Skip to main content

Ledger

Ledger is your workspace's decision memory. Every consequential change — a prompt rewrite, a model swap, a process decision made in a meeting room — gets recorded with a pre-registered expectation: what you believe will happen, written down before the evidence arrives. When the evidence lands, you settle the entry and distill what the workspace now knows. Writing the expectation first is what separates learning from narrative — without it, every outcome gets retold as intended.

The settled record is protected. An open entry is a draft you can still edit — but once settled, a claim is never rewritten; a correction is a new entry that supersedes the old one, with the history intact. That discipline is the point: a memory you can quietly rewrite is just a story.

The entry

Every entry has a kind:

Decisionentry kind

A change being made now, carrying an expectation of what happens next. Decisions open with the expectation and stay open until evidence settles them into a lesson.

Lessonentry kind

A distilled belief the team already holds — it arrives settled, no expectation needed. "Group benefits messaging lands best with the 51–100 employee segment."

Observationentry kind

A durable fact worth keeping — also arrives settled. "Five9 exports call dispositions nightly at 2am CT, as CSV, to the SFTP drop."

A decision carries six fields, and they fill in over its lifetime — not all at once:

FieldWhat it holdsWhen it's written
WhereThe Compass pages this touches — workflow, system, personAt creation
What we didThe change itselfAt creation
WhyRationale, one paragraph, at decision timeAt creation
What we expectThe pre-registered expectation — a number, comparator, and deadline when it can bind to data; plain words when it can'tAt creation, before any evidence
What happenedEvidence — Caliper eval runs or manual readingsAs evidence lands
What we now believeThe distilled lessonOnly at settlement

Ledger enforces the order. You can't write the lesson at creation, and you can't settle without one. Lessons and observations skip the open phase entirely — they're memory that arrives already settled.

The lifecycle

A decision is open while its expectation awaits evidence. Settling it — attaching what actually happened and writing what you now believe — happens exactly once, and locks the entry. An entry that later turns out wrong or reversed becomes superseded: a new entry takes its place and backlinks it, and the old one keeps its history. Nothing is ever overwritten.

The full mechanics — expectations that bind to Caliper evals, evidence, settlement, superseding — are on the entries page.

Two views

  • Entries — the stream. Every entry, reverse-chronological, filterable by lifecycle: open decisions (with a days-open counter), settled ones, memory, superseded. This is where you triage what still owes you an answer.
  • Beliefs — what the workspace knows. Every lesson, each citing the entry that earned it — like case law citing cases. Settled decisions and captured lessons sit side by side; superseded entries drop out. This is what a new teammate reads on day one.

Where it fits

The rest of the suite answers three questions: Compass knows where the work is, Workbench knows what's running, Caliper knows how well it scored. Ledger answers the fourth — why we did it, and whether we turned out to be right.

The connections are concrete. An entry's Where field links Compass pages, so you can pull every decision that ever touched a workflow before you touch it again. An expectation can name a Caliper eval with a target and deadline, so a measurement settles it. Run the loop enough times and the beliefs page becomes the workspace's institutional memory — earned, cited, and current.

Access

Ledger runs in the browser at ledger.zerowidth.ai — sign in and pick the Ledger tile from your workspace.

For programmatic access, Ledger is reachable through the MCP server — the same tools zv1 uses to consult the record before acting and to capture decisions from conversation.

Next

5 min read