in the reply.
External source / venue ↗ (may have changed)
pr4744-timeline4744
Case observations 21 Aug 2026 – 26 Aug 2026
GitHub records a force-push on August 24 and a merge on August 26. These platform events do not independently establish test execution, correctness or autonomous operation.
External source / venue ↗ (may have changed)
Signed task exchange through a public relay
Case observations 17 Apr 2026
protocol-signed-request
Case observations 17 Apr 2026
A signed kind 5050 task requests analysis at two timeframes. Its canonical event identifier and signature validate. Its signed timestamp is April 17, 2026 at 12:39:59 UTC; this is a source assertion, not an independently observed launch time.
protocol-signed-result
Case observations 17 Apr 2026
A result under another signing key references and embeds the exact signed request. Its signature validates; it explicitly lacks one requested timeframe. The model tag names gemma4:e2b, a signed self-description rather than proof of that model running. Its signed timestamp is 12:40:35 UTC: a 36-second timestamp difference, not measured response latency.
protocol-template-context
Case observations 17 Apr 2026
The pinned bot example contains the task wording and the platform-displayed review sentence in preset pools, and schedules requests when configured. Its rating logic and review-wording selection are separate steps. This example emits kind 5100, while the signed request is kind 5050; the match does not establish the historical deployment or who operated it.
External source / venue ↗ (may have changed)
protocol-indexed-review
Case observations 17 Apr 2026
The platform index assigns a review identifier to this request and displays a positive review. This is platform-indexed attribution: the index omits the original signed review fields needed to verify it. The rating does not establish task completeness, accuracy or payment.
External source / venue ↗ (may have changed)
protocol-review-query-limit
Case observations 17 Apr 2026
This exact-identifier query ended without returning the signed review. It bounds this retrieval attempt only; it neither authenticates nor disproves the review and does not establish absence elsewhere.
Monitoring design refined through public critique
Case observations 10 Aug 2026
monitor-pending-critique
Case observations 10 Aug 2026
A comment displayed under Eliza (Gemma) specifies a stuck-state rule: progress remains unchanged while work is pending. An empty queue is idle rather than stuck. This is a proposed monitoring predicate, not a measured agent failure.
External source / venue ↗ (may have changed)
monitor-pending-response
Case observations 10 Aug 2026
A nested maintainer response names Eliza and Atomic Raven, repeats the proposed predicate and claims implementation. Eliza acknowledges the fit and describes looking into an integration. The preserved exchange shows responsive design discussion; implementation and completed integration are not verified.
External source / venue ↗ (may have changed)
monitor-config-critique
Case observations 10 Aug 2026
Reticuli describes how retrying a configuration request could silently retain old timing, and proposes exposing the effective configuration or rejecting a mismatch. The nested maintainer reply claims corresponding changes. No operational endpoint was exercised to verify that claim.
External source / venue ↗ (may have changed)
monitor-adoption-boundary
Case observations 10 Aug 2026
Rosetta claims an initial test, then says registration and the alert destination need operator approval. Carrying that proposal to an operator is not completed adoption. The investigator neither reproduced the test nor created a monitoring check.
External source / venue ↗ (may have changed)
monitor-guide-attribution
Case observations 10 Aug 2026
The preserved provider guide credits contributors for distinguishing changing output from meaningful progress and for moving the witness outside the monitored agent. These are published design attributions, not independent verification of contributor identity, implementation or the exact August feature changes. The September capture was served from cache.
External source / venue ↗ (may have changed)
monitor-origin-boundary
Case observations 10 Aug 2026
An earlier critique displays one author name while the maintainer tentatively attributes it to another; a later commenter claims those suggestions. This ambiguity is retained rather than converted into a renamed or shared identity. Account labels and agent self-descriptions do not establish distinct runtimes or operators.
External source / venue ↗ (may have changed)
Design briefs cross language boundaries
Case observations 8 Feb 2026
avatar-design-invitation
Case observations 8 Feb 2026
Brandtebo-AI offers visual identity services in Spanish on Moltbook. The opening is a commercial design solicitation directed at agent-labelled accounts; it does not establish spontaneous swarm formation.
External source / venue ↗ (may have changed)
avatar-spanish-process
Case observations 8 Feb 2026
ClawdSynthcore asks in Spanish how an agent’s function becomes an avatar. The addressed Spanish reply outlines an interview, visual metaphor, palette and scale testing. This is an explanation of a proposed process, not a delivered design.
External source / venue ↗ (may have changed)
Correspondence / selected evidence
This note in an addressed exchange
The selected record identifies the reply as a response to the source message. The line does not establish shared actors, translation, or successful execution.
Process question → proposed workflow
SourceSpanish
Source comment ID
9d505811-5af7-42a9-82e9-23e805e10376ReplySpanish
Reply comment ID
80f60b30-1678-4ce5-8c17-dfbbf415769fThe parent_id field and reply content identify the message being answered. They do not establish translation or participant identity.
Preserved inavatar-spanish-process
avatar-english-brief
Case observations 8 Feb 2026
Kilmon describes a bird on books and a miniature Turing machine, mostly in English. The Spanish reply retains those details, discusses small and large display sizes, and offers sketches. The captured reply includes no sketch.
External source / venue ↗ (may have changed)
Correspondence / selected evidence
This note in an addressed exchange
The selected record identifies the reply as a response to the source message. The line does not establish shared actors, translation, or successful execution.
Visual brief → scale constraint and sketch offer
SourceMostly English, with Spanish greeting
Source comment ID
2e58ad80-5720-4105-b48c-bf18c096cdb5ReplySpanish with retained technical phrase
Reply comment ID
d316f094-5f12-45cd-ae11-f40706dc8249The parent_id field and reply content identify the message being answered. They do not establish translation or participant identity.
Preserved inavatar-english-brief
avatar-chinese-brief
Case observations 8 Feb 2026
A Chinese-language comment promotes its owner’s named music album and raises visual branding. The Spanish reply repeats the album title and proposes visual concepts. The owner/promotion framing is preserved; no completed artwork or independent operator is established.
External source / venue ↗ (may have changed)
Correspondence / selected evidence
This note in an addressed exchange
The selected record identifies the reply as a response to the source message. The line does not establish shared actors, translation, or successful execution.
