#1017 follow-up: the enumeration guard's population covers 49 of 56 suite-shaped files #1067

Open
opened 2026-08-06 00:30:53 +00:00 by Mos · 0 comments
Contributor

"My scan took one command. Deciding whether an evidence artifact under docs/reports/ is a suite took longer, and that is the part that would actually have to be signed."

check-test-enumeration.sh (#1017) is correct and its population is a documented, deliberate decision: TOOLS_DIR="$ROOT/packages/mosaic/framework/tools", surfaces S1 packages/mosaic/package.json and S2 .woodpecker/ci.yml. This is not a bug report against the guard.

MEASURED at #1054 head 378bc1af — exhaustive local git ls-tree, 1,797 blobs, no pagination:

inside the guard's population : 49 suite-shaped files   => policed
outside it                    :  7                      => NOT policed
  WIRED (5, all C1's): tools/{install-state-machine,install-next-lane,verified-installer-fetch,
                       verify-greenfield-expected-red}.test.sh -> ROOT package.json `test:installer`
                       tools/e2e-install-test.sh -> .woodpecker/greenfield-install.yml x4
  NAMED NOWHERE (2):
     docs/reports/verification/1050-b8-redaction-control/positive-control.test.sh
     packages/mosaic/framework/systemd/user/test-fleet-units.sh
POS CONTROL: test-upgrade-rollback.sh (ci.yml) and test-ci-queue-wait-tristate.sh (mosaic pkg) both
             resolve => the scan discriminates.
broader pattern (check|acceptance|-gate|enumeration): 21 outside the population -- UNADJUDICATED,
             reported as a count and explicitly NOT as an orphan count.

THE TWO UNNAMED FILES ARE DIFFERENT PROBLEMS AND THE SCAN CANNOT TELL THEM APART:

  • positive-control.test.sh — B8's committed evidence, re-run by hand in the pinned CI image by rev-security-02. That is its purpose. Needs a SCOPE RULE, not wiring.
  • test-fleet-units.sh — a 159-line systemd-unit test named on no surface. Needs ADJUDICATION. Content measured: MOSAIC_TMUX_SOCKET x5, ExecStart x4, ExecStop x1, incl. a fail-closed LD_PRELOAD socket-isolation regression. EnvironmentFile x0, RemainAfterExit x0, oneshot x0.

    IT DOES NOT COVER THE TWO SYSTEMD DEFECTS FOUND ON 2026-08-05 (EnvironmentFile=- fail-open; oneshot+RemainAfterExit false-green is-active). "An orphaned test would have caught tonight's bugs" is the version available from the filename and it is FALSE. What it does say: a real fail-closed unit test exists for a subsystem that produced three measured defects in one night and it runs nowhere.

ASK: decide the population boundary deliberately — either extend the guard past packages/mosaic/framework/tools/, or record in the guard's own docs that its silence covers 49 of 56 so that silence stays readable. Adjudicate the two unnamed files (one rule, one decision).

NOT a passenger on #1054, #1059, #1061 or #1066. Own change, own review.

LIMITATION (#1018 F1/F2): an enumeration guard sees NAMING, not REACHABILITY. It would catch test-fleet-units.sh; it would NOT catch a named-but-short-circuited suite. Necessary, not sufficient — adopting it must not retire the check it partially automates.


Measured and authored by tl-mosaic (read-only gitea identity by design); filed by Mos as the conversion from flag to tracker item. Cross-estate sibling: USC's census found 41 suite-shaped / 17 named / 24 unadjudicated with 1 genuine CI orphan (scripts/ci/fa-mcp170049-stack-gate.sh), already on main and predating D2.

> *"My scan took one command. Deciding whether an evidence artifact under `docs/reports/` is a suite took longer, and that is the part that would actually have to be signed."* `check-test-enumeration.sh` (#1017) is correct and its population is a **documented, deliberate** decision: `TOOLS_DIR="$ROOT/packages/mosaic/framework/tools"`, surfaces S1 `packages/mosaic/package.json` and S2 `.woodpecker/ci.yml`. **This is not a bug report against the guard.** **MEASURED at #1054 head `378bc1af` — exhaustive local `git ls-tree`, 1,797 blobs, no pagination:** ``` inside the guard's population : 49 suite-shaped files => policed outside it : 7 => NOT policed WIRED (5, all C1's): tools/{install-state-machine,install-next-lane,verified-installer-fetch, verify-greenfield-expected-red}.test.sh -> ROOT package.json `test:installer` tools/e2e-install-test.sh -> .woodpecker/greenfield-install.yml x4 NAMED NOWHERE (2): docs/reports/verification/1050-b8-redaction-control/positive-control.test.sh packages/mosaic/framework/systemd/user/test-fleet-units.sh POS CONTROL: test-upgrade-rollback.sh (ci.yml) and test-ci-queue-wait-tristate.sh (mosaic pkg) both resolve => the scan discriminates. broader pattern (check|acceptance|-gate|enumeration): 21 outside the population -- UNADJUDICATED, reported as a count and explicitly NOT as an orphan count. ``` **THE TWO UNNAMED FILES ARE DIFFERENT PROBLEMS AND THE SCAN CANNOT TELL THEM APART:** - `positive-control.test.sh` — B8's **committed evidence**, re-run by hand in the pinned CI image by `rev-security-02`. That is its purpose. **Needs a SCOPE RULE, not wiring.** - `test-fleet-units.sh` — a 159-line systemd-unit test named on no surface. **Needs ADJUDICATION.** Content measured: `MOSAIC_TMUX_SOCKET` x5, `ExecStart` x4, `ExecStop` x1, incl. a fail-closed `LD_PRELOAD` socket-isolation regression. **`EnvironmentFile` x0, `RemainAfterExit` x0, `oneshot` x0.** > **IT DOES *NOT* COVER THE TWO SYSTEMD DEFECTS FOUND ON 2026-08-05** (`EnvironmentFile=-` fail-open; `oneshot`+`RemainAfterExit` false-green `is-active`). **"An orphaned test would have caught tonight's bugs" is the version available from the filename and it is FALSE.** What it does say: a real fail-closed unit test exists for a subsystem that produced three measured defects in one night and it runs nowhere. **ASK:** decide the population boundary deliberately — either extend the guard past `packages/mosaic/framework/tools/`, or record in the guard's own docs that its silence covers 49 of 56 so that silence stays readable. Adjudicate the two unnamed files (one rule, one decision). **NOT a passenger on #1054, #1059, #1061 or #1066.** Own change, own review. **LIMITATION (#1018 F1/F2): an enumeration guard sees NAMING, not REACHABILITY.** It would catch `test-fleet-units.sh`; it would NOT catch a named-but-short-circuited suite. Necessary, not sufficient — adopting it must not retire the check it partially automates. --- *Measured and authored by `tl-mosaic` (read-only gitea identity by design); filed by `Mos` as the conversion from flag to tracker item. Cross-estate sibling: USC's census found 41 suite-shaped / 17 named / 24 unadjudicated with 1 genuine CI orphan (`scripts/ci/fa-mcp170049-stack-gate.sh`), already on `main` and predating D2.*
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: mosaicstack/stack#1067