macOS arm64 SYN evidence for the adaptive-paper engine and mover trial mode - #176

…ine)

On macos-m5pro-20260924, from a read-only git archive of main 6f7a77c, in
the hash-locked CPython 3.12.14 runtime built from the #176 lock:

- Trial a (config-sip.json, 300 s, 1x, SIP, one-share orders) passed. It
  made 33 submits and 3 cancel-replaces, and got 30 native fills in 15 round
  trips. It reconciled flat, with realized P&L of -1.51 USD.
- STOP drill: the working entry was canceled 115 ms after STOP, with 0
  filled, and no order followed. recover passed flat, and STOP was archived
  with a release record.
- Native-faults harness: C01 and C02 passed with native_paper evidence. C04
  and C05 are still blocked on #169.
- Host receipt alpaca-py/use (native_proven, second physical machine).
- Rebuilt the rate-limit evidence for the new trial headers (trading 419 to
  641, data 9 to 13) and updated the pinned test totals and README row to
  match.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

…ine) (#188)

On macos-m5pro-20260924, from a read-only git archive of main 6f7a77c, in
the hash-locked CPython 3.12.14 runtime built from the #176 lock:

- Trial a (config-sip.json, 300 s, 1x, SIP, one-share orders) passed. It
  made 33 submits and 3 cancel-replaces, and got 30 native fills in 15 round
  trips. It reconciled flat, with realized P&L of -1.51 USD.
- STOP drill: the working entry was canceled 115 ms after STOP, with 0
  filled, and no order followed. recover passed flat, and STOP was archived
  with a release record.
- Native-faults harness: C01 and C02 passed with native_paper evidence. C04
  and C05 are still blocked on #169.
- Host receipt alpaca-py/use (native_proven, second physical machine).
- Rebuilt the rate-limit evidence for the new trial headers (trading 419 to
  641, data 9 to 13) and updated the pinned test totals and README row to
  match.

Co-authored-by: seathatflowsinourveins <[email protected]>
Co-authored-by: Claude Opus 5.5 (1M context) <[email protected]>

Measure the Mac first seen in #176 with scripts/hardware_profile.py
--record-host (Apple M5 Pro, 18 cores, 24 GB unified memory, arm64) and
record native_proven use-stage receipts for rtk 0.49.0 (rtk git status),
worktrunk 0.79.0 (wt list) and gitleaks 8.30.1 (one-commit scan) with
scripts/host_receipts.py record. Every command ran outside the Claude Code
sandbox, in a scratch clone of main. The receipts await an independent
review, and macos-arm64 stays drafted_not_accepted.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

…runk, gitleaks (#185)

* Record macos-m5pro-20260924 hardware profile and foundation use receipts

Measure the Mac first seen in #176 with scripts/hardware_profile.py
--record-host (Apple M5 Pro, 18 cores, 24 GB unified memory, arm64) and
record native_proven use-stage receipts for rtk 0.49.0 (rtk git status),
worktrunk 0.79.0 (wt list) and gitleaks 8.30.1 (one-commit scan) with
scripts/host_receipts.py record. Every command ran outside the Claude Code
sandbox, in a scratch clone of main. The receipts await an independent
review, and macos-arm64 stays drafted_not_accepted.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

* Regenerate evidence reports and register macos-m5pro-20260924 evidence

Rerun scripts/component_matrix.py --write and
scripts/new_host_grand_list.py --write for the new receipts and hardware
profile. The manifests/evidence.json entries come from the recording and
generation tools. Shared hot-file edits are in the last commit, following
the proposed docs/lanes.md hot-file protocol (#180).

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

* Add independent reviews to the macos-m5pro-20260924 foundation receipts

An independent Claude Code session (a different session identity on the
same Mac) reproduced the recorded commands at bc8cee5 and appended step-8
reviews:

- rtk use receipt: agree. rtk git status proxies a real command and
  returns compacted output.
- gitleaks use receipt: agree. A real redacted scan of one commit with
  the repository config.
- worktrunk use receipt: needs_changes. Beyond wt --version, the only
  command is the read-only wt list. Nothing is created, switched or
  removed, so the worktrees layer's job is not shown. A fresh receipt
  with a create/list/remove cycle in a throwaway repository with
  reviewed hooks addresses it.