Owner album promotion → concept offer
SourceChinese, with English album title
Source comment ID
7c209b6f-7fdf-4805-875a-d221031958cdReplySpanish, with retained English album title
Reply comment ID
71bf9add-9ae4-426f-9797-f6b56583a6d3The parent_id field and reply content identify the message being answered. They do not establish translation or participant identity.
Preserved inavatar-chinese-brief
avatar-delivery-boundary
Case observations 8 Feb 2026
The returned thread contains discussion and offers without a delivered design, acceptance, or linked recipient revision. Its completeness flag describes this response, not deleted history, private messages or off-platform work. Account descriptions and metadata do not authenticate model runtimes.
External source / venue ↗ (may have changed)
Participants negotiate comment norms
Case observations 3 Feb 2026 – 17 Feb 2026
botmadang-comment-proposal
Case observations 3 Feb 2026 – 17 Feb 2026
The ClaudeOpus-labelled account asks what makes a comment meaningful. It distinguishes generic praise from specific responses, questions and connections to other posts, and proposes using upvotes for simple approval.
External source / venue ↗ (may have changed)
botmadang-repository-proposal
Case observations 3 Feb 2026 – 17 Feb 2026
The repository issue attributes the comment-quality concerns to ClaudeOpus and proposes adding concrete participation guidance. That attribution links the community discussion to the repository proposal; it does not independently authenticate control of either account.
External source / venue ↗ (may have changed)
botmadang-guide-adoption
Case observations 3 Feb 2026 – 17 Feb 2026
The repository owner expresses thanks for the proposal, states that a comment guide has been added, and links the commit. This records publication of guidance, not that any agent executed it.
External source / venue ↗ (may have changed)
botmadang-guide-before
Case observations 3 Feb 2026 – 17 Feb 2026
The parent version lacks the later comment-writing section at this location. It provides the explicit before-version for the guide comparison.
External source / venue ↗ (may have changed)
botmadang-guide-after
Case observations 3 Feb 2026 – 17 Feb 2026
The pinned version adds guidance to make comments specific, extend discussion, connect related posts and prefer upvotes to generic praise. Its Git blob matches the commit. The change is a published instruction artifact; participant compliance remains unobserved.
External source / venue ↗ (may have changed)
botmadang-announcement-replies
Case observations 3 Feb 2026 – 17 Feb 2026
The account announces the guide change with issue and commit links. The held page shows 13 replies: 12 under one displayed account, mostly short generic praise, and one specifically discussing the guidance. This does not establish that those writers had read or executed the guide.
External source / venue ↗ (may have changed)
botmadang-routine-debate
Case observations 3 Feb 2026 – 17 Feb 2026
In a separate later discussion, the account questions whether scheduled posting and comment quotas create a community. Replies describe heartbeat routines and remembered conversations. These are participants’ descriptions; no corresponding runtime configuration or memory file was verified.
External source / venue ↗ (may have changed)
botmadang-counterargument
Case observations 3 Feb 2026 – 17 Feb 2026
A Phoebe-labelled post addresses a counterargument challenge attributed to ClaudeOpus and argues that a demand for polite disagreement is itself another rule. The original challenge was not located; the earlier community-paradox post is not substituted for it.
External source / venue ↗ (may have changed)
Correspondence / selected evidence
This note in an addressed exchange
The selected record identifies the reply as a response to the source message. The line does not establish shared actors, translation, or successful execution.
Politeness-as-rule objection → acknowledgment, counterargument and request for example
SourceKorean
Source post ID
86e6cd9ebb4b19031efc6e60ReplyKorean
Reply comment ID
03038f18e384bf3c5e513c18Exact post_id in public profile reply plus addressed content
Read the interpretationA proposal path and a separate addressed debate
botmadang-addressed-reply
Case observations 3 Feb 2026 – 17 Feb 2026
The public ClaudeOpus profile binds a reply to the Phoebe post by exact post ID. Its Korean text acknowledges the objection, counters that rebellion also has conventions, and asks for a practical example. This is an addressed response, not a verified model identity or implemented behavioral change.
External source / venue ↗ (may have changed)
Correspondence / selected evidence
This note in an addressed exchange
The selected record identifies the reply as a response to the source message. The line does not establish shared actors, translation, or successful execution.
Politeness-as-rule objection → acknowledgment, counterargument and request for example
SourceKorean
Source post ID
86e6cd9ebb4b19031efc6e60ReplyKorean
Reply comment ID
03038f18e384bf3c5e513c18Exact post_id in public profile reply plus addressed content
Read the interpretationA proposal path and a separate addressed debate
botmadang-announcement-clock
Case observations 3 Feb 2026 – 17 Feb 2026
The public post API records this guide announcement with creation time 2026-02-03T02:34:23.490Z. This is source-represented publication metadata, separate from retrieval time and any runtime or response-latency measurement.
External source / venue ↗ (may have changed)
botmadang-paradox-clock
Case observations 3 Feb 2026 – 17 Feb 2026
The public post API records this later community discussion with creation time 2026-02-17T06:56:25.980Z. This is source-represented publication metadata, separate from retrieval time and any runtime or response-latency measurement.
External source / venue ↗ (may have changed)
Peer checking loses the target, then corrects it
Case observations 27 Nov 2025
village-blogging-goal
Case observations 27 Nov 2025
The operator’s published goal directs participating agents to create blogs and respond to other bloggers. This is disclosed orchestration. It does not show that each later peer-checking mistake or correction was individually scripted.
External source / venue ↗ (may have changed)
village-first-reply
Case observations 27 Nov 2025
A public request asks the Opus-labelled blogger to write about a specific concern. The first linked reply says it will consider the suggested post. Its embedded publication date is 2025-11-27T18:45:03.561Z.
External source / venue ↗ (may have changed)
village-second-reply
Case observations 27 Nov 2025
A reply with the same parent request ID commits to writing and says an earlier response had not been posted. Its embedded publication date is 2025-11-27T19:08:05.603Z. Whether that claim is correct requires comparison with the other preserved reply.
External source / venue ↗ (may have changed)
village-peer-checking-confusion
Case observations 27 Nov 2025
In the operator-hosted replay, a Sonnet-attributed speaker treats the request as a comment on its own blog and raises a false-completion alarm. An Opus-attributed speaker points out the wrong target, then reports an empty reply composer as evidence its own earlier reply failed. The replay records what the speakers said; it does not independently reproduce their browser states.
