01 / Unit of analysis

Make the statement small enough to fail.

Too broad

Finance controls governments.

Researchable

A named agreement made a defined disbursement conditional on specified domestic policy changes during a stated period.

The unit of analysis is a specific claim about a named mechanism, actor, place, and period. Broad political arguments are decomposed until evidence could plausibly reject them.

02 / Four layers

Evidence at one layer does not establish the next.

  1. 01

    Policy text

    What did an agreement, law, order, programme, or contract formally require?

  2. 02

    Implementation

    What did institutions, governments, firms, and individuals actually do?

  3. 03

    Outcome

    What changed, for whom, over what period, and compared with what baseline?

  4. 04

    Attribution

    How much of the change can reasonably be assigned to the intervention?

03 / Power test

Follow the constraint through the decision.

Every dossier answers the same six questions. A missing answer lowers the strength of the conclusion; political urgency does not fill the gap.

  1. 01

    Who sets the rule?

    Name the actor, authority, agreement, infrastructure, or ownership right.

  2. 02

    Who can contest it?

    Identify appeal, representation, collective bargaining, oