part nine · prediction writing
Prediction Writing
Baseline, target, check date, method, owner.
A prediction is a falsifiable claim about a measurable change, written before the cycle runs. It is the chain's smallest unit of honest commitment. A brief that does not contain one is not a brief — it is a description.
This part of the volume is short, dry, and exact. The discipline does not vary.
The five fields
Every prediction has the same five fields. They are not negotiable.
| Field | What it is | What goes wrong without it |
|---|---|---|
| Baseline | The current measured state. Number, with sample size and date. | The check has nothing to compare to. |
| Target | The state we expect after the change. Number, range, or threshold. | The check cannot say yes or no. |
| Check date | The day someone will run the check. | The cycle ends and no one runs the check. |
| Check method | The specific way we will measure. | The cycle ends and people argue about what counted. |
| Owner | The named person responsible for running the check. | The check date arrives and no one shows up. |
The fields are anchored to measurable things. We expect users to be happier fails three out of five — no baseline, no target, no method.
A prediction worth writing
Prediction: Gal completes a grading cycle in under 15 minutes.
Baseline: 47 minutes (mean), 38 minutes (median), n=12,
captured 2026-04-22 by direct observation.
Target: <15 minutes (mean) across n>=8 observed cycles
OR <12 minutes (median) across n>=8.
Check date: 2026-06-15.
Check method: Three observation sessions across three named
graders, in the field, with a stopwatch and the
existing time-on-task instrumentation as cross-check.
Minimum 8 cycles total. Same observers as Discovery.
Owner: Alex (PO).A prediction that fails the five fields
Prediction: Make grading easier and faster for users.No baseline. No target. No date. No method. No owner. This is not a prediction. It is a marketing claim.
Range, threshold, ratio
Predictions can be expressed three ways. Pick the one the change supports.
| Form | When | Example |
|---|---|---|
| Specific number | The change has a single dominant effect | Cycle time falls to <15 minutes mean |
| Range | The change has high variance across the population | Time saved is in the range 25–40 minutes per cycle for graders with >50 cycles/week |
| Threshold | The change must clear a binary criterion | Hebrew names render correctly across 100% of test corpus |
A prediction that hides behind we'll see what happens is not predicting. It is observing.
Honesty in baseline
The baseline is the most-skipped field. Teams write we will improve X by 30% without saying what X currently is. The check then becomes we improved by 30% relative to nothing, which is the chain's most common form of self-deception.
The corpus pattern: no baseline, no execution. If the team does not know the current state with sample-size precision, the team has not finished Discovery. The first execution week of any cycle is sometimes spent capturing the baseline that should have been captured during Discovery.
Honesty in check method
The check method is the most-skipped field after baseline. Teams write predictions and execute, then arrive at the check date and discover the metric they need was not instrumented. The check becomes a survey, which is the chain's second most common form of self-deception.
The corpus pattern: the check method is named in Discovery. If the method requires instrumentation that doesn't exist, that instrumentation is one of the cycle's stories — as Alex (PO), I want to know how long Gal spent grading, so that I can run the check on 2026-06-15.
Predictions feed the check, the check feeds the model
This is the chain. The prediction is written in Volume II. It is kept alive through Volume III's stories and Volume IV's instrumentation. It is run in Volume V Part 2 — Signal and the Prediction. The result feeds the model update — Volume V Part 6.
A team that does this for three cycles in a row is already most of the way to a chain. A team that skips even one of the five fields will, within those three cycles, hit the kind of not checked outcome that the corpus identifies as the only result with no value.