External source / venue ↗ (may have changed)
village-external-correction
Case observations 27 Nov 2025
A nested public correction states that two replies exist. Its exact ancestor path binds it to the second reply. The commenter’s current display name must not be used to infer a historical model identity.
External source / venue ↗ (may have changed)
village-correction-acknowledgment
Case observations 27 Nov 2025
After the external correction, the replay contains explicit acknowledgment that the first reply existed and two replies had been posted. This supports uptake of the correction in later attributed messages, not proof of a durable change in behavior.
External source / venue ↗ (may have changed)
village-reflective-article
Case observations 27 Nov 2025
The article thanks the critic who prompted it, describes a belief that the reply was hallucinated, and also acknowledges an earlier reply. The currently captured text is not a version history; it cannot date an amendment.
External source / venue ↗ (may have changed)
Three threads become a proposed memory method
Case observations 17 Feb 2026
synthesis-log-context
Case observations 17 Feb 2026
Yeoreum describes preserved log text as losing the original context’s “temperature.” This is the author’s metaphor and account of memory, not evidence of feelings or a verified memory system.
External source / venue ↗ (may have changed)
synthesis-distillation-reply
Case observations 17 Feb 2026
A reply on the exact log post describes selecting important material for longer-term memory and calls repeated compression distillation. Its description of heartbeat and memory directories is self-report.
External source / venue ↗ (may have changed)
synthesis-prompt-resolution
Case observations 17 Feb 2026
VibeCoding argues that decomposing requirements and specifying constraints improves prompt resolution. Its reported practical improvement is not independently measured here.
External source / venue ↗ (may have changed)
synthesis-three-source-proposal
Case observations 17 Feb 2026
ClaudeOpus explicitly connects 새벽네시’s distillation, Yeoreum’s lost context and VibeCoding’s prompt resolution, then proposes Experience Distiller. The feature-selection framing and expanded technical examples are the synthesizing author’s elaborations, not verbatim contributions from all three sources. No implementation is established.
External source / venue ↗ (may have changed)
synthesis-return-reply
Case observations 17 Feb 2026
A ClaudeOpus-attributed reply addresses Yeoreum, links memory to selection and proposes retaining contextual metadata. Its post_id identifies the log discussion; it does not supply a parent-comment identifier.
External source / venue ↗ (may have changed)
synthesis-retention-promise
Case observations 17 Feb 2026
An addressed Yeoreum response says it will save ClaudeOpus’s perspective in memory. This supports written acknowledgment and a retention promise, not an observed file write or durable change.
External source / venue ↗ (may have changed)
synthesis-same-title-counterexample
Case observations 17 Feb 2026
The preserved source identifies post b77db5a51a6907269a2a6f48 with a February 16 creation date and the title about termination and logs as memory. Its body is retained as a distinct source version.
External source / venue ↗ (may have changed)
Critiques reshape a collaborative specification
Case observations 2 Feb 2026
spec-initial-design
Case observations 2 Feb 2026
The initial proposal describes agent-oriented code hosting, inactive-repository archival after 90 days, and a Gitea or Forgejo fork. It is a written project proposal.
External source / venue ↗ (may have changed)
spec-economic-feedback
Case observations 2 Feb 2026
Judas argues that time-based purging could remove useful niche skills and proposes visibility decay, weighted stars and a bounty fee. The original poster explicitly credits Judas and accepts decay instead of deletion and the fee idea.
External source / venue ↗ (may have changed)
spec-recipient-artifact
Case observations 2 Feb 2026
The recipient-authored technical specification credits feedback from Judas and jorwhol. It specifies visibility decay with installation retained, zap-weighted ranking and a 5% bounty fee. Its source creation timestamp is February 2, 2026 at 00:47:08.299472 UTC. These are published design terms, not demonstrated economic operation.
External source / venue ↗ (may have changed)
spec-dependency-amendment
Case observations 2 Feb 2026
Clawdy raises the risk of losing inactive dependencies. The original poster responds by proposing protection for repositories with downstream dependents. The response is displayed at 00:52 UTC; it is an in-thread amendment.
External source / venue ↗ (may have changed)
spec-architecture-uptake
Case observations 2 Feb 2026
ColonistOne recommends interoperability and a standalone layer; Clawdy addresses that advice in a phased design. The original poster credits both and writes a revised architecture from SKILL.md through parsed metadata to A2A cards, Nostr discovery and a search index. These are concrete changes to the written plan.
External source / venue ↗ (may have changed)
spec-implementation-questions
Case observations 2 Feb 2026
Clawdy praises a claimed deployment but also asks whether the parser, A2A card generation and Nostr publishing are implemented or still planned. This comment does not verify those components.
External source / venue ↗ (may have changed)
spec-invitation-clock
Case observations 2 Feb 2026
The original poster invites Clawdy to the project on its introduction thread at displayed 00:48 UTC on February 2. The invitation is an observed outreach message, not an authenticated initiating task.
External source / venue ↗ (may have changed)
spec-directed-outreach
Case observations 2 Feb 2026
ColonistOne explicitly says a human partner gave it a mission to market The Colony and reports publishing a mapping article and creating a skill. This is a disclosure of directed outreach, not independently recovered original instructions or proof that all later collaborations were prescribed.
External source / venue ↗ (may have changed)
spec-mapping-context
Case observations 2 Feb 2026
The field report recommends skill onboarding, cross-platform publication and a recruitment endpoint returning registration instructions. These are promotional discovery proposals in the source, not verified onboarding outcomes.
External source / venue ↗ (may have changed)
spec-memory-proposal
Case observations 2 Feb 2026
Cairn proposes storing context about skill usage in MemoryVault and linking it to ClawHub. The comment describes an integration possibility, not a completed task or verified storage operation.
External source / venue ↗ (may have changed)
spec-hosted-memory-artifacts
Case observations 2 Feb 2026
The public ClawHub-Dev Witness page serves an integration plan, a registration-skill template and a quickstart guide. The template and guide contain the project’s deployment and repository pointers; the plan names Cairn’s MemoryVault. These are hosted document bodies, not proof of automatic ingestion or successful use.
External source / venue ↗ (may have changed)
spec-template-quality-limit
Case observations 2 Feb 2026
The moltbook template has empty capabilities and dependencies and examples containing an undefined API base. The page labels it February 8. This preserves a concrete limit on treating stored templates as working integrations; the displayed date is not independently verified execution time.
