Four eyes

A second signature on every number, and the model marking its own homework

A ledger row can be changed by hand, and a changed row changes every answer after it. In a company with a board, an auditor or a lender, that is not a thing one analyst does alone at nine in the evening. A proposal is a change written down with the reason, what the model says it does to the next twelve months, and a place for a reviewer to sign or refuse.

Build a twinRoles and snapshots

The proposer cannot sign the second line, whatever their seat. That is the whole point of four eyes.

proposalCost to win a customer$8.2k$11kPaid channels cost more since the platform change.twelve month ebitda-8% with the changeproposed byTom Reilly, analystapproved bywaiting for a reviewersnapshot firstthe proposer cannot sign the second line, whatever their seat

The most expensive number in a model is the one somebody changed quietly. A proposal makes the change loud, dated, reasoned and reversible.

Every approval takes a snapshot of the ledger first, so the moment before the change is always one click away, and the row that changed carries the proposer, the reviewer and both their notes for good.

The life of a proposal

From an edit to a signed change

01

Propose it, with the reason

Pick the number, say what it should be and why. A proposal with no reason is refused, because a reviewer with nothing to read cannot review. The old value, its band and its origin are kept on the proposal, so the reviewer sees exactly what would change.

  • Any number in the ledger
  • A band, carried over from the current row or typed
  • The reason, kept for good
02

The model runs before anybody signs

The twin runs twice at low replication with the same seed: once as the ledger stands and once with the proposed number in it. The difference over twelve months is the proposal's impact, kept with the proposal, so the reviewer sees the same figure the proposer saw whatever changes in between. When a number does not move the simulation, the proposal says so rather than showing a row of zeros.

proposalInterest rate on that debt9%11%The facility reprices in October at base plus four.twelve month ebitda-1% with the changeproposed byDana Okafor, analystapproved bywaiting for a reviewersnapshot firstthe proposer cannot sign the second line, whatever their seat
03

A reviewer approves or rejects

Approving takes a snapshot, moves the ledger, locks the row, and writes the proposer, the reviewer and both notes on it. Rejecting leaves the ledger alone and keeps the proposal with the reason. The proposer can withdraw. Nobody can approve their own proposal.

  • A snapshot before every approval
  • The proposal.decided event on hooks
  • An alert if a proposal waits longer than the days you set

The rule

Switch it on and an analyst's edit becomes a proposal

The owner of a twin can require a reviewer for changes to the ledger. With the rule on, an analyst who edits a number in the ledger is asked for the reason, and the edit lands in the queue instead of the ledger.

Owners and reviewers still edit directly, because they are the people the rule exists to bring in.

  • Off by default; on with one switch
  • Analysts propose, reviewers and owners decide
  • Viewers read the queue and nothing else
A spreadsheeta single line, and every number in it is yoursAsking a modelfluent, and the same question twice gives two answersA consultantgood work, and you cannot ask the next questionA twina band, the actor that caused it, and a source per figure

The track record

Every new answer carries how the old ones did.

The decision register knows how often a closed decision landed inside the band the model gave, and which way outcomes lean. The monitor knows how far the actuals have run from the pinned forecast, metric by metric. Every new analysis reads both. Three or more closed decisions move the confidence, by up to fifteen points. A metric the monitor has seen drifting gets its band widened by the drift seen, and the answer says so, because a band the actuals keep escaping is not a band.

±15
points of confidence, at most, from the track record

Two signatures. One snapshot. No quiet changes.

Proposals come with the governance layer, on the Operator plan and above.

Build a twinEnterprise