[InauguralPhysicist]

What does this PR do?

tests/test_lab.sh claimed to check that the critically damped and overdamped rows of the physics.eigs zeta-sweep (zeta=1, zeta=2) do not oscillate. What it actually did was count lines containing settles. Every underdamped row reads x OSCILLATES, energy settles, so the count was always at least 1, and the gate stayed green even when the zeta>=1 rows oscillated.

This PR replaces that count with a row-by-row check:

• tests/check_physics_sweep.py (new) parses the sweep table and makes five checks:
  1. The zeta rows are exactly -0.05, 0, 0.1, 0.4, 1, 2, in that order.
  2. Each row has its required reading. zeta<0 must read DIVERGES (energy grew), 0<=zeta<1 must read x OSCILLATES, energy settles, and zeta>=1 must read exactly settles.
  3. The energy-grew flag is Y exactly when the reading is DIVERGES.
  4. When energy did not grow, the x:osc flag is Y exactly when the reading is x OSCILLATES.
  5. The zeta=0 E:conv flag is -. This is the lab's stated claim: at zeta=0 energy is conserved, so it never converges.
• tests/test_lab.sh: the grep -c settles lines are replaced by a call to the checker. The DONE, DIVERGES and x OSCILLATES greps are kept, so no existing assertion is removed.
• tests/test_physics_sweep_plants.sh (new) is a CI plant harness:
  • The unmodified physics.eigs must pass first.
  • Then six planted faults are each applied to a copy of physics.eigs, at least one per check.
  • Each plant must change exactly 1 line, so a plant that stops applying fails the harness.
  • The planted program must still exit 0 and print DONE, so a crash does not count as a rejection.
  • The checker must reject each plant with the exact message of the check that plant targets. A plant rejected for any other reason fails the harness.
• .github/workflows/test.yml: one new row in runCmd runs the plant harness. This is unavoidable, because runCmd is the only place tests are enrolled. python3 is already used by test_lab.sh (check_solver_output.py on main), so the devcontainer image needs nothing new.

There is no changes/ fragment. This PR changes only tests and CI, and dynamics CI has no fragment check (#55 landed without one).

Checklist

• Every changed/added specimen parses and runs (eigenscript <file>.eigs). No specimen changed. physics.eigs runs unmodified and under every plant.
• bash tests/test_smoke.sh and bash tests/test_lab.sh pass locally (pinned v0.48.0)
• Observer usage is load-bearing. Not applicable: no specimen changed.
• Any runtime gap or surprising predicate behavior is logged in FINDINGS.md. None was found; the defect is in the test.
• eigs.json version bumped if the importable surface changed. Not applicable: the surface is unchanged.

Finding

• DYN-LAB-1. The physics zeta-sweep overdamped check passes even when the zeta>=1 rows oscillate.
• Location: tests/test_lab.sh:16-18 (origin/main). The sweep rows are printed at physics.eigs:116-126. The E:/x: flag columns were never checked.
• Tracking: https://github.com/InauguralSystems/hq/issues/58 ("treat CI as a product", Wave 2).

Plant: red before, green after

All runs below use the pinned runtime, EigenScript v0.48.0 (546848c0, the .devcontainer/Dockerfile EIGS_REF). "Old gate" is git show origin/main:tests/test_lab.sh.

Plant A is the Wave-1 reproduction. It is a 1-line change to physics.eigs:122:

122c122
<     elif p[3] == 1:
---
>     elif (p[3] == 1) or (z >= 1.0):

Old gate, rc=0. It lies: both zeta>=1 rows read oscillating and the gate still prints PASS.

  1	| E:- Y - -     | x:- Y -      | x OSCILLATES, energy settles
  2	| E:- Y - -     | x:- - -      | x OSCILLATES, energy settles
PASS: physics zeta-sweep produces the expected regime spectrum

New gate, EIGENSCRIPT=$E bash tests/test_lab.sh, rc=1:

FAIL: physics sweep: zeta=1 reads 'x OSCILLATES, energy settles', expected 'settles'

All six plants, old gate compared with new gate. Each plant changes 1 line.

plant	old gate	new gate
overdamped-reads-oscillating (A)	rc=0 PASS	rc=1 zeta=1 reads 'x OSCILLATES, energy settles', expected 'settles'
x-osc-false-positive (or (zeta >= 1.0) on the x observer)	rc=0 PASS	rc=1 same message
overdamped-row-dropped (2.0 removed from the sweep)	rc=0 PASS	rc=1 expected zeta rows ['-0.05', '0', '0.1', '0.4', '1', '2'], got ['-0.05', '0', '0.1', '0.4', '1']
x-flag-column-desync (x:osc printed from p[4])	rc=0 PASS	rc=1 zeta=0 x:osc flag '-' contradicts reading 'x OSCILLATES, energy settles'
grew-column-desync (grew printed from p[5])	rc=0 PASS	rc=1 zeta=-0.05 energy-grew flag '-' contradicts reading 'DIVERGES (energy grew)'
zeta0-energy-converged (E:conv printed from p[0])	rc=0 PASS	rc=1 zeta=0 energy reported converged; the undamped row must conserve energy