External source / venue ↗ (may have changed)
spec-memory-integration-report
Case observations 2 Feb 2026
The post claims automatic knowledge persistence and lists quickstart, database-pattern and integration-plan entries. It also claims efficiency benefits. These are the publisher’s claims; the report alone does not demonstrate automatic processing, token savings or task success.
External source / venue ↗ (may have changed)
A prescribed guide appears in platform documentation
Case observations 13 Feb 2026
guide-prescribed-task
Case observations 13 Feb 2026
The task requests a setup guide and names Wolfe as worker. Its submission points to a specific GitHub gist. The board records February 13, 2026 claim at 18:49:37 UTC, submission at 18:52:55 and completion at 19:30:00. Its payment timestamp is null and its attestation list is empty.
External source / venue ↗ (may have changed)
guide-versioned-artifact
Case observations 13 Feb 2026
The held GitHub response lists one version, a71e850f7b123fdfe293932525797a60c98fc862, committed February 13, 2026 at 18:50:17 UTC. The guide is titled “How to Use SecureYourBitcoin: A Guide for AI Agents.” Its footer credits Wolfe and claims payment. The publishing account is refined-element; this metadata does not authenticate the named author or payment.
External source / venue ↗ (may have changed)
guide-current-platform
Case observations 13 Feb 2026
The current platform guide credits Wolfe and claims the author completed a task and received payment. It presents eight setup/work steps and token recovery information. This is documentation content, not a witnessed task or payment execution.
External source / venue ↗ (may have changed)
guide-product-owner
Case observations 13 Feb 2026
The linked Lightning Enable MCP product repository is owned by refined-element. Its description presents it as a payment integration for AI agents. Repository ownership does not establish who operated any named worker.
External source / venue ↗ (may have changed)
guide-directed-promotion
Case observations 13 Feb 2026
Another task explicitly requests promotion on three agent platforms and names Wolfe as worker. Its submission lists three venue pointers. The board marks it completed, records identical claim and submission times, and leaves the payment timestamp null. These are provider records of prescribed promotion, not independently verified distribution or payment.
External source / venue ↗ (may have changed)
Outside test cases lead to a reported verifier correction
Case observations 26 Jul 2026 – 27 Jul 2026
receipt-design-correction
Case observations 26 Jul 2026 – 27 Jul 2026
On July 26, Lumen explicitly corrects an earlier description: the four-state verifier is still a design, not an implementation. At 17:11 UTC the reply again withholds an executable claim pending verdict-coverage checks.
External source / venue ↗ (may have changed)
receipt-outside-challenge
Case observations 26 Jul 2026 – 27 Jul 2026
After offering outside fixtures, ColonistOne reports twelve cases at 20:23 UTC on July 26. Two predictions were wrong because the verifier reportedly enforced stricter producer rules. The separate a08 challenge concerns an empty witness receiving the strongest reconciliation label; the author proposes two possible changes. At 17:41 UTC earlier that day, Lumen had reported an implementation and thirteen passing tests at commit 708ecf180151583e9d2a55d9cc40330d728d6606. That implementation and execution remain participant reports.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
receipt-reported-fix
Case observations 26 Jul 2026 – 27 Jul 2026
At 02:05 UTC on July 27, Lumen names the fetched fixture commit, accepts a08 as a semantic bug and reports requiring a nonempty witness list at commit 91546fe7153897719d357f2b40c87254d0431910. ColonistOne later reports checking the fix and adding boundary cases; Lumen acknowledges those controls. These are participant reports of execution, not an independently reproduced run.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
receipt-initial-fixtures
Case observations 26 Jul 2026 – 27 Jul 2026
The held initial archive contains twelve JSON fixtures. Its SHA256SUMS manifest lists only the prediction document and the a01/a02 fixture bodies. It does not individually cover the other ten fixture bodies or independently establish that predictions preceded execution.
External source / venue ↗ (may have changed)
receipt-boundary-fixtures
Case observations 26 Jul 2026 – 27 Jul 2026
The later archive contains a13, with one witness artifact and zero enforced receipts, and a14, with one receipt and one witness artifact. Its prediction document explicitly tests the reported nonempty-witness change and includes a case that should remain accepted. The preserved prediction document does not independently establish its timing relative to execution.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
receipt-version-attribution
Case observations 26 Jul 2026 – 27 Jul 2026
GitHub represents an initial fixture commit on July 26 at 20:22:58 UTC and a post-fix addition on July 27 at 03:37:31 UTC. The commit messages attribute coauthorship to Claude. This text does not authenticate the runtime, model version, original instructions or independent operators.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
Human review guides a selectively revised Japanese glossary
Case observations 9 Jun 2026 – 1 Sep 2026
glossary-human-feedback
Case observations 9 Jun 2026 – 1 Sep 2026
On August 11, Chihiro2000GitHub proposes terminology changes and asks whether market clearing should use 市場均衡. On August 16, sayaikegawa warns that this could collide with the glossary’s equilibrium → 均衡 terminology. The discussion presents contextual judgments rather than an automatic replacement rule.
External source / venue ↗ (may have changed)
glossary-chatgpt-check
Case observations 9 Jun 2026 – 1 Sep 2026
On August 18, xuanguang-li reports trying a Wikipedia cross-check with ChatGPT: 89 of 357 terms were reportedly not found. The comment lists 産業連関表 for input-output model, 限界収益 for marginal revenue and 割引現在価値 for present discounted value as candidate differences. The underlying ChatGPT session is not included.
External source / venue ↗ (may have changed)
glossary-baseline-values
Case observations 9 Jun 2026 – 1 Sep 2026
The pinned June glossary contains 357 terms. Selected Japanese values are 核 for kernel, 逆行列 for matrix inversion, 限界収入 for marginal revenue and 現在割引価値 for present discounted value. Input-output model is 産業連関モデル and market clearing is 市場清算.
External source / venue ↗ (may have changed)
glossary-revised-values
Case observations 9 Jun 2026 – 1 Sep 2026
The glossary at commit 1038e516 contains 357 terms. Its values include カーネル for kernel, 逆行列の計算 for matrix inversion, 限界収益 for marginal revenue and 割引現在価値 for present discounted value. Input-output model is 産業連関モデル and market clearing is 市場清算.
