/pace — Parallel Agent Consensus Engine
You need a verified deliverable — not just an answer, but one that's been independently checked.
When you need this
- Writing a methods section you'll defend in front of a committee
- Verifying that a numerical claim in your paper is arithmetically correct
- Any task where an error would be hard to catch without a second opinion
What it does — and what it won't
/pace routes your task through two independent agents (Players A and B) who work without seeing each other's output. Each is reviewed by a Coach. A Cross-Reviewer then compares both teams. You get a consolidated result that explicitly surfaces where the two teams agreed (high confidence) and where they diverged (genuine uncertainty).
Note: /pace runs 5 agents and produces substantial output — best suited for tasks where correctness matters enough to justify the cost. For advisory questions with no single correct answer, /coa is a better fit.
The rule of thumb: use /pace when there is a correct answer, use /coa when there isn't one. If you're verifying a calculation or drafting a methods section, use /pace. If you're choosing between two valid research designs, use /coa.
/pace vs. /coa — which one do you need?
| /pace | /coa | |
|---|---|---|
| Question type | One correct answer exists | Multiple valid answers exist |
| Goal | Verified, cross-checked deliverable | Structured disagreement & synthesis |
| Output | Consolidated result ready to use | Advisory synthesis — you decide |
| Best for | Numerical verification, proof checking, document drafting | Research design, methodology choices, strategic decisions |
| Agents | 2 independent players + 2 coaches + cross-reviewer | 3–6 specialists with distinct expert lenses |
Worked example
Maya has written a methods section claiming her difference-in-differences analysis achieves a "12% reduction in demographic parity gap." She wants the arithmetic and experimental design verified before submitting to a workshop.
/pace Verify the DiD methods section in dissertation/methods.qmd — check every numerical claim against the evaluation logs in results/Players A and B independently verify each claim. Player B catches that the "12%" is computed on the test split but the paper's Figure 3 shows the validation split — an inconsistency. The Cross-Reviewer flags it as the highest-priority finding. Maya gets a consolidated report: 2 verified claims, 1 discrepancy requiring correction, 1 claim needing a confidence interval.
Try it
/pace Verify [claim] against [source]
/pace Write and cross-check [deliverable]Run /dailysummary after a PACE run to capture the findings — convergence results and any discrepancies resolved are decision context worth preserving.