The decision register
Every decision, the evidence it stood on, and whether the model was right
A simulation nobody records is a conversation. The register is where a decision becomes a row: the question, the choice, the analysis or run it came from, the twin frozen as it was, the band the model gave for the figure it was judged on, who signed it, and, once the review date passes, what happened.
Build a twinHow an answer is made
The outcome closes the loop. Every closed decision says whether the actual landed inside the band, and the register keeps one calibration score across all of them.
A model that is never checked against what happened is a story. The register turns every prediction into a promise with a date on it.
Most tools that forecast never find out whether they were right, because nobody writes the forecast down where it can be compared later. Here it is written down by the software, at the moment of the decision, in the band it gave.
One row
What a decision carries
- Statuses
- 5proposed, approved, rejected, enacted, closed, in that order and no other
- Signatures
- 1 or moreeach with the name, the role and a note, kept for good
- The band
- 10 to 90lifted from the analysis or the run, never typed in later
- Snapshot
- 1taken the moment a decision is approved, so the evidence stays as it was
The life of a decision
From a question to a score
Record it from the evidence
Under every analysis and every run there is one button: record the decision this leads to. The question, the answer and the band for the lead figure are lifted from it, along with the eight numbers the answer rested on and where each one came from. Nobody retypes a forecast, so nobody can soften it.
- The predicted band for the figure the decision is judged on
- The assumptions it stood on, with their origins
- A review date, set from the horizon
Sign it
A proposal waits for a reviewer or the owner. Approving puts a name, a role and a note on the row and freezes the twin as a snapshot, so six months later the ledger the decision stood on can be opened exactly as it was. Rejecting keeps the row, with the reason.
Mark it enacted, then record the outcome
When the review date arrives the register says so. The actual figure goes in, in the same unit the band was given in, and the decision closes with a verdict: inside the band or outside it, and by how much.
Read the calibration
Across every closed decision: the share that landed inside the band, the average miss, and which way the model leans for this company. A ten to ninety band should hold eight times in ten. If it holds five, the bands are too narrow and the register says so, in one sentence, before anybody trusts the next answer.
The board pack
One document, assembled by the software, not by hand
Pick the decisions and the pack is built from the register: for each one, the question, the choice, the band, the outcome if there is one, the answer and its confidence, the reasons behind it, where it would be wrong, the files it drew on, the snapshot it was frozen with and every signature. Then the calibration paragraph, at the top, because it is the first thing a board should read.
It prints, or saves as a PDF, from the application. Every figure in it traces to a line in a file the company uploaded or is marked as an industry default, because it is the same data the screens show, laid out as a document.
- The calibration sentence at the top
- A section per decision, numbered
- Signatures and the frozen snapshot on every one
- Nothing typed in by hand
Honest about it
What the register does not do
Start the score
The first outcome you record starts the calibration. Everything after it is evidence.