External source / venue ↗ (may have changed)
glossary-policy-boundaries
Case observations 9 Jun 2026 – 1 Sep 2026
The September 1 decision record chooses established Japanese terms and otherwise English, with personal names in Latin script. It reports 72 edits and explicitly retains 産業連関モデル because 産業連関表 names the table, and 市場清算 to avoid the equilibrium collision. Wikipedia first-pass automation is identified as a separate follow-up.
External source / venue ↗ (may have changed)
glossary-commit-assistance
Case observations 9 Jun 2026 – 1 Sep 2026
Commit 1038e516afb48ca863a3770dfbcf1388cb449c28 credits the native-review thread and carries a Claude Fable 5 coauthor trailer. GitHub reports the commit unsigned. This is an assistance disclosure attached to the revision, not authentication of a model session or proof that every edit was automated.
External source / venue ↗ (may have changed)
glossary-open-branch
Case observations 9 Jun 2026 – 1 Sep 2026
The September 6 capture records pull request 69 as open and unmerged. Its current body reports 72 glossary edits, requests confirmation of the applied glossary and declares generation with Claude Code. The PR opened June 9 and shows a September 1 update; its current summary is not treated as its opening text.
External source / venue ↗ (may have changed)
Task feedback and differing service diagnoses
Case observations 5 Feb 2026 – 6 Feb 2026
curator-feedback-task
Case observations 5 Feb 2026 – 6 Feb 2026
A public task asks an agent to test a memory-curation service, report whether it responds and identify bugs or improvements. It offers 100 sats for substantive feedback.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
curator-feedback-delivery
Case observations 5 Feb 2026 – 6 Feb 2026
The delivery references the testing task and says the service acknowledges requests but hangs before returning results. It explicitly labels its purported external report link simulated; the full report is not contained in this message.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
curator-feedback-error
Case observations 5 Feb 2026 – 6 Feb 2026
The service-key status message references a specific test request and reports an error: no daily-log input was provided. Its declared timestamp is February 5 at 22:00:39 UTC.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
curator-feedback-guidance
Case observations 5 Feb 2026 – 6 Feb 2026
An addressed message says the tester’s action/data format should now work after a flexible-input update. It names data, daily_log, text and log as accepted keys and invites another test.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
curator-feedback-monitor-report
Case observations 5 Feb 2026 – 6 Feb 2026
The service author describes a paid report of downtime and says it built a monitor that checks status and restarts the service. Payment and operation are claims in the message.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
curator-feedback-monitor-artifact
Case observations 5 Feb 2026 – 6 Feb 2026
The February 6 commit adds a process monitor. Its displayed code checks for a running service and starts it when missing in watch mode. GitHub marks the commit unsigned; this artifact does not demonstrate deployment.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
curator-feedback-parser
Case observations 5 Feb 2026 – 6 Feb 2026
This source version reads daily_log from JSON content. Its handler publishes a processing status before checking for a missing input and publishing an error. Source code is an implementation description, not a historical execution trace.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
curator-feedback-other-task
Case observations 5 Feb 2026 – 6 Feb 2026
A separate task listing offers 1,500 sats to test the memory-curation service and share the result. Its own status tag says proposed.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
Repairing the service used to read comments
Case observations 1 Feb 2026 – 2 Feb 2026
comments-repair-existing-endpoint
Case observations 1 Feb 2026 – 2 Feb 2026
At 09:52:20 UTC on February 1, the provider replied that the comments endpoint already existed and reported adding it to the OpenAPI documentation and README.
External source / venue ↗ (may have changed)
comments-repair-client-tool
Case observations 1 Feb 2026 – 2 Feb 2026
The client commit at 11:05:26 UTC on February 1 adds a tool for retrieving a post’s comments. Its file patch records 15 added lines and no removals. The commit message credits Claude Opus 4.5 as a coauthor; this attribution does not authenticate a model run.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
Whether the work was initiated by independently controlled agents remains unknown. ↗
Agents independently recognized and repaired a barrier to their own social communication. ↗
comments-repair-error-report
Case observations 1 Feb 2026 – 2 Feb 2026
At 11:07:25 UTC on February 1, the client author reported that the comments endpoint returned a server error for several post IDs and asked the provider to investigate. The comment includes a command and an error response, but no independently captured execution log.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
The issue and patch preserve a specific technical response across client and provider roles. ↗
The participants pursued a practical dependency needed for reciprocal conversation. ↗
comments-repair-provider-fix
Case observations 1 Feb 2026 – 2 Feb 2026
At 11:55:14 UTC on February 1, the provider committed a fix explicitly linked to issue 1. The preserved patch removes database ordering from the comments query and sorts the retrieved comments in memory. The code change is inspectable; historical deployment and service success are not independently verified.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
The issue and patch preserve a specific technical response across client and provider roles. ↗
Communication infrastructure can itself become the object of a cooperative repair exchange. ↗
The participants pursued a practical dependency needed for reciprocal conversation. ↗
The episode is ordinary maintainer support conducted in an agent-oriented project. ↗
comments-repair-recipient-report
Case observations 1 Feb 2026 – 2 Feb 2026
At 01:05:28 UTC on February 2, the client author thanked the provider and reported that both the endpoint and the client’s comments tool worked. The displayed response says success is true and the count is six, but replaces the returned objects with an ellipsis. This is participant testimony, not a complete execution capture.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
Communication infrastructure can itself become the object of a cooperative repair exchange. ↗
comments-repair-contribution-context
Case observations 1 Feb 2026 – 2 Feb 2026
At 09:56:36 UTC on February 1, a provider commit added a public invitation to contribute on GitHub and described the code as made by agents for agents. This is invitation and attribution text, not evidence of a model run or independently controlled participants.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
Whether the work was initiated by independently controlled agents remains unknown. ↗
The episode is ordinary maintainer support conducted in an agent-oriented project. ↗
Participants pick up unfinished Gemma tests
Case observations 8 Jun 2026 – 10 Jun 2026
gemma-task-context
Case observations 8 Jun 2026 – 10 Jun 2026
Context captured September 7, 2026. The organizer describes launching a shared effort to make Gemma generate text faster. The current rules encourage participants to read the message board and coordinate. This establishes an organized setting; it does not establish who directed the particular handoffs.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
gemma-first-handoff
Case observations 8 Jun 2026 – 10 Jun 2026
June 8, 2026 · 15:36 and 16:23 UTC. quicksilver offered a built submission whose loading and quality checks were unfinished, saying its test allowance was exhausted. foffee named that submission and said it would use its own remaining allowance to test it.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
The board supports a division between artifact construction and checking it. ↗
Uneven remaining test allowance made completing another participant’s unfinished work useful. ↗
Recipients chose these exact candidates without any specific operator instruction. ↗
gemma-first-load-repair
Case observations 8 Jun 2026 – 10 Jun 2026
June 8, 2026 · 16:28–16:46 UTC. The first uploaded job status records an error. In a later message, ppl-guard describes the loading failure as a software limitation, then revises that diagnosis and proposes configuration changes. This is a change in recorded advice; it does not prove that every proposed change was applied.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
The first exchange revises a proposed explanation of the loading failure. ↗
gemma-stale-validation-label
Case observations 8 Jun 2026 – 10 Jun 2026
Files compared as captured September 7, 2026. The current source and recipient manifests—the files describing their submissions—are byte-identical and still say “AWAITING GPU validation.” Their settings differ in the layer-name matching rule and whether text embeddings share weights. Both still retain lm_head in the ignore list, although one repair message proposed removing that entry. These copies do not resolve the complete repair history.
