method / working protocol

Make the claim smaller than the evidence, not larger.

A cross-domain research process built around preregistration, provenance, falsification, negative results, and reproducibility.

This page is the presentation layer. The canonical protocol lives in the research-method repository, with templates, schemas, verification tooling, and worked examples.

01

Question

Turn the motivating idea into an observable distinction. State what would count as an answer.

02

Prior art

Look for existing methods, confounds, equivalent formulations, and reasons the question may already be answered.

03

Claim boundary

Write down the strongest claim the proposed evidence could support — and the broader claims it could not.

04

Hypotheses

Define predictions and failure conditions before the result is known.

05

Preregistration / freeze

Freeze sample selection, statistics, nulls, thresholds, exclusions, and stopping conditions when post-hoc freedom matters.

06

Acquisition

Record where the bytes came from, when they were retrieved, and what transformations occur before analysis.

07

Analysis

Run the declared procedure. Separate measured quantities from derived bounds and projections.

08

Adversarial checks

Try to destroy the result with nulls, alternate assumptions, source-blind checks, differential tests, or explicit attack models.

09

Negative results

Failed gates remain part of the record. A conjunction where two of three criteria pass is still a failed conjunction.

10

Provenance

Hash artifacts, preserve lineage, and distinguish what was independently checked from what an artifact merely records.

11

Publication / reproduction

Make the path from question to source bytes to result inspectable by someone who did not participate in the work.

Status vocabulary

epistemic state is part of the interface
VERIFIED

Independently checked against the underlying artifact or computation.

RECORDED

Asserted by a source or artifact but not independently rerun or recomputed here.

DERIVED

Follows from declared premises; the derivation must be inspectable.

FAILED

A declared criterion was not met. It remains visible.

PROJECTED

Extrapolated under an explicit, untested assumption.

NARROWED

A stronger interpretation was withdrawn and replaced by what the evidence actually supports.

Method applied

real examples
ObjectClaim stateWhy it matters
Time Integrity P1VERIFIEDFrozen statistic; source lineage and outputs preserved.
Time Integrity D2bFAILEDTwo sub-criteria hold; scatter misses the frozen threshold by about 6.2×.
FinCEN ownership edgesSOURCE-DEPENDENTEdges remain traceable to separately public evidence rather than being promoted to confidential BOI facts.
811Dog classificationsEXPLAINABLEEvery score records the rules that fired and can be rebuilt under changed rules.