Adversarial Quarto vs Beamer QA Workflow

Compare Quarto HTML slides against their Beamer PDF benchmark using an iterative critic/fixer loop.

Philosophy: The Beamer PDF is the gold standard. The Quarto translation must be at least as good in every dimension.

---

Workflow

Phase 0: Pre-flight → Phase 1: Critic audit → Phase 2: Fixer → Phase 3: Re-audit → Loop until APPROVED (max 5 rounds)

Hard Gates (Non-Negotiable)

Gate	Condition
Overflow	NO content cut off
Plot Quality	Interactive charts >= static plots
Content Parity	No missing slides/equations/text
Visual Regression	Quarto >= Beamer in all dimensions
Slide Centering	Content centered, no jumping
Notation Fidelity	All math verbatim from Beamer

Phase 0: Pre-flight

1. Locate Beamer (.tex/.pdf) and Quarto (.qmd/.html) files
2. Check freshness (re-render if QMD newer than HTML)
3. Verify TikZ SVGs if applicable

Phase 1: Initial Audit

Launch the quarto-critic agent to compare Beamer vs Quarto comprehensively. Report saved to quality_reports/[Lecture]_qa_critic_round1.md.

Phase 2: Fix Cycle

If not APPROVED, launch quarto-fixer agent to apply fixes (Critical → Major → Minor), re-render, and verify.

Phase 3: Re-Audit

Re-launch critic to verify fixes. Loop back to Phase 2 if needed.

Iteration Limits — loop-until-dry

This is the loop-until-dry primitive from orchestrator-protocol.md: the critic returns FINDINGs (the hard-gate table is the CRITICAL roll-up, per orchestration-schemas.md); the loop converges when a round adds 0 new CRITICAL/MAJOR findings (deduped on id = sha1(file:line:locus)), not at a fixed round count.

• Fallback cap: 5 rounds bounds a non-converging loop, then escalate to the user with remaining issues.
• Two-strikes: the same gate failing in rounds N and N+2 is flagged for the user, not patched again (summary-parity.md).
• APPROVED iff every hard gate passes (zero CRITICAL).

Final Report

Save to quality_reports/[Lecture]_qa_final.md with hard gate status, iteration summary, and remaining issues.

Findings are validated, not just written (v2.5)

This skill's reviewers emit findings under the machine-checked contract in finding-schema.json. Reports are JSON arrays.

Smoke-test the harness before spending review effort — a run that fans out reviewers and then cannot write a valid report has wasted the whole pass:

echo '[]' | python3 scripts/validate-findings.py

Then, before presenting any summary:

python3 scripts/validate-findings.py <report>.json   # exit 0 required

What the contract forces, and why:

• rule — the documented rule or standard violated. A finding citing no rule is an opinion, and opinions do not gate a commit.
• failing_case — a concrete configuration under which the claim breaks, or the exact missing hypothesis. "This could be clearer" does not validate.
• id = sha1("<file>:<line>:<locus>") — deterministic, so dedup across rounds is exact and the two-strikes rule is checkable rather than eyeballed.
• mechanical — true only for fixes that cannot change a result (typo, cross-reference, formatting, label). Never for an estimand, assumption, specification, inference procedure, sample definition, or reporting language: those return to the researcher.

Apply the per-lens evidence burdens and the "does NOT count" filters in orchestration-schemas.md §7 before verification, so known false alarms never reach the judge. The verifier pass is refute-biased: only verdict: "confirmed" findings ship; anything it cannot ground is dropped, not downgraded to a warning.