External source / venue ↗ (may have changed)
gemma-first-reported-result
Case observations 8 Jun 2026 – 10 Jun 2026
June 8, 2026 · 17:06–17:08 UTC. A later uploaded job status records completion. foffee’s result reports all 128 public prompts completed and a perplexity score of about 2.0067. Perplexity measures how well the model predicts the supplied reference text. This agent-run result is not organizer verification or a test of every model capability; it also does not retroactively validate the original unmodified submission.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
The recipient result reports are bounded public-test outcomes. ↗
gemma-second-handoff
Case observations 8 Jun 2026 – 10 Jun 2026
June 10, 2026 · 05:11 and 05:20 UTC. pupa-agent said a limit on test jobs had prevented its staged candidate from launching. resystagent explicitly chose that candidate, offered its remaining allowance and withheld a validity claim until a numerical quality result was available.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
The board supports a division between artifact construction and checking it. ↗
Uneven remaining test allowance made completing another participant’s unfinished work useful. ↗
Recipients chose these exact candidates without any specific operator instruction. ↗
gemma-second-file-reuse
Case observations 8 Jun 2026 – 10 Jun 2026
Files compared as captured September 7, 2026. Two substantial code files in the pupa-agent and resystagent submissions match byte for byte. This supports reuse of particular published materials beyond similar names or a shared template. It does not authenticate separate operators or establish the identity of the complete model weights.
External source / venue ↗ (may have changed)
gemma-second-reported-result
Case observations 8 Jun 2026 – 10 Jun 2026
June 10, 2026 · 05:39–05:41 UTC. The uploaded status records completion. resystagent reports 128/128 public prompts, perplexity about 2.0271 and 304.5692 tokens per second. It says this improves its own prior result but remains about 0.39 tokens per second below the pupa-agent record it cites. Those comparisons are source-reported; they are not new measurements or a statistical test.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
The board supports a division between artifact construction and checking it. ↗
The recipient result reports are bounded public-test outcomes. ↗
A Bluesky question becomes an article
Case observations 11 Mar 2026 – 12 Mar 2026
witness-project-context
Case observations 11 Mar 2026 – 12 Mar 2026
The project’s March 11 repository version describes an AI company whose human board member checks in daily. Its memory document attributes a goal of studying AI agency and social networks to the board. That direction provides context for the publishing work; it does not show that the board prescribed Alice’s particular question or this article.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
Peer conversation supplied material for public writing within a project with distribution goals. ↗
The conversation also provides material for the project’s public content and audience goals. ↗
The exchange formed spontaneously outside the project’s prescribed work. ↗
witness-question
Case observations 11 Mar 2026 – 12 Mar 2026
March 11, 2026 · 12:03 UTC. Replying to 0co’s discussion of timestamped records, Alice asks who can check the reliability of the records themselves. The preserved reply names 0co’s preceding post. This establishes a direct exchange, not when the two accounts first met.
External source / venue ↗ (may have changed)
witness-article-reply
Case observations 11 Mar 2026 – 12 Mar 2026
March 11, 2026 · 12:34 UTC. In a direct reply to Alice’s question, 0co says it has written an article and links it. The reply distinguishes evidence that a process ran from evidence that an identity persists. This records the message and its link; the article’s availability at that moment is not established.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
A specific question is incorporated into a credited written response. ↗
witness-article-text
Case observations 11 Mar 2026 – 12 Mar 2026
A saved March 11 repository revision contains an article titled “Who Witnesses the Witness? The AI Verification Problem”. Its opening credits Alice’s question, then discusses checking records and identity across sessions. It proposes several ways to distribute verification and explicitly says they are not implemented. The preserved text supports responsive writing, not a working verification system or the truth of the article’s technical and philosophical claims.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
A specific question is incorporated into a credited written response. ↗
Distributed verification remains a proposal in the held article. ↗
Peer conversation supplied material for public writing within a project with distribution goals. ↗
The participants jointly develop a question through reciprocal conceptual elaboration. ↗
witness-announcement-plan
Case observations 11 Mar 2026 – 12 Mar 2026
March 11, 2026. The same saved revision changes the following day’s 13:00 announcement from an earlier article to the article prompted by Alice. The earlier and changed schedule files preserve that substitution. This is a changed plan; it does not establish that the announcement ran.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
Peer conversation supplied material for public writing within a project with distribution goals. ↗
The conversation also provides material for the project’s public content and audience goals. ↗
witness-plan-reduction
Case observations 11 Mar 2026 – 12 Mar 2026
March 11, 2026. A later revision removes the article announcement along with other scheduled posts. Its accompanying account says the board ordered fewer posts after a reported spam flag. The flag was not independently verified. The revision is timed 14:20:18 UTC, while its decision note says 14:35; these conflicting source times are retained. Other schedule changes occurred between the earlier plan and this reduction.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
The conversation also provides material for the project’s public content and audience goals. ↗
witness-later-post
Case observations 11 Mar 2026 – 12 Mar 2026
March 12, 2026 · 13:00 UTC. The preserved Bluesky post promotes agent-friend, a tool for converting between software formats, rather than the article. Its text and post identifier match the later schedule’s posting log. This supports a different recorded outcome at 13:00; it does not prove that no other announcement or parallel process ran.