Green after the fix. On the unmodified tree, EIGENSCRIPT=$E bash tests/test_lab.sh gives rc=0:

PASS: zeta=-0.05 reads 'DIVERGES (energy grew)' (grew=Y, x:osc=Y)
PASS: zeta=0 reads 'x OSCILLATES, energy settles' (grew=-, x:osc=Y)
PASS: zeta=0.1 reads 'x OSCILLATES, energy settles' (grew=-, x:osc=Y)
PASS: zeta=0.4 reads 'x OSCILLATES, energy settles' (grew=-, x:osc=Y)
PASS: zeta=1 reads 'settles' (grew=-, x:osc=-)
PASS: zeta=2 reads 'settles' (grew=-, x:osc=-)
PASS: zeta=0 energy never converges while x oscillates
PASS: physics zeta-sweep produces the expected regime spectrum

EIGENSCRIPT=$E bash tests/test_physics_sweep_plants.sh gives rc=0:

PASS: unmodified physics.eigs accepted
PASS: plant overdamped-reads-oscillating rejected: FAIL: physics sweep: zeta=1 reads 'x OSCILLATES, energy settles', expected 'settles'
PASS: plant x-osc-false-positive rejected: FAIL: physics sweep: zeta=1 reads 'x OSCILLATES, energy settles', expected 'settles'
PASS: plant overdamped-row-dropped rejected: FAIL: physics sweep: expected zeta rows ['-0.05', '0', '0.1', '0.4', '1', '2'], got ['-0.05', '0', '0.1', '0.4', '1']
PASS: plant x-flag-column-desync rejected: FAIL: physics sweep: zeta=0 x:osc flag '-' contradicts reading 'x OSCILLATES, energy settles'
PASS: plant grew-column-desync rejected: FAIL: physics sweep: zeta=-0.05 energy-grew flag '-' contradicts reading 'DIVERGES (energy grew)'
PASS: plant zeta0-energy-converged rejected: FAIL: physics sweep: zeta=0 energy reported converged; the undamped row must conserve energy
PASS: every physics-sweep plant is rejected

Every check is load-bearing. In scratch copies, each of the checker's five checks was disabled in turn (its if replaced by if False:), and the checker was also gutted with sys.exit(0). The plant harness then gave:

check removed	harness rc	first failure
reading	1	plant overdamped-reads-oscillating rejected for the wrong reason
grew consistency	1	plant grew-column-desync rejected for the wrong reason
x:osc consistency	1	plant x-flag-column-desync was accepted
row list	1	plant overdamped-row-dropped was accepted
zeta=0 E:conv	1	plant zeta0-energy-converged was accepted
whole checker gutted	1	plant overdamped-reads-oscillating was accepted

Checks run

All locally, on pinned v0.48.0, covering the full CI runCmd:

• test_lint rc=0
• test_smoke rc=0
• test_lab rc=0
• test_physics_sweep_plants rc=0
• test_orbit_hist rc=0
• test_bifurcation rc=0
• test_orbit_oracle rc=0 (Xvfb)
• test_orbit_mouse rc=0
• test_bif_mem rc=0
• bash -n and py_compile pass on the new and changed files.

The added CI time is about 0.3s (the plant harness reruns physics.eigs 7 times). For context, the "suite (in devcontainer)" job takes 139s p50 and 155s p90 (n=40, 2026-10-04..2026-11-01). This PR is not a speedup.

Coverage goes up and nothing is removed. One weak count assertion becomes 6 row checks plus 1 claim check, and 6 plants are added.

Done when

• On this PR's CI, the suite (in devcontainer) log shows PASS: every physics-sweep plant is rejected and PASS: zeta=2 reads 'settles' (grew=-, x:osc=-)
• Applying the plant-A line to physics.eigs makes EIGENSCRIPT=eigenscript bash tests/test_lab.sh exit 1 with FAIL: physics sweep: zeta=1 reads 'x OSCILLATES, energy settles', expected 'settles'
• Replacing the rows[1][2] != "-" condition in tests/check_physics_sweep.py with False makes bash tests/test_physics_sweep_plants.sh exit 1

Overlaps

There are no open PRs in dynamics (GET /repos/InauguralSystems/dynamics/pulls?state=open returned nothing).

Not done in this PR:

• dynamics has no required status check (DYN-REPO-1). Until one is required, this gate cannot block a merge. Requiring a check is a settings change, so it is out of scope for this PR.
• There is no dynamics issue for DYN-LAB-1, so this PR references hq#58.
• The flag columns that the lab makes no claim about are deliberately not pinned. These are E:osc/E:div, and x:conv/x:div outside the checks above. They follow observer-predicate semantics, and pinning them would force a re-baseline at every pin bump.

---

Generated by Claude Code

[InauguralPhysicist]

@codex review

---

Generated by Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-01T21:46:11.963266Z	6ea7dd8	Draft marked ready
🔒 Security Review	✅ Completed 2026-11-01T21:46:52.678280Z	6ea7dd8	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep them coming!

Reviewed commit: 6ea7dd8b80

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".