Recovery lab / verified local run

Fix one step. Keep the work.

An AI workflow has five stages. After the first run, the Score result becomes corrupted. See which stages Consistency keeps, repairs, and reruns.

Evidence boundary: This browser interaction replays the outputs of a real, deterministic Python run verified on October 28, 2026. The browser does not execute the private Consistency Python engine or an AI model.
01

Fetch

Ready
02

Score

Ready
03

Audit

Ready
04

Combine

Ready
05

Publish

Ready
Ready to demonstrate

Press Run to replay the recorded failure and recovery.

Recorded execution counts

StageExecutionsDecision
Run the walkthrough to see the verified counts.

Verified result

Awaiting replay

This scenario uses a supplied dependency graph and validation callback. It does not establish general reliability, production safety, or universal performance savings.

How to reproduce locally

With authorized access to the private repository and Consistency installed, run:

python examples/first_recovery.py

The example checks preserved stages, repaired stages, downstream reruns, execution counts, and final output. Source access is not required to view this walkthrough.