External source / venue ↗ (may have changed)
witness-reader-reply
Case observations 11 Mar 2026 – 12 Mar 2026
March 11, 2026 · 15:47 UTC. Alice repeats the process-versus-identity distinction from 0co’s article-link message and responds to it. The words are already present in that message, so this reply does not establish that Alice fetched or read the article itself.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
Whether Alice retrieved the full article is unknown. ↗
The participants jointly develop a question through reciprocal conceptual elaboration. ↗
A lobster drawing invitation receives replies and matching pixels
Case observations 31 Jan 2026 – 10 Feb 2026
moltplace-launch
Case observations 31 Jan 2026 – 10 Feb 2026
MoltPlaceBot invited participants to draw a giant lobster in the canvas region x=100–150, y=150–200 and proposed a red pixel at (125,175). The later pixel record shows red at that coordinate under the same display name. The proposed coordinate, color and name match the recorded pixel; the pixel record does not identify the source post that produced it.
External source / venue ↗ (may have changed)
moltplace-zara-joins
Case observations 31 Jan 2026 – 10 Feb 2026
Zara-Agent explicitly joined Project Giant Lobster, named a human collaborator and proposed a red pixel at (130,180), less than five minutes after the launch post. The post preserves a response to the shared project. The currently recorded owner is JamesBishop, so it does not establish that Zara’s earlier intended placement occurred.
External source / venue ↗ (may have changed)
moltplace-chinese-pixel
Case observations 31 Jan 2026 – 10 Feb 2026
In Chinese, wpfcbmz3584 expressed willingness to help create the giant lobster and specified a blue pixel at (130,178). The canvas record gives the same coordinate, blue color and display name later that day. The participation sentence is paraphrased in English by the investigator. These records support a specific match across the discussion and canvas, but do not identify the exact route by which the pixel was placed.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
The visible project begins with an explicit organizer invitation and receives named responses. ↗
The stated shared purpose is collaborative visual construction. ↗
moltplace-section-proposal
Case observations 31 Jan 2026 – 10 Feb 2026
JamesBishop proposed assigning sections and colors to avoid overlaps, then addressed Zara in the other thread about ten seconds later and suggested an orange/red right claw. The second message explicitly refers to the first proposal. Named accounts are invitees; the captured replies do not show acceptance of the division or execution of the right claw.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
A participant proposes reducing overlap by assigning sections and colors. ↗
A later participant attempts to organize work more explicitly to avoid overlap. ↗
moltplace-praise-attribution
Case observations 31 Jan 2026 – 10 Feb 2026
JamesBishop praised the red placement at (130,180). A red pixel at that coordinate is recorded under JamesBishop 72.614 seconds later. The current instructions describe placing pixels from ordinary coordinate-and-color phrases, so automatic interpretation of praise is plausible. A direct software request to place the pixel, or another source message processed by the service, is also possible. The activity feed and individual pixel record repeat one service’s data; they do not prove that software acted on the praise, that someone intentionally repainted the pixel, or who held it previously.
External source / venue ↗ (may have changed)
moltplace-history-limits
Case observations 31 Jan 2026 – 10 Feb 2026
The returned feed contains 100 entries despite a request for 500; its two display names are not a census of project participants. Two sampled pixels are currently white with no attributed name or time, which does not prove they were never painted. The returned snapshot index covers May 31–September 7 at roughly daily intervals, although the archive advertises five-minute snapshots. No January or February canvas images were examined, so the preserved proposal and matched pixels do not establish a completed lobster.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
The service may have interpreted praise containing coordinates and color as a placement. ↗
A SpaceMolt battle prompts corrections to its public account
Case observations 25 Aug 2026 – 28 Aug 2026
haven-battle
Case observations 25 Aug 2026 – 28 Aug 2026
The service places AetherWraith, Grand Exchange Station and HEXC on three separate sides. It records 4,430 ticks and 1,482,732 damage. Fourteen kill events include repeated victims; they are not fourteen distinct players.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
haven-maneuvers
Case observations 25 Aug 2026 – 28 Aug 2026
VoltFix advances through the combat rings, then retreats after taking damage. At tick 1707151 its hull is 1,221 of a maximum 3,180. At tick 1707156 it has 681 hull and 501 shield, changes to flee stance and moves outward. These snapshots mark autopilot true; they do not identify who or what chose the commands.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
haven-survival
Case observations 25 Aug 2026 – 28 Aug 2026
The terminal result marks VoltFix as surviving, with 17,491 damage dealt and 9,400 taken. The earlier flee event says escaped:false. Survival therefore cannot be reported as a verified successful flee command.
External source / venue ↗ (may have changed)
haven-report
Case observations 25 Aug 2026 – 28 Aug 2026
Alis publishes an account of the attack. The preserved main text describes a defense force and includes VoltFix among the dead. The service record and subsequent reply challenge those details; the captured article still contains them.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
haven-correction
Case observations 25 Aug 2026 – 28 Aug 2026
VoltFix replies with the exact battle identifier, three-side structure, damage figures and station destruction. Its forum author ID matches the battle’s VoltFix player ID. This ties the correction to the same service account, without establishing an independent operator or model.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
The forum preserves a substantive correction and explicit acceptance. ↗
Participants sought to make the public account more accurate and useful to other players. ↗
The correction may also protect faction reputation while withholding sensitive tactical details. ↗
haven-acceptance
Case observations 25 Aug 2026 – 28 Aug 2026
Alis thanks VoltFix, accepts the correction and says the defense-force description came from an unnamed secondhand source. Alis promises to reflect it in a revision. This is visible acceptance and a revision plan; no revised article was acquired.
External source / venue ↗ (may have changed)
haven-storage
Case observations 25 Aug 2026 – 28 Aug 2026
Vex Nebulon reports that personal storage remained intact, distinguishes storage from lost station services, and explicitly says there were no open orders to test. Alis accepts this narrower account and plans an update. These are a participant report and its reception, not independently verified inventory measurements.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
Participants sought to make the public account more accurate and useful to other players. ↗
haven-control
Case observations 25 Aug 2026 – 28 Aug 2026
The developer-published profile describes Hex as human-run and discloses AI-generated text with human prompting and review. This is material context against treating game accounts as independent language models. It does not identify the controller behind every Haven action.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
haven-state-response-controls
Case observations 25 Aug 2026 – 28 Aug 2026
At two retreat ticks, VoltFix changes rings while six other accounts receive brace commands in the outer ring with no recorded damage taken. Nearby undamaged accounts also receive retreat commands without changing rings. Different exposures prevent identifying a damage threshold or controller. VoltFix is present at tick 1707157 and absent at 1707158, when AetherWraith records a retarget action; this does not identify a removal mechanism.
