Interactive synthetic proof
Build an evidence ledger without storing content
Add fictional evidence and inspect how the envelope states source, temporal certainty and privacy receipts. Secret-shaped input is refused rather than silently redacted.
Ready. Nothing leaves this browser.
No synthetic evidence recorded yet.
This demonstration models Worktrace semantics but does not use SQLite, install hooks, read shell history, access a filesystem or contact Personal RAG. All events disappear when the page is closed.
Implemented architecture
Canonical ledger, derived surfaces
explicit record or bounded collector -> input normalization and secret screening -> event envelope with source + temporal semantics -> local SQLite ledger (canonical) -> search / show / Markdown report / JSONL export -> explicit Personal RAG projection and admission boundary separate lane: declared project context -> attribution claim -> reconcile with observed evidence
Evidence and limits
What the source supports
| Claim | Evidence class | Boundary |
|---|---|---|
| Local-first Python CLI and SQLite ledger | Implemented in private source | No system-wide or cloud telemetry claim |
| Envelope version 2 with explicit temporal semantics | Normative event contract | Consumers must reject unknown or contradictory versions |
| Automatic Claude Code observation and RAG outbox recovery | Implemented and documented in v0.4.2 | No daemon, scheduler or browser-history collector |
| 962 tests, 0 skipped, CI on Python 3.10–3.13 | Source-reported at implementation receipt | Not re-run during this static publication change |
| Native Windows runtime | Declared dependency support | Source explicitly says execution was verified on WSL/Linux, not native Windows |