Evidence Compiler

Vericite

Vericite turns scattered records into source-traced timelines: claims become evidence cards, every card carries a trust score, and private material stays outside the public product repo.

vericite.cristiandominguez.com
Vericite evidence timeline interface
Status
Live site / protected proof of concept
Role
Product design, frontend, worker auth, deployment model
Core idea
Claims checked against source records, not memory alone
Boundary
Product code is public; evidence data is separate and private

Why it exists

Vericite started from a hard product problem: people often have evidence scattered across messages, emails, transcripts, screenshots, documents, and notes, but the useful artifact is a clean timeline that shows what can actually be proven.

The product does not ask the user to trust a summary. It compiles a record where each event traces back to sources, carries a confidence level, and makes gaps visible instead of hiding them.

That framing matters. Vericite is a documentation and evidence-organization tool, not a legal service and not a system for fabricating certainty.

What I built

The shipped surface combines a public explanation page, early-access capture, password-protected hosted timelines, a synthetic public demo, source cards, trust scoring, PDF-oriented output, and a dedicated Cloudflare Pages/Worker deployment.

Evidence timeline viewer

Events are grouped chronologically with expandable context, category filters, search, keyboard navigation, source cards, and print/PDF behavior.

Trust scoring

The interface distinguishes reported, corroborated, documented, and multi-source facts so strong evidence is not diluted by weaker claims.

Provenance model

Sources are treated as chains from original record to compiled card, including platform, timestamp, file reference, and verification context.

Separated deployment

The product runs on its own Cloudflare Pages project with Worker-based auth/data routes, separate from the personal site and separate from private evidence.

How the compiler thinks

The product workflow is intentionally conservative: a claim is only useful if it can be matched against source material, contradicted by source material, or labeled honestly as thin.

ClaimA plain-language memory, event, quote, or pattern the user wants to check.
RecordsMessages, emails, transcripts, documents, screenshots, and metadata are searched for support or contradiction.
TimelineThe output is a sourced event card with date, context, trust score, and visible limits.
1Thin
2Reported
3Single source
4Corroborated
5Documented

Decisions that mattered

Separate product from evidence. The GitHub repo contains platform code and public assets. User evidence lives outside the repo and is staged only during protected deployment flows.

Make uncertainty visible. The timeline is stronger because low-confidence cards are labeled. The product does not pretend that every recollection is equally documented.

Own the deployment boundary. Vericite was moved onto a dedicated Cloudflare Pages project so personal-site routing changes cannot accidentally serve or break the product.

Use synthetic demos publicly. The public demo can show the interface and workflow without exposing private source material or real case details.

The important product move is not "AI summarizes your life." It is "AI helps compile a record you can audit."

Vericite is an evidence product built around source discipline, access control, privacy boundaries, and UX for material where certainty has consequences.

Why it holds up

Vericite is strongest when it refuses to turn uncertain evidence into confident prose. The timeline has to expose sources, score confidence, make gaps legible, and preserve privacy boundaries.

The operational side matters just as much: dedicated Cloudflare ownership, worker-level access control, explicit smoke tests, and product architecture that keeps sensitive data out of public repos.