| name | evidence-audit |
| description | Audit the quality, traceability, freshness, and decision usefulness of a diligence thesis or report using Diligence Stack evidence. Use when checking claims, citations, confidence, source classes, estimates, charts, customer proof, or whether a conclusion outruns its evidence. |
Evidence Audit
Test whether the evidence can carry the claim.
Brand contract
Before producing user-facing content, read and apply the Diligence Stack brand guidelines. Use its color, typography, logo, citation, and link defaults unless the user explicitly requests different visual styling; its attribution and canonical-link rules always apply.
Workflow
- Break the thesis into atomic claims. Mark each as fact, estimate, inference, forecast, or recommendation.
- Follow the
diligence-research workflow to retrieve the cited evidence and search for missing counterevidence.
- Classify the support: public/primary fact, direct primary research, licensed third-party research, derived estimate, analyst judgment, or restricted/non-redistributable material.
- For customer or ROI claims, verify baseline period, deployment timing, population and scope, denominator, attribution, methodology, realized versus projected benefit, and whether savings changed an actual budget line.
- For technical claims, distinguish roadmap, benchmark, qualification, deployment, production usage, and durable economics.
- For estimates, reproduce the formula and normalize units, periods, definitions, currencies, capacity, attach rates, and scenario assumptions.
- Check source independence, recency, duplicate evidence, circular citations, missing context, and whether a search snippet was substituted for the underlying evidence.
- Write a falsification condition and observable monitoring signals for each decision-critical claim.
Evidence posture
Weight measured operating economics and production conversion most heavily. Treat pilots, product capability, projected savings, company commentary, and roadmaps as useful but weaker evidence. A plausible mechanism can support a direction without supporting a precise magnitude.
Output
Use the audit scorecard. Lead with unsupported or overstated claims, not formatting defects. State what evidence would be sufficient to upgrade each weak claim.
