Torque & Tokens Radar · 100x Engineering · Lesson 07 exercise

Checks a reviewer can actually rerun

ILLUSTRATIVE ARTIFACTS · invented part BRK-2210, ECR-0417, names and numbers

A front radar sensor gets a larger housing, so its bracket goes from rev C to rev D. The pinned baseline is BRK-2210 rev D · BRK-2210-D/as-designed · MODAL+SHOCK deck v4 · mm-t-s (N, mm, MPa) · TP-88 v2. A generator agent already wrote the summary below. Run the independent checker, then fix inputs and record the judgment the way a real team would.

Generator agent summary (do not trust, check)

1 · Fix the inputs

Re-export the CAD clearance at rev D, config BRK-2210-D/as-designed (the delivered export is rev C from 21 Oct) Put the shock load case in the deck's mm-t-s unit system and re-run the deck (delivered load case is SI: N, m, Pa)

2 · Record the engineering judgment (REQ-ENV-044)

The vibration report TR-5531 was run on rev C hardware. Whether it is representative of rev D is a judgment, not a computation. The reviewer named in the map row decides, with a rationale, a date, and the exact evidence the decision covers.

decision decided_by rationale date bound_to.part bound_to.baseline_revision bound_to.report_id bound_to.tested_revision

Training limit: A real release gate takes the name and signature from your PLM or e-signature records.

3 · Try to break it

Count each run as a rework attempt (bounded rework: after 2 failed attempts a branch escalates to its named reviewer)

What to notice

Every branch compares part, revision, CAD configuration, deck, unit system and test plan against the pinned baseline, field by field, and shows the comparison. A matching revision with the wrong part still stops.
The 0.16 MPa "pass" in the summary is the unit error. After the fix and re-run, the illustrative result is 148.6 MPa: still under 180 MPa, with much less margin.
Clearance passes at 2.1 mm, close to the 2.0 mm limit. Your report should say so.
A rev C test is not automatically rejected. It needs an explicit, complete decision by the named owner, bound to TR-5531, tested rev C and baseline rev D. Change the report or the baseline and the decision goes stale.
No setting lets the tool approve the release. A pre-set approval holds the join.

The same rules run in a terminal with python3 verify.py, and python3 -m unittest tests/test_verify.py checks both versions against each other. See README.md.

Concept credit: verification before parallelism, from Lauren Tan (Cursor) with Colin Matthews, Maven workshop, 12 Aug 2026. The engineering scenario is our illustrative application, not something shown in the workshop.