Record host mac-coordinator-64gb-20260925: hardware profile and five use receipts

[seathatflowsinourveins]

Scope

• What this PR changes: records the replacement coordinator Mac as host mac-coordinator-64gb-20260925 through the new-host loop (docs/contributing-evidence.md section 0): its measured hardware profile and five native_proven use-stage receipts for the components that its replacement checks exercised, plus the regenerated matrix and grand list.
• Base commit: 720e294b5c10211d8f75a939bc478af734279d74 (the receipts' catalog_revision, where they were recorded); the branch is rebased onto f036e96e (Bind host receipts to landscape winner ids: refuse stack-id aliases, list unbound Mac receipts #214), with the generated files and manifests/evidence.json regenerated there
• Owned paths touched: evidence/hosts/mac-coordinator-64gb-20260925/, evidence/artifacts/hw-profiles/mac-coordinator-64gb-20260925/, the new hosts[] entry in adoption/hardware-profiles.json, and the generated manifests/evidence.json, catalogs/landscape/component-evidence-matrix.json, catalogs/landscape/new-host-grand-list.json, docs/component-evidence-matrix.md and docs/new-host-grand-list.md.

Recorded from a fresh clone of the default branch, not from a pinned install checkout, as adoption/update.md ("Moving a host to a new release", step 4) requires. The host ID was confirmed by the user on 2026-09-25. adoption/launchd/launchd-agents.sh install was not run. Every command ran on this Mac (macOS 27.0 26A428, arm64) from a Claude Code session outside the Seatbelt sandbox.

Evidence-class table

Claim	Evidence class	Command / receipt
Hardware: Apple M5 Pro, 18 cores, 64 GB unified memory, arm64	native_proven	hardware_profile.py --record-host; evidence/artifacts/hw-profiles/mac-coordinator-64gb-20260925/profile.json
Qdrant 1.19.1 (pin) on the running launchd service created a temporary collection, upserted three passages embedded by this host's Ollama (qwen3-embedding:4b, 2560 dims), ranked the backup passage first, deleted the collection (GET 404, collection count unchanged)	native_proven	evidence/hosts/mac-coordinator-64gb-20260925/mac-coordinator-64gb-20260925--qdrant--use--20260925.json
QMD 2.8.3 (pin) indexed adoption/ into a temporary index (Indexed: 17 new, from its own collection add output); release tag re-pin returned adoption/update.md first; a nonsense query returned 0 hits	native_proven	...--qmd--use--20260925.json
restic 0.19.1 (pin): backup, check --read-data, restore --verify (each exit code printed), byte-identical restore with no extra files; one flipped byte in a pack makes check --read-data exit 1 (positive control); reads selected fields of the backup script's latest receipt on this host (repository on /Volumes/ORICO, exFAT, snapshot 5c15e0b04f3f, all four verification flags true; it does not show which restic binary that script ran)	native_proven	...--restic--use--20260925.json
SocratiCode 1.14.0 (pin; a yes/no check confirms the Codex MCP entry launches that build) indexed a three-file temporary project over MCP stdio (59 chunks, Ollama embeddings into Qdrant), answered three questions with the expected file first, then codebase_remove left the Qdrant collection list unchanged	native_proven	...--socraticode--use--20260925.json
ai-memory server 2.4.0 answered status (live capture, full FTS coverage, both local providers ok) and MCP initialize, tools/list, scoped memory_status and scoped memory_query (hits returned; counts only in the receipt)	native_proven, unbound version	...--ai-memory--use--20260925.json

The ai-memory receipt is recorded with --component-version 2.4.0 --allow-unbound-version: the winner pin is 2.3.2, so it counts toward no derived status (the generated matrix still tallies it as a macOS pass, untested [1/0/0/0], because component_matrix.py counts unbound passes). scripts/receipt_staleness.py lists it under unbound (version 2.4.0). It is here as dated evidence that the service runs on this host.

Not recorded, because there is no catalog component or because the check is not a component's job: Ollama 0.34.4 (no manifests/stack.json or landscape id; it appears as the embedder inside the qdrant and socraticode receipts), and the agent-ecosystem health.py, native_policy.py and access_doctor.py results (repository and configuration checks of that project, not component use). Codex and Claude Code run versions that differ from the pins, and no functional check of them was made here.

