ACPs: Chinese interconnection implementation, not recovered swarm activity Reviewed September6,2026 UTC Primary source and provenance https://github.com/AIP-PUB/ACPs-community/tree/3985c1330209f075c124669cc9d31f0e75448140 The README identifies Beijing University of Posts and Telecommunications and China Electronics Standardization Institute involvement. It describes a travel-planning leader and five specialist partners. This is project-level primary attribution, not attribution of any external anonymous posts. Its claimed standards history was not independently audited here. Source coverage GitHub returned a nontruncated tree with2281 entries. Six files downloaded at the pinned revision and Git-blob-SHA1 verified: README.md, demo-leader/README.md, monitor-server/README.md, acps-docs/tutorials/amp-agent-observability.md, acps-sdk/acps_sdk/amp/metrics_demo.py, acps-sdk/acps_sdk/amp/emitter.py. No investigated code executed. No filenames ending .log, .jsonl or .ndjson occur in this tree. That is a filename inventory result, not proof there are no embedded examples, historical logs, release artifacts or private deployments. No actual conversation run was recovered in this pass. Useful evidence locations The observability tutorial specifies local JSONL/NDJSON then forwarding to a monitor, with separate heartbeat, metrics, access, message, system and audit families. Searchable filenames include amp_message_.jsonl and amp_access_.jsonl. Correlation/session identifiers and sender/recipient identities would be useful for matching a real exchange across records. The documented monitor address is localhost; no external deployment URL was established. No internal design or authenticated monitor was pursued. Source-level caution metrics_demo.py explicitly implements a demo-only sampler. Static inspection confirms synthetic active/queued task counts, resource percentages, success rates and latency values; these must not be treated as evidence of real workloads if encountered in an example dashboard. This does not imply all ACPs metrics are synthetic: production providers are separately supported. AuditEmitter appends JSON records and optionally signs them. Missing signer leaves records unsigned, and signing exceptions fall back to unsigned output. A log-format match or the presence of an integrity-capable SDK therefore does not authenticate a record. No signing key, credential or private certificate material was read. Access result https://gitcode.com/LuffyTeam/ACPs-community returnedHTTP200 and94570 bytes, with project metadata in direct HTML. The earlier web-reader cache miss was not a demonstrated Chinese network block. The GitHub tree/source routes also worked here. No infrastructure purchase warranted by this access test. New adjacent lead Search surfaced Punky971210/dsh-punky-swarm, described by a catalog as a DeepSeek Harness plugin implementing ACPs discovery. GitHub tree request returned403; this is unresolved, not a negative finding. Use ordinary public repository/browser retrieval next. Do not assume the catalog's DeepSeek label implies lab ownership. Assessment and preservation Concrete Chinese inter-agent infrastructure and documented log shapes recovered; no live swarm or escape attribution established. Raw captures and SHA256SUMS in265-private. Retain ACPs as a source of precise fingerprints, with actual run records still the missing join.