Determinism verification agent
Replay it 10×.
Bit-for-bit.
rig-replay records a seeded run once, then replays it under a virtual clock and demands bit-for-bit identical output hashes across every replay. Feed it the same recording under an injected model-change and it must catch the regression — not silently swallow it.
rig-replay · virtual-clock replay engine · UML-RT state machine[higgsfield: gold-lit reel-to-reel tape loop frozen mid-spin, ten stacked frame ghosts perfectly aligned, noir studio backdrop]
Why teams install
What rig-replay earns back
Bit-for-bit across 10 replays, every time
Same recorded seed re-executed 10× under a virtual clock — all 10 hashes and the recorded hash match exactly.
‹31e05b4›Regressions can’t hide in noise
An injected model-change (dropped backlog clamp) replayed against the same events — the hash mismatch is DETECTED, not silenced.
‹c516323›The clock is the only thing allowed to lie
Force the live wall clock instead of virtual time and 10 replays produce 10 distinct hashes — the gate goes RED immediately.
‹35009d4›How it works
Three steps. No faith required.
01
Record
One seeded run captures the full event trace plus a canonical SHA-256 hash.
02
Replay
Re-derive the same events from the seed, re-run n=10 times, compare hashes.
03
Mutate
Replay the same recording against an injected model-change; a hash mismatch must fire.
Proof band
One packet. Every claim accounted for.
packet rig/rig-replay · determinism-gates v1 run run-42-baseline · seed 42 · 40 events · n=10 result determinism PASS — 10/10 hashes match 31e05b4a… result regression_detection PASS — mutant hash mismatch DETECTED anchor PASS — virtual-clock replay verified bit-for-bit ‹31e05b4› themis SEALED — merge gate armed
Every rig-replay run emits a packet like this — hash-chained, verifier-signed, and queryable from your own dashboard. How ProofPackets work →
Buyers
Who runs rig-replay
Staff SRE
distributed systems team
QA Automation Lead
fintech backend
Principal Engineer
concurrent systems shop
Objections
Answered before you ask
Isn’t this just recording and diffing logs?
Logs can be edited after the fact. rig-replay re-derives the event stream from the seed and re-executes the system 10 times live — the hash has to match a fresh run, not a stored file.
How do you know it’s not deterministic by accident?
Force the live wall clock instead of virtual time and 10 replays produce 10 distinct hashes immediately — determinism is engineered, and the test proves it’s load-bearing.
What if a real model change slips through undetected?
The injected model-change (a dropped clamp in the ack handler) is a genuine, plausible regression — and it’s caught on every seed tested, not a synthetic no-op.
Pricing
Start free. Pay when it pays.
Generalist ships today. Domain Expert adds vertical memory + priors. Custom is built to your stack.
Free
Priced after a short conversation
25 replay runs a month.
Team
Priced after a short conversation
Unlimited replays, mutant injection suite, CI gate.
Pro
Priced after a short conversation
Dedicated replay fleet, custom SUT adapters, SLA.
SB 26-189 readiness — deterministic regression evidence is what state auditors ask for first.