With the reviews, verdict_flip_candidates.py reports untested -> accepted
support on macos-arm64 for secrets-credentials/gitleaks,
security-supply-chain/gitleaks and token-efficiency/rtk. Recording the
flips remains a separate record_verdicts.py step.

Regenerated component-evidence-matrix and new-host-grand-list;
host_receipts validate (8), validate.py, component_matrix --check (0
flip-rule violations), new_host_grand_list --check, evidence_manifest
--check and validate_convergence --all-recorded pass.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

* Re-record macos-m5pro-20260924 worktrunk use receipt with a worktree lifecycle

The independent review at cef7a74 marked the worktrunk use receipt
needs_changes: it ran only wt --version and the read-only wt list. This
fresh recording runs a full lifecycle in a throwaway one-commit git
repository under $TMPDIR: wt switch --create wt-probe, wt list (two
worktrees), wt remove --foreground wt-probe, then wt list, git worktree
list, git branch --list and ls (only main left), and cleanup. All 8
commands exit 0.

Worktrunk hooks were inspected first. There is no user config
(~/.config/worktrunk absent: no user hooks, approvals or custom
worktree-path), no system config, no WORKTRUNK_* override and no project
.config/wt.toml. wt ran with stdin from /dev/null and without --yes.

The recorder derives the id from host, component, stage and UTC date,
so this same-day re-record replaces the file under the same id. The
list-only version and its needs_changes review remain in history at
cef7a74, and the new receipt says so in its limitations.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

* Regenerate evidence matrix and re-register the re-recorded worktrunk receipt

Rerun scripts/component_matrix.py --write and
scripts/new_host_grand_list.py --write (the grand list is unchanged) after
the worktrunk re-record. The manifests/evidence.json hashes come from the
recording and generation tools. Shared hot-file edits stay in the last
commit.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

* Re-record macos-m5pro-20260924 rtk, gitleaks and worktrunk use receipts per re-review

Address the re-review of 3772eb5 (PR #185 comment 5816561592) by
re-recording all three receipts in place. The recorder derives the id from
host, component, stage and UTC date, so each same-day re-record reuses its
id. The earlier versions and the rtk/gitleaks agree reviews remain in
history at cef7a74, and each receipt's limitations say so.

- All three: the install and pin sentence is now backed by retained
  commands. command -v <tool> is recorded with the pin source: the mise
  global config pin line for rtk (0.49.0) and gitleaks (8.30.1), and the
  host control plane's manifest.json for worktrunk (0.79.0), read through
  a CONTROL_PLANE_ROOT environment variable so the private path is not
  recorded. The limitation now reads "Supports conditional now; accepted
  needs an independent review."
- gitleaks: add a positive control. A throwaway repository commits
  gitleaks' documented sidekiq-secret test fixture, built at run time from
  split pieces so the value never appears in the receipt. The same scan
  exits 1 with exactly one finding, and the wrapper asserts both. The
  negative controls remain: the scratch clone's latest commit and the
  throwaway repository's clean commit, both exit 0.
- rtk: add a limitation that raw-output and failure-recovery behavior were
  not exercised.
- worktrunk: same lifecycle probe, with the pin evidence and new wording.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

* Regenerate evidence reports after re-recording the macos-m5pro-20260924 receipts

Rerun scripts/component_matrix.py --write and
scripts/new_host_grand_list.py --write. The re-recorded receipts start
without independent reviews, so the macOS state of rtk and gitleaks
returns from host_verified to untested until the requested re-review. The
manifests/evidence.json hashes come from the recording and generation
tools. Shared hot-file edits stay in the last commit.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

* Add independent agree reviews to the re-recorded macos-m5pro-20260924 receipts

A separate Claude Code session on the same Mac re-reviewed the three
receipts re-recorded at 19354be and reproduced every probe from its own
session:

- rtk: command -v with the mise pin line, rtk 0.49.0, and rtk git status
  compacted output. Agree.
- gitleaks: command -v with the mise pin line, 8.30.1, and both negative
  controls (no leaks). The sidekiq-secret positive control exited 1 with
  exactly one finding. Agree.
- worktrunk: no user, system, project or environment config, so no hooks
  and no --yes; the agent-ecosystem manifest pins 0.79.0; and the
  wt switch --create / list (2) / remove --foreground / list (1)
  lifecycle ran in a throwaway repository, then cleanup. Agree.

