CIVITAS

Zeta

A research program for claim-to-evidence discipline.

Reproducible runs. Clear scope. Audit-ready artifacts.

What Zeta is

Zeta is a methodology and reporting format for evaluating information, model outputs, or system claims against verifiable artifacts. It supports review by making evidence paths explicit.

Zeta does not declare truth globally. It produces a bounded assessment inside a defined scope and dataset.

Outputs are designed to be human-reviewable and procurement-friendly, with explicit limits and uncertainty disclosure.

What Zeta measures

Claim-Artifact Mapping

What it asks: For every claim, what artifact supports it?

What evidence looks like: Links to files, logs, datasets, screenshots, hashes, or deterministic command outputs.

Reproducibility Path

What it asks: Can a reviewer re-run and get the same outputs?

What evidence looks like: Versioned commands, pinned inputs, fixed seeds, and environment notes.

Scope Boundaries

What it asks: What is explicitly in and out of scope?

What evidence looks like: Written assumptions, exclusions, and known limitations stated upfront.

Uncertainty Disclosure

What it asks: What is uncertain, and why?

What evidence looks like: Confidence notes, missing-data notes, and sensitivity to assumptions.

Provenance and Change Control

What it asks: Where did the data come from and how did it change?

What evidence looks like: DOI, SHA, and commit references with dated change logs.

Reviewer Ergonomics

What it asks: Can a human reviewer audit without heroics?

What evidence looks like: Short executive summary plus traceable appendix with minimal jargon.

Robustness Checks

What it asks: Do conclusions hold across plausible variants?

What evidence looks like: Alternative runs, ablations, and counterexamples logged when relevant.

Integrity Posture

What it asks: Are there implied endorsements or unverifiable statements?

What evidence looks like: Explicit no-implied-partnership language and clean citations.

Deliverables

Evidence Map

Claims mapped to artifacts with clear traceability and scope references.

Reviewer Pack

Executive summary and appendix designed for procurement and audit review.

Repro Bundle

Commands, pinned inputs, and hashes to reduce verification debt.

Minimal example

Claim: Feature X reduces failure rate under condition Y.

Artifacts: run log, dataset hash, command line, chart export, commit id.

Notes: scope boundaries and uncertainty statement attached to the claim.

Research posture

MethodsBenchmarksRequest intake