Ledger quickstart

You'll record a real decision with a pre-registered expectation, attach the evidence when it arrives, and settle the entry into a lesson the whole workspace can cite. The example is a prompt change to a ticket-triage classifier — swap in whatever you're actually changing this week.

1. 1
   Open Ledger

   Go to ledger.zerowidth.ai, sign in, and pick your workspace. You land on Entries — the stream. Empty, for now.

2. 2
   Record a decision

   Click New entry. Ledger asks what kind of memory this is — pick Record a decision: you're changing something and predicting what happens.

   Fill the three fields that exist at decision time:

   • What we did — "Shortened the triage classifier prompt to the three load-bearing routing rules."
   • Why — "Long prompts were drowning the routing rules; accuracy plateaued two evals ago."
   • What we expect — this is the one that matters. On Measurable, pick Caliper eval as the source and write the expectation as a claim the data can check: >= 85 % accuracy, deadline two weeks out. If the outcome can't bind to a number — a room decision, a market call — switch to Freeform and write it in plain words instead.

   Optionally link the Compass pages this touches under Where, so anyone working on that workflow later finds this entry.

   Click Record entry. The decision is now open — an expectation on the record, waiting for reality.

3. 3
   Notice what you can't do

   Open the entry. What we now believe stays empty until you settle it. While the entry is open you can still edit it — but settlement locks the claim, so the expectation you wrote can't be quietly adjusted once the numbers come in.

4. 4
   Attach the evidence when it lands

   Two weeks later the Caliper eval runs. Open the entry and click Add evidence — reference the eval run and summarize the reading: "Run scored 88.2% routing accuracy on the v3 dataset." Evidence accumulates under What happened; an entry can collect several pieces before you're ready to call it.

5. 5
   Settle it

   Click Settle. One field: What we now believe. Don't restate the number — distill it. "Prompt length wasn't the bottleneck — routing examples were. Shorter prompts help only when the examples stay."

   Settlement happens exactly once. The entry locks, the expectation closes, and the lesson becomes part of the workspace's record — right or wrong, and knowing which is the value.

6. 6
   Capture something you already know

   Not every memory starts as an open question. Click New entry → Capture what we learned and record a lesson the team already holds, or a fact worth keeping — "the nightly export lands at 2am CT, so anything downstream is day-fresh at best." No expectation, thirty seconds, straight into the record.

7. 7
   Read the Beliefs page

   Switch to the Beliefs tab. Both lessons are there — the one you earned through settlement and the one you captured directly — each citing the entry that produced it. This page is the payoff: as decisions settle, it becomes the current, cited answer to "what do we actually know here?"

That's the loop: expect, observe, settle. Each pass costs a minute or two; the compounding happens on its own.

What's next