Local commands run

$ python3 scripts/hardware_profile.py --record-host mac-coordinator-64gb-20260925 --label "macOS arm64, Apple M5 Pro, 18 cores, 64 GB unified memory (measured)"
{"status": "recorded"}
$ python3 scripts/host_receipts.py record ... (five times; every command in every receipt exits 0)
$ python3 scripts/component_matrix.py --write      -> {"flip_rule_violations": 0, "rows": 32, "status": "written"}
$ python3 scripts/new_host_grand_list.py --write   -> {"status": "written", "layers": 32, "winners": 66}
$ python3 scripts/host_receipts.py validate        -> {"receipts": 42, "status": "passed"}
$ python3 scripts/validate.py                      -> status passed (5447 hashed files after the rebase)
$ python3 scripts/component_matrix.py --check; python3 scripts/new_host_grand_list.py --check   -> checked / passed
$ python3 scripts/validate_catalogs.py; python3 tools/sota-convergence/build_verdicts.py --check; python3 scripts/build_ecosystem.py --check   -> passed
$ adoption/tools/gitleaks-guarded-macos dir evidence/hosts/mac-coordinator-64gb-20260925 --no-banner --redact   -> no leaks found (exit 0)
$ adoption/tools/gitleaks-guarded-macos dir evidence/artifacts/hw-profiles/mac-coordinator-64gb-20260925 --no-banner --redact   -> no leaks found (exit 0)
$ git diff --check   -> clean
$ python3 -m unittest   -> Ran 5169 tests in 617.5s, OK (skipped=746), exit 0, at 27b79a9b before the rebase (a detached worktree); after the rebase onto f036e96e: tests.test_host_receipts, test_component_matrix, test_new_host_grand_list, test_receipt_staleness and test_hardware_profile OK

scripts/verdict_flip_candidates.py lists no row from these receipts: they carry only the recorder's self review.

Decision record

None: evidence only. No verdict, pin or platform_status changes.

Host evidence

• python3 scripts/host_receipts.py validate passes for every new/changed receipt.
• Independent review is explicitly requested in this PR: each receipt carries only the recorder's self review. A receipt-level review (scripts/host_receipts.py review, contributing-evidence step 8, four points) needs a session other than the recording one; a subagent of the recording session would inherit its session id and be refused. The PR diff itself was reviewed by a separate headless reviewer session; its findings and their resolution are in a PR comment.
• No platform_status change is made from a host receipt alone.

Receipts (all native_proven, stage use, second_physical_machine: true, hardware_profile_ref: mac-coordinator-64gb-20260925):

• evidence/hosts/mac-coordinator-64gb-20260925/mac-coordinator-64gb-20260925--qdrant--use--20260925.json
• evidence/hosts/mac-coordinator-64gb-20260925/mac-coordinator-64gb-20260925--qmd--use--20260925.json
• evidence/hosts/mac-coordinator-64gb-20260925/mac-coordinator-64gb-20260925--restic--use--20260925.json
• evidence/hosts/mac-coordinator-64gb-20260925/mac-coordinator-64gb-20260925--socraticode--use--20260925.json
• evidence/hosts/mac-coordinator-64gb-20260925/mac-coordinator-64gb-20260925--ai-memory--use--20260925.json (unbound version)

Checklist

• New/changed GitHub Actions are pinned to a full commit SHA with a version comment (no floating tags). No workflow changes.
• New/changed workflows declare top-level permissions: contents: read. No workflow changes.
• No secrets are printed, logged or committed; no new required secret was added. The Keychain was not read; the ORICO repository was not opened.
• No new paid hosting, subscription or billing surface was introduced.
• Peer-owned untracked files and worktrees were preserved.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.

[seathatflowsinourveins]

Independent PR-diff review by a separate headless Claude Code session (Opus, read-only tools, at head 7989aaa9). It ran host_receipts.py validate, validate.py, component_matrix.py --check and new_host_grand_list.py --check (all pass) and reported no blocking issue. This is a diff review, not a receipt-level host_receipts.py review; that is still requested.

#	Finding	Resolution at 27b79a9b (receipts re-recorded, not hand-edited)
1	restic claim said "first" backup to "the ORICO drive" and "ran restic 0.19.1", which command 4's output did not show	Command 4 now also prints the volume mount (/Volumes/ORICO) and repository directory name; the claim drops "first", and the limitation says the receipt does not show which restic binary the backup script ran
2	QMD's 17-file count came from Python counting files on disk	Command 2 keeps QMD's own collection add line (Indexed: 17 new ...)
3	SocratiCode claim named Qdrant 1.19.1 with no command recording it	Command 1 prints the Qdrant and Ollama versions
4	restic "exit 0" text was fixed in the print statement	Command 2 captures and prints the real backup, check and restore exit codes
5	ai-memory "2.5.0 pre-release" wording	Now "install directory is labelled 2.5.0 at source commit 19b6429"
6	QMD_CONFIG_DIR could override the temporary XDG config	Command 2 unsets it
7	SocratiCode command 1 printed client configuration values	Replaced with a yes/no check that the Codex entry launches tools/socraticode-1.14.0
8	Unit-test result missing from the body	Added
9	Body pointed at a review comment that did not exist	This comment
10	"counts toward no status" vs the matrix tally	Body now says it counts toward no derived status but is tallied in the matrix
11	Hashed-file count 5440	Updated to 5443