ClisTa ClisTa Decision provenance
Middle rungs

Start small.
Climb to a real run.

A full audited run on a real, hard-to-reverse decision is a cliff for a cold start. These rungs are the staircase: low personal risk, quick to complete, each producing something that demonstrates the format.

0 30 sec
shipped
30-Second Replay Rung 0

Clean-room reproduction of the bundled example. Verifies determinism and ends with Clean-room replay PASSED. The universal first action.

git clone …/clista.git && cd clista/packages/protocol && npm run replay

1 ~20 min
One-Paste Sealed Run Recommended first rung

One prompt, pasted into your own agent, with any decision swapped into the scenario block. It spawns a maker and an adversarial checker, forces every step through an append-only hash-chained gate, seals the thread, and emits a claim-cited report — then the verifier checks it mechanically. No install, no workflow to learn.

  • Copy the prompt from the reference run (genesis_prompt.md) and swap in your decision.
  • Paste it into any capable agent; let the maker and checker fight through the gate.
  • Run python3 gate.py verify on the sealed thread — pass criteria are mechanical, no judgment.

Graduates to Rung 4: the same prompt on a real, hard-to-reverse decision you own is a submittable external run. If your checker changes nothing, or the run falls apart — that is wanted evidence too.

2 15–25 min
Scenario React The critic's seat

Engage with the format as a critic on a real — but not your own — decision. Zero personal stakes; forces engagement with the objection-handling discipline.

  • Clone the repo and read packages/protocol/examples/scenario-demo/expected-state.json.
  • Pick one Critic lane (e.g. Privacy/Compliance or Operations).
  • Write 3–5 numbered objections against the claims, using the state vocabulary (CLAIM, OBJECTION, SEVERITY).
  • Optionally fill the relevant rows of LEDGER_TEMPLATE.md.

Report: post as a comment on the scenario, or open a lightweight "Scenario critique" issue.

3 30–40 min
Toy Decision The "is the overhead worth it?" test

Run the full pack format on a low-stakes but real decision you actually face. Full ceremony, zero real consequence — this is where people feel the overhead honestly.

  • "Should we add dark mode to the internal tool this quarter?"
  • "Approve or reject the new contractor for the Q3 design-system audit."
  • "Upgrade the team's LLM provider before the next kickoff?"

Use the full roles and close protocol. Produce at least the ledger table, a Transfer State, and a short failures.md. No real execution required.

Report: a "Toy run" issue with a separate label, so it does not pollute the run ledger.

4 the summit
Full External Run The real thing

A real, hard-to-reverse decision with full artifacts — LEDGER.md, failures.md, cost.md, outcome.md — verified mechanically on receipt. This is what the run ledger records.

File a run report

The ladder turns an all-or-nothing ask into a funnel without diluting the target: only Rung 4 runs on real decisions enter the run ledger. Everything below builds the judgment to get there.