FutureFault rehearses every release before it ships. It proves what a change could break, fixes it, verifies the fix, and lets a deterministic policy decide.
Loading a recorded rehearsal…
Green CI is a claim. A rehearsal is evidence. Every conclusion links back to the record that proves it.
The bundled change passes every test it ships with. Here is what its own pipeline saw, next to what the rehearsal recorded.
no run recorded yet
An ordinary pipeline would merge this.
Eleven persisted steps, each resumable and idempotent. Scroll through the loop.
Agents propose falsifiable failures, each citing the exact lines that triggered it, with severity and likelihood rated on an ordinal scale, never a fake probability.
Only approved, registered experiments run, in child processes with a scrubbed environment and hard timeouts. Each records what it asserted and what it observed.
Confirmed defects get minimal patches plus a regression test that must fail before the fix and pass after it.
app/main.py · class TransferRequest - amount: float+ amount: float = Field(gt=0, le=1_000_000) + tests/test_ff_regression_validation.pyfails before the patch · passes after it
A fresh copy re-runs the original experiment, the full test suite and a security rescan. A fix that only looks right is rejected.
A digest-pinned policy reads only recorded evidence. Missing evidence counts as missing, never as a pass. No model decides.
The loop runs hands-off, but no model output ever becomes a command, a path or a verdict.
A five-agent GitLab Duo flow proposes hypotheses and fixes as validated data.
Allowlisted scenario IDs only, in child processes with a scrubbed environment and timeouts.
Twelve deterministic rules under a hashed profile. The exit code gates the pipeline.
Every tool call and transition is logged with a correlation ID and exported in the bundle.
Each sample is a real change to the same small service. Rehearse any of them in the live demo; the outcome is whatever the experiments record.
Auth bypass, unbounded amounts, blind readiness and credentialed CORS. All tests pass; the rehearsal reproduces all four.
A readiness refactor looks risky. A rehearsed database outage proves it still fails closed, so it ships with no fix.
Refunds drop the positive-amount bound. The fix is applied, then checked by an independent re-run before staging.
Cheaper readiness probes stop checking the database. Only an injected outage shows traffic still routed to a dead instance.
A destructive migration with no registered experiment. The policy refuses to guess and asks a human to approve.
A stage counts only when the run produced the evidence for it. The dashboard computes this live for every rehearsal.
Risk-ranked rehearsal plan from change impact
Targeted patches with regression tests
Experiments and independent re-verification
SAST, dynamic auth and CORS probes, audits
Reproducible artifact with SHA-256 manifest
Twelve-rule deterministic promotion gate
Health routes, non-root, bounded resources
Synthetic probes and fault-injected recovery
Audit trail, policy digest, evidence bundle
Watch a rehearsal reproduce, fix and verify real defects in about thirty seconds. No account needed.
Free plan includes 50 rehearsals a month · See pricing