[aarontrowbridge]

What

Three loop-methodology skill cards updated with measured laws from the diraq-esr-realism research campaign (2026-10-22/23, loops 3–10 of ledger/campaigns/campaign-20261022-diraq-esr-realism.md in the campaign repo):

debugging — four new Solver Diagnosis Patterns:

• The Vacuous Solve (silent control-drop): an optimizer that runs thousands of iterations, moves variables, and leaves the objective bit-identical to the seed across every option and formulation change is decoupled from its dynamics — Piccolo 2.2.0's template integrator auto-resolution measured doing exactly this. Countermeasure: evaluate the objective at the seed and assert it MOVED; pass integrators explicitly; trust independent re-rollouts, never internal readouts (a stale fidelity(qtraj) masks it in the other direction).
• The Silent Edit — six measured instances of no-op transformations (str.replace/sed succeed on no-match by design; swallowed kwargs; pattern-revert stacking). Every edit asserts its pattern matched and grep-verifies the critical line; after multi-edit confusion, rewrite clean with git as arbiter.
• The Silent OOM — death right after optimizer initialization with no error = the OOM killer; dense unitary-trajectory Hessians scale as (2·d²·N)².
• Background reaping — plain nohup dies at sandboxed tool-call exit; setsid survives (measured both ways).

research — the anti-gaming contract gains wiring-gate-per-axis (three measured instances of unwired axes banking wrong verdicts with all gates green, plus the symmetric ±σ telegraph free-validator idiom) and director-verifies-analyzer; Failure modes gains the silent-no-op-edit and unattended-run laws (setsid, chain cells through a driver, ledger-before-night).

verification — Common Failures gains two rows: Edit applied (pattern asserted + line grep-verified, not "the command returned") and Options reached their consumer (probe with tiny max_iter, not "no error was raised") — an unverified edit is an unverified claim.

Why these, why now

Every failure class above was measured, named, and paid for during one overnight research run — the same classes will recur in any campaign that edits files programmatically, launches solvers, or banks measured numbers. The counter-doctrine (assert every transformation; machine-arbitrate every number; wire a positive gate to every measured axis; score the representation the hardware plays) is now pre-installed for future sessions instead of re-learned.

Notes

• Companion physics-card amendments (setup/solve/tool-selection/harmony) landed in the team vault root (armonissima skills/, commit 8150fd3, lint 0 errors) per the placement doctrine — amico-plugin was not used (corrected mid-session).
• Zero new lint errors: node packages/extension/scripts/lint-skills.mjs --dir packages/extension/skills — the only errors mentioning touched files are pre-existing companion-file errors.
• Vault knowledge cards for the physics findings landed in the personal vault (promote via the nightly dream pass).

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: 4f62ba42-42b2-478c-9aa7-4343fada03fd

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}