skills.rest
krzemienski
krzemienskiCOMMUNITY
Explore all skills by this author →
Website𝕏@nkrzemienskiGitHub

consensus-synthesis

Synthesize per-validator PASS/FAIL reports into consensus verdicts with confidence scoring.

2|Updated Apr 8, 2026
Search Tags:#consensus#evidence#synthesis#confidence#validators#e2e-validation#disagreement-analysis
One-click install
npx skills add https://github.com/krzemienski/validationforge --skill consensus-synthesis
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: consensus-synthesis
Source: https://github.com/krzemienski/validationforge/tree/main/skills/consensus-synthesis
Command: npx skills add https://github.com/krzemienski/validationforge --skill consensus-synthesis

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This skill turns independent per-validator PASS/FAIL reports into a single, auditable consensus verdict per journey and an overall run verdict, preventing silent dismissal of dissent and enforcing strict read-only synthesis rules.

Core Features & Use Cases

  • Vote Tabulation: Collects per-journey and per-criterion votes from N validator reports and builds agreement matrices.
  • Synthesis & Confidence: Applies synthesis states and an agreement-ratio confidence formula to emit UNANIMOUS, MAJORITY, or SPLIT outcomes with HIGH, MEDIUM, or LOW confidence tiers.
  • Disagreement Handling: Triggers disagreement analysis for non-unanimous journeys, records dissenting evidence, and escalates unresolved splits to humans.
  • Use Case: Aggregate three independent end-to-end validators' reports for a payment flow, resolve disagreements, and produce a single evidence-backed run report.

Quick Start

Synthesize consensus from validator reports in e2e-evidence/consensus/ and write the consolidated report to e2e-evidence/consensus/report.md.

Frequently Asked Questions about consensus-synthesis

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I synthesize multiple end-to-end validator reports into a single consensus verdict?▼

You synthesize multiple validator reports by tabulating per-journey and per-criterion votes from all N reports, then applying an agreement-ratio confidence formula to emit UNANIMOUS, MAJORITY, or SPLIT outcomes.

What happens when end-to-end validators disagree on a test journey?▼

When validators disagree on a test journey, the system triggers a disagreement analysis for non-unanimous journeys, records dissenting evidence, and escalates unresolved splits to humans for review.

Do all validator reports need to be present to calculate consensus confidence?▼

Yes, all N non-empty validator reports must be present before synthesis begins, ensuring the agreement-ratio confidence formula has complete data to prevent silent dismissal of dissenting evidence.

How is confidence scoring calculated for synthesized e2e test verdicts?▼

Confidence scoring uses an agreement-ratio formula applied to per-criterion votes, assigning HIGH, MEDIUM, or LOW confidence tiers to UNANIMOUS, MAJORITY, or SPLIT journey-level and run-level synthesized reports.

What is the best way to aggregate multiple independent e2e validation reports into an auditable run report?▼

The best way is to use a read-only synthesis engine that builds agreement matrices, triggers disagreement analysis for non-unanimous journeys, and writes a single auditable consensus report file.

Can I write validator reports directly into the consensus directory during synthesis?▼

No, the synthesis process enforces strict read-only rules and avoids writing to validator-N directories, only emitting the final consolidated output to the designated consensus report file.

Related Skills

Explore More Software Engineering Skills→
krzemienskikrzemienski
2

consensus-disagreement-analysis

Compare validator evidence and emit per-criterion consensus resolution reports.

Community
Advanced
krzemienskikrzemienski
2

evidence-dashboard

Generate evidence dashboards with per-journey verdicts and historical trends.

Community
Advanced
baekenoughbaekenough
13

omcustom:agora

Orchestrate adversarial reviews among three or more LLMs to identify design flaws.

Community
Advanced
vTRKAvTRKA

dispatching-parallel-agents

Determine parallel dispatch safety for candidate tasks with overlapping scopes.

Community
Advanced