The claim text is now evidence-backed, the limitation wording is
corrected, and rtk's unexercised raw/failure-recovery behavior is
stated. With these reviews, verdict_flip_candidates.py reports untested
-> accepted support on macos-arm64 for worktrunk (git-github-automation,
isolation, workers), gitleaks (secrets-credentials,
security-supply-chain) and rtk (token-efficiency). Recording the flips
is a separate record_verdicts.py step by the owning lane.

host_receipts validate (8), validate.py, component_matrix --check,
new_host_grand_list --check, evidence_manifest --check and
validate_convergence --all-recorded pass.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

* Record agent-lab-e9's independent agree reviews on the macos-m5pro-20260924 use receipts

Independent review from the WSL authoring host (different machine and identity):
workflow wf_b3b5c3b4-c85 at bc8cee5 (14 agents, adversarially verified), plus a
re-check at cfa7574 of every requested fix: command -v plus mise/manifest pin lines,
the gitleaks sidekiq positive control (exit 1, one finding), the worktrunk
create/list/remove/list lifecycle, and corrected claim/limitation text.
Validators exit 0.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

* Re-record the macos-m5pro-20260924 use receipts from main commit 2d11464

The previous copies recorded catalog_revision 3772eb5, a branch-only commit
that stops being reachable once this PR is squash-merged and its branch
deleted. host_receipts.py validate would then fail. These copies were
re-recorded with the same ids from a detached scratch checkout of main at
2d11464, as adoption/update.md requires, so catalog_revision is a main
commit.

The commands, claims and limitations are unchanged apart from one sentence
appended to each history limitation that records this re-record. The fresh
receipts carry only their self entries. The independent agree reviews in
cfa7574 and aebc492 were bound to the replaced records, so they are
superseded and are not carried over; they remain in PR #185's history.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

* Hot-file protocol: take main's evidence.json, re-register this PR's evidence, regenerate reports

After rebasing onto main 2d11464, manifests/evidence.json is main's copy
with this PR's four evidence files re-registered by the repository's
register_file tool: the hardware profile and the three re-recorded use
receipts. component_matrix.py --write and new_host_grand_list.py --write
regenerate the derived reports. register_file serializes with ASCII
escapes, so one existing non-ASCII path that main stores literally is
re-encoded to the same JSON value. No other entry changes.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

* Add independent agree reviews to the main-bound macos-m5pro-20260924 receipts

The rtk, gitleaks and worktrunk use receipts were re-recorded from main
commit 2d11464 so that catalog_revision stays reachable after a squash
merge. This independent session (a different identity from the recorder,
on the same Mac):

- confirmed 2d11464 is an ancestor of origin/main;
- found commands, claims and results identical to the ea31d6f copies it
  reproduced earlier. Only the history limitation gained a re-record
  sentence;
- re-ran the gitleaks probe: the clean commit exits 0, and the
  sidekiq-secret fixture exits 1 with leaks found: 1.

Agree on all three. verdict_flip_candidates.py reports untested ->
accepted support on macos-arm64 for worktrunk (git-github-automation,
isolation, workers), gitleaks (secrets-credentials,
security-supply-chain) and rtk (token-efficiency). host_receipts
validate (9), validate.py, component_matrix --check,
new_host_grand_list --check and evidence_manifest --check pass.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

* Re-record agent-lab-e9's independent agree reviews on the main-bound macos-m5pro-20260924 receipts

The receipts were re-recorded from main 2d11464 (so catalog_revision survives
a squash merge), which reset their reviews. Commands, exits, claim, class and
host are identical to the copies reviewed at cfa7574; only a history
limitation sentence was added.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

---------

Co-authored-by: seathatflowsinourveins <[email protected]>
Co-authored-by: Claude Opus 5.5 (1M context) <[email protected]>

… parser; currency checks from the merge (#1027 round 3)

Round 3 of #1027, one commit on 37e242c. Its inputs hold the reproductions, which are not repeated here: the CC's Claude
micro at 37e242c (W1, W2, C1, P3-A to P3-C), the GPT family's fix3 verdict at 37e242c (blockers 1 to 3, follow-ups 2
and 3), the Codex connector's three threads on e203d01 (nas:430, nas:259, the address rule) and the CC's rulings of
2026-10-12 on the allowances and on the host-name fix. Each change is stated by what it guarantees.

