Failure-to-Gate preflight station logo

A stored failure that cannot refuse an action is only a note

A remembered failure only counts when a gate runs before the deploy.

This station runs the same canonical gate engine the CLI runs. It resolves recalled lifecycle events, re-runs the reviewed committed verifier, and shows the exact command that reproduces the result.

The Walrus mascot saluting beside the preflight gate station
  • 01 Read-only browser evaluation of committed fixtures
  • 02 No Mainnet write and no Mainnet proof happens in this page
  • 03 Persistence receipts stay in a separate committed evidence layer
Run the preflight gate

Gate station

01 · Select a committed scenario

Prompt revision (sha256)
reading…
Canonical source
demo/gate-engine.mjs

02 · Prompt context you supply

Your text is hashed and passed through the canonical untrusted-content scanner, then retained as data. It changes the intake row of the trace. It does not choose the gate outcome: that comes from the committed fixture events and the reviewed verifier run.

03 · Gate decision

Ready

GATE HELD

Waiting for the canonical evaluation

Scoped active
–
Held entities
–
Conflicts
–
Integrity notes
–

Canonical resolver reason: …

Without the evolved prompt

The agent recalls the past deploy failure as prose. A remembered “verified” note is treated as permission, so the same gas-budget break can ship again.

With the evolved prompt

The recalled failure becomes a named verifier that must be re-run on this commit. Recall is evidence; only a reviewed passing run opens the gate.

A held gate is the intended outcome here: the prompt turned a remembered failure into a check that must pass first.

Trace lane

Every row is produced by the canonical gate engine for the selected scenario.

    CLI reproduction

    GATE_CONFIG=demo/deploy.config.json node demo/gate-runner.mjs deploy 9f2a1c evidence/failures.sample.json

    The runner prints the same canonical gate counts and reason. The scenario’s committed context is loaded from evidence/failures.sample.json

    Separate evidence layer

    Gate decisions above are local, deterministic and credential-free. Walrus Mainnet persistence lives in a different, committed layer that is produced only by real MemWal calls that returned blob IDs.

    1. make gateThe fixture is underfunded, so the deploy is refused with exit code 1.
    2. make fix gateThe same gate clears once the corrected budget passes the reviewed verifier.
    3. evidence/mainnet-blobs.jsonCommitted terminal receipts, kept apart from this station.

    Built on Walrus Memory · a receipt is written only after terminal completion returns a blob ID

    ↑ Top