From evidence to a decision.
Eight steps. Three phases. A conclusion you can trace, challenge and revisit.
NCB-004 · v0.3 ↗
Establish the evidence
- 01
Find the evidence
Define the question and choose the sources.
Software-assisted - 02
Extract the claims
Separate dated observations from commentary.
Software-assisted - 03
Grade the sources
Assess quality and independence separately.
Rule-guided
Work out the implications
- 04
Connect the signals
Map relationships and grade their confidence.
Rule-guided - 05
Name the uncertainty
Find the unknown that could change the conclusion.
Accountable judgment - 06
Weigh the choices
Show which options the evidence opens or closes.
Accountable judgment
Publish and review
- 07
Write the brief
Use the same seven-part format each time.
Rule-guided - 08
Review and correct
Sign off and record any later changes to the assessment.
Accountable judgment
People run the published workflow. Software supports retrieval and formatting; a named reviewer owns the conclusion.
A brief with its reasoning attached.
A seven-part brief with sources, unanswered questions and a dated record of review.
Read a published example- 01
The conclusion
What the evidence means for a specific decision.
- 02
The evidence behind it
Sources, relationships and documented disagreement.
- 03
The next test
What is uncertain and what would change the conclusion.
- 04
The dated record
Who reviewed it, and what changed after publication.
Inspect the reasoning.
Follow a conclusion back to its sources.
Challenge the conclusion.
See the uncertainty and the evidence that would overturn it.
Keep the history.
Compare later outcomes with the original rule.
The rules and their limits.
Expand a section to check the rules or their limits.
Six audit rulesHow the record stays checkable
Pre-registration
Before tracking begins, FP1 publishes each assessment, its confidence level, threshold and the evidence that would prove it wrong. If a threshold changes during that period, the signal loses its registered status. It is publicly retired and the change is logged.
Source and independence grading
Every claim has separate grades for source quality and independence. Five outlets repeating one wire story count as one source, not five.
Relationship mapping
Connections between signals get their own confidence grades. Strong evidence for two facts does not prove a relationship between them.
Disagreement preserved
Each analytical method keeps its own reasoning. The brief records where they agree, where they disagree, or where evidence is insufficient. Editing must not erase a disagreement.
Grading discipline
Review dates are fixed in advance. The review date and publication date are shown separately, so delays stay visible. Reviewers must choose from five published verdicts: held, held and strengthened, held and watched, tripped, retired. They cannot invent a softer label for a failed assessment.
Self-falsification
An assessment fails when its stated test is met. The method itself must be revised publicly if thresholds change during tracking, a test is met without earlier warning on the chart, the chosen historical comparison cannot describe the change, or reviewers cannot resolve conflicting grades.
Four stated limitationsWhat the method cannot guarantee
Choosing historical comparisons takes judgment
Analysts can choose different historical cases for comparison. FP1 names the cases it uses so readers can challenge the choice.
Some histories are thin
Where a threshold is anchored to a benchmark rather than to a base rate, it is labelled as such and carries at most medium conviction.
Chart positions are not probabilities
A chart shows ranking and direction, not probability. A marker twice as far from a line is not twice as safe, and nothing here claims otherwise.
FP1 reviews its own work
FP1 sets the thresholds and reviews the results. Separate roles, fixed verdicts and public records make that work easier to check. They do not replace an external audit.
Software and human reviewCurrent workflow versus development
Current published material is produced by people working through all eight stages. Software supports retrieval and formatting. Rule-guided steps apply published standards; they do not delegate judgment to a program.
A computational engine intended to assist stages 1–4 is in development. It does not produce current published material. Consequential judgments and corrections remain attributable to a named reviewer.