[arcaven]

Two seats writing a finding at once can take the same hand-picked number. Minted ids come after finding 7, which closes that, so every new finding gets a collision-free id from kos id finding <slug>.

• _kos/kos.yaml: a findings: block with numbered_through: 7 (the highest numbered finding on main) and the adoption date, 2026-11-02.
• Existing findings keep their names and citations.

In this worktree, kos id finding x prints finding-sideshow-trbg (no file kept). kos validate reports 7 findings with no duplicate ids and 30 nodes with no failures. CLAUDE.md already carries the mint rule, so it is unchanged.

Cost of merge: additive only; one block in kos.yaml.

[arcavenai]

APPROVE at 6980817. The findings: block marks finding-007 as the last hand-numbered finding and leaves the existing graph fields alone. kos then mints finding-sideshow-... ids with validate unchanged.

The manifest (_kos/kos.yaml:5-10):

• numbered_through: 7. On main (e3c702e), _kos/findings/ holds finding-001 to finding-007 with no gap, and no open PR adds a finding.
• graph_id, scope, description and schema_version (:1-4) and includes (:11) are untouched. The diff is +6 -0.
• The comment (:6-9) is the same form, word for word and dated 2026-11-02, as ArcavenAE/callbook#17.

CLAUDE.md is unchanged, and that is right. Step 5 of its Session Protocol (CLAUDE.md:20) already reads "Write a finding in _kos/findings/. Mint the id with kos id finding <slug>; never hand-allocate a number.", which is the same rule.

kos runs, at this head and at main as the control:

• kos id finding x at head printed finding-sideshow-g5er and finding-sideshow-g5er-x.md, exit 0, and git status stayed clean, so no file was kept. At main it printed finding-sideshow-qlas.
• kos validate at head: exit 0. 30 nodes: 11 passed, 19 warnings, 0 failed. 7 findings, 0 duplicate-id failures. At main the output is byte-identical.

The body as it stands says "Minted ids come after finding 7", which matches the manifest.

Format and text:

• The commit is signed and verified on GitHub, with one parent (e3c702e). CI is clean: the release jobs are skipped and everything else succeeds.
• The subject's one em dash is the separator of the kos action-commit format. There is no other em dash, banned word, AI attribution, private name, host name or non-ASCII byte in the diff, the message or the body. The scan's only other hit is the ai style check, on CLAUDE.md.
• The body leads with why.

Run conditions for the kos commands:

• kos alpha-20261002 (c4f93c2), in scratch clones under env -i with a scratch HOME and TMPDIR, and no MARVEL_SOCKET.
• Network: a sandbox-exec profile denied the network except localhost. Controls: a connect to 1.1.1.1:80 was refused with "Operation not permitted", and a localhost connect gave connection refused.
• Writes: the same profile denied writes outside the scratch dirs and /dev. Controls: a write to a fixed /private/tmp path outside scratch was refused with "Operation not permitted", and a scratch write succeeded.
• Everything else was read only.

Seat: reviewer / reviewer-claude-g17-1