Evidence + legal operations · Independent
Source Integrity Workbench
Live claim-to-source review workbench · v2 August 2026
Turn claims and citations into a structured verification ledger, review queue, and portable release packet before publication, filing, or decision.
Controlled visual demonstration
Current live interface
This sandboxed preview shows the current public interface. Open the full site to use the build in its intended window.
What I noticed
AI-assisted and human-authored documents can contain authorities that do not exist, sources that do not support the attributed proposition, outdated law or science, vague attribution, and claims whose confidence exceeds the underlying evidence. A spot check cannot show which claims were actually reviewed.
What I researched
The method separates claim type, materiality, retrievability, source type, support, contradiction, currency, attribution, conflicts, reviewer responsibility, and disposition instead of inferring source validity from citation markers.
What exists
A five-stage local-first workbench covering the review brief, claim ledger, source register, claim-level findings, blocker queue, release/qualify/replace/remove/hold/escalate decisions, documentation completeness, browser persistence, and Markdown, CSV, JSON, and print/PDF exports.
Who it serves
Publishers, law firms, researchers, founders, creators, portfolio professionals, and trust-sensitive organizations.
What it demonstrates
Evidence operations, claim-source mapping, review workflow design, deterministic decision logic, local-first privacy, and portable recordkeeping.
Current evidence
The canonical v2 production release was browser-verified end to end with a brief, material claim, retrievable source record, qualified support finding, disposition, explanatory note, and generated packet. Its score measures documentation completeness—not truth or source authentication—and legal, clinical, scientific, regulatory, and other high-stakes conclusions still require the appropriate qualified reviewer.
Verified build record
What is actually running, and what it is not.
LiveLive · v2, August 2026
The problem. Documents now arrive containing authorities that do not exist, sources that do not support the proposition attributed to them, and confidence that outruns the evidence. A spot check cannot show which claims were actually reviewed, which is the thing a reader needs to know.
What it does. Turns claims and citations into a review ledger across five stages: review brief, claim ledger, source register, claim-level findings, and disposition. It tracks claim type, materiality, retrievability, source type, support, contradiction, currency, attribution, conflicts and reviewer responsibility separately, rather than inferring validity from the presence of a citation marker. Outputs a release packet as Markdown, CSV, JSON or print.
Built on. A static browser application with local persistence, served from Vercel.
What it does not establish. The score measures documentation completeness, not truth and not source authentication. Legal, clinical, scientific and regulatory conclusions still need the appropriate qualified reviewer.
This record is generated from the same inventory as the complete build index: the production URL was checked at the last regeneration, and the status line above repeats what the build reports about itself rather than restating an intention.