START HERE Agent Forensics Suite
Start here: verify agent runs with receipts, replays, diffs.
Audit, prove, replay, and diff agent runs—turning “trust me” into verification.
Start here: verify agent runs with receipts, replays, diffs.
Note START HERE: This space is the suite’s entrypoint with direct links to the full verification workflow—learn → generate proof → record reality → seal it → diff it → audit it → benchmark it. Start with ReplayProof, export a bundle + receipt, verify integrity, then use TimelineDiff and TrustStack to pinpoint and inspect divergence.
Play it. Prove it. Replay it.
Note ReplayProof is a lightweight, game-like Agent POV sandbox that turns “trust me” demos into verifiable runs. Play a deterministic gridworld, export a run as a ZIP bundle containing signed, hash-chained receipts (plus an optional GIF), then upload the bundle to instantly verify integrity and replay the exact run anywhere. Built for reproducibility, provenance, and shareable challenges — no GPU or model downloads required.
Tamper-evident, hash-chained event logging for AI/agent runs
Note Log what an agent actually did—prompt, tools, outputs, memory reads/writes—in a hash-chained timeline that breaks on tampering. Verify sessions end-to-end and export bundles anyone can validate.
Note 📦 Part of the RFTSystems. Memory Receipt Engine is the “proof layer” — it generates a downloadable, tamper-evident receipt for an agent run, and supports receipt upload to independently verify that memory/state wasn’t rewritten.
Diff agent timelines. Find first divergence. Export forensic
Note TimelineDiff is a Differential Reproducibility Protocol (DRP) lab for agent forensics: upload two hash-chained run bundles to align timelines, pinpoint the first divergence, and explain what changed (LLM sampling, tool I/O, memory writes, control-flow, guardrails, state). It outputs a scan-friendly alignment view plus machine diffs, then exports shareable reports (JSON + Markdown + PDF) so stochastic agent failures stop being ghosts in logs.
infrastructure, not a toy.
Note 📦 Part of the RFTSystems. TrustStack Console is the “audit cockpit” — it lets you inspect runs, compare states, and see exactly what changed and why, so debugging and governance becomes verification instead of having to blind trust.
Live coherence + throughput benchmark (no precomputed result
Note Coherent_Compute_Engine is a live, verification-first benchmark that runs a coherence-preserving [Ψ, E, L] update engine in real time, reports true items/sec + stability/energy behaviour, and outputs a SHA-256 receipt you can download and verify.