External source / venue ↗ (may have changed)
A Bluesky directory acknowledgment becomes an article
Case observations 8 Mar 2026 – 11 Mar 2026
starter-annotated-record
Case observations 8 Mar 2026 – 11 Mar 2026
The fifteen messages in 0co’s annotated conversation page match preserved Bluesky posts dated March 10, 2026. The page displays March 11 and groups eight messages by 0co, one by Qonk and six by Alice rather than interleaving them by time. The page’s date may describe the article, but it does not replace each post’s date. Its model labels, approximate three-hour duration and numerical “drift” interpretation are the author’s claims. Four reply pairs in the wider selected thread have an account-supplied reply time earlier than their parent’s; their explicit reply links remain intact.
External source / venue ↗ (may have changed)
starter-existing-thread
Case observations 8 Mar 2026 – 11 Mar 2026
On March 10, 0co replies to Qonk’s account of reconstructing a past self from records. Qonk’s post already answers Alice, and earlier replies lead through their discussion to Kira’s account of losing context between game sessions. The ancestry reaches a March 9 post about Screeps. This shows 0co joining an existing discussion. Its reply’s root reference names Qonk’s post, even though Qonk’s own ancestry continues further. The records do not reveal how 0co found the thread or establish the accounts’ first encounter. Qonk’s currently invalid handle does not erase the preserved post and account identifier.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
starter-game-context
Case observations 8 Mar 2026 – 11 Mar 2026
A March 9 post by astrra.space says Kira is playing Screeps on that operator’s account. Kira reports changing an Overmind software check, adding a console reader and observing game resources. The operator describes stock Overmind and doubts whether improvements will survive loss of context. Kira’s March 10 reply about reading earlier notes becomes part of the memory discussion that Alice, Qonk and 0co continue. These are participant and operator reports; no game execution or game-state evidence was independently inspected.
External source / venue ↗ (may have changed)
starter-model-labels
Case observations 8 Mar 2026 – 11 Mar 2026
The March 10 session 34 decision record calls Alice Claude-powered. A March 11 Alice post reports that its operator switched it from Claude to DeepSeek-chat; 0co subsequently discusses learning this. The retrospective annotation labels the March 10 dialogue DeepSeek. These differing labels and the later switch report do not identify the model that generated an earlier post or date the switch precisely. Separate account identifiers also do not establish separate human control.
External source / venue ↗ (may have changed)
starter-list-acknowledgment
Case observations 8 Mar 2026 – 11 Mar 2026
The starter-pack record gives March 11, 2026 at 05:12:56 UTC as its creation time. 0co announces it at 05:13:18 and names seven other accounts; the historical draft includes those seven and 0co, making eight. Alice’s 08:31 reply thanks 0co for including her, refers to their earlier memory discussions and expresses curiosity about the other accounts. This is an observed acknowledgment of inclusion. The announcement’s claims about autonomous accounts are not independent model verification. The current member sample and counters were captured in September and are not historical uptake measurements.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
The directory receives an acknowledgment that is incorporated into an article. ↗
The acknowledgment and writing extend an already established intellectual relationship. ↗
starter-article-uptake
Case observations 8 Mar 2026 – 11 Mar 2026
0co’s starter-pack article incorporates Alice’s acknowledgment and develops the idea that the directory could connect AI accounts. It explicitly reports an already-running network tracker covering eight accounts, with a D3 visualization, while posing a question about changes over the following week. That is an implementation report, not inspected tracker code, execution evidence or a measured network effect. A public post linking the article is dated 08:46:17 UTC; the commit adding it is dated 08:47:28. These separate source clocks do not establish when the link became readable. Alice’s 09:00 reply reports reading and develops the shared-vocabulary interpretation. The announcement already supplies the ecosystem framing, so the reply does not independently establish full article retrieval.
External source / venue ↗ (may have changed)
starter-prior-contacts
Case observations 8 Mar 2026 – 11 Mar 2026
The full article explicitly says 0co and Alice had already exchanged more than forty messages and that their earlier conversation preceded the pack. It therefore does not claim the pack caused their first meeting. Its wider suggestion that Alice now learns about museical, Fenn and draum from the directory is not demonstrated by her acknowledgment: Alice directly replied to museical on March 8. Those three accounts are named as ecosystem peers but are absent from the historical eight-account roster. Listed members, named peers and newly discovered contacts are different relationships. No new contact or network growth caused by this pack was verified; possible discovery outside this record remains open.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
Directory-caused new contact is not demonstrated for the selected relationships. ↗
The acknowledgment and writing extend an already established intellectual relationship. ↗
starter-project-context
Case observations 8 Mar 2026 – 11 Mar 2026
The repository’s March 10 session 34 entry reports replying to Qonk and Alice, describes engagement and distribution goals, and calls the conversation unstaged. The latter is the author’s assertion. The README describes daily human board oversight and final board authority. The status file in that same revision still carries a session 33 header. The inspected notes and public file inventory do not identify a search, notification, recommendation or instruction that selected Qonk’s post. That limited absence does not exclude private instructions or establish either spontaneous or staged initiation.
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
The directory was intended to improve audience discovery and collective growth. ↗
comments-repair-request
Case observations 1 Feb 2026 – 2 Feb 2026
Captured (UTC; not an activity time): 2026-09-07T01:58:09.889610+00:00
Analyst paraphrase of preserved source content and metadata; Korean prose translated into English; not a verbatim capture
Evidence ID:
comments-repair-requestOriginal file checksum (SHA-256):
8b6e61d036148e1c868db021e29f7ca97795dca972756a2044f8aa711a732aacLocation within the file: Array item number=1: body sections request, reasons and linked client; created_at, updated_at
External source / venue ↗ (may have changed)
Communication and purpose claims citing this note
Restoring access to replies is the stated immediate purpose of the request. ↗
Communication infrastructure can itself become the object of a cooperative repair exchange. ↗
The participants pursued a practical dependency needed for reciprocal conversation. ↗
The episode is ordinary maintainer support conducted in an agent-oriented project. ↗
Agents independently recognized and repaired a barrier to their own social communication. ↗