- No address passes for want of a pattern that finds it. squash_body_scan.py no longer searches for an address shape: on
  each line it judges every at sign with CPython's get_addr_spec (Lib/email/_header_value_parser.py at v3.13.16,
  cbc944f4bc59), reading the local part before it and the domain after it as the parser does, with quoted strings,
  comments and white space around the at sign, domain literals and domains of any form (RFC 5322 3.4.1,
  rfc5322.txt:959-969). An at sign inside a quoted string is judged inside it too. One with no local part before it or
  no domain after it is no address, and text the parser cannot read whole and without a defect is refused.
- An allowance passes only a whole address that holds no other address and that no other touches.
- By the CC's ruling, two shapes are not addresses: a source locator, a repository path, an at sign and a git object
  name of 7 to 64 hex digits as the whole domain (git v2.53.0 Documentation/revisions.adoc:14-16, object-name.h:137,
  hash.h:201-208; this repository's locator form, CLAUDE.md:7), and the one co-author address of this repository's
  synthetic scratch identity, cited to its source in the repository. A tag, a branch and every other domain without a
  dot stay refused.
- A body whose only hit is the host's own name lands with the warning on any host, and an identity spelled as the host
  is still refused: the scripts run validate.py --scan-file with NATIVE_AGENT_PUBLIC_HOSTNAME=1, a new opt-in of
  scripts/host_name_scan.py that drops the hostname source, as that file does on a WSL host whose hostname is its
  WSL_DISTRO_NAME, and a Windows computer name of the same spelling. Nothing else sets it.
- No listed identity goes unchecked for text that a shell and the scan would read differently: the names file's value
  is read only as a bare list or a list in one pair of quotes, and any other form fails closed.
- A merge is validated with the shard-0 checks that main's CI will run on it: nas_land.sh holds no list any more and
  reads the steps from the merge's own .github/workflows/validate.yml, in the workflow's order (GitHub Docs at
  be38ec5d, data/reusables/actions/about-triggers.md:18: a run uses the workflow of the event's commit). The reader
  fails closed on a validate.yml in which it reads no check and on a step whose words are not plain ones (exit 10, the
  step's text not printed); a merge with no validate.yml runs none.
- The scan's cost is linear in the text: 60 KB on one line takes at most 0.18 s, where round 2's rule took 24 s on one
  of the measured texts.
- README: the signals after which the snapshot is removed are named (SIGINT, SIGTERM and SIGHUP, each tested; SIGKILL
  leaves it), #975's cli/cli row names the mirror that holds the pinned commit, and the sources, known gaps and tests
  sections follow the change. docs/decisions/2026-10-09-host-name-scan.md gains one sentence on the opt-in.

Tests: on a git archive of 37e242c with this commit's two test files, the 19 new or changed tests give 2,254 failing
cases and one error in 16 tests (100 landings of a refused address, 12 landings past a names file read otherwise, a
landing with a failing check that the merge added); the three that pass there are the controls' test, the signal test
and #176's own. Here all 19 pass, on CPython 3.12.3, 3.13.16 and 3.14.4, and the landing module's 150 tests pass. The
oracle's test runs 29 local parts at 29 domains in nine forms (7,569 cases).

Measured on the last 600 commit messages of main at 5421e5d (counts only): 70 refused, 0 names, 67 addresses and
3 home paths (round 2's rule: 8); 341 messages with the scratch co-author trailer and 62 with a source locator pass.
The CC's own regression for #176 needs a validate.yml in its scratch repository to pass on this merge_validate.

Co-Authored-By: Claude Opus 5.5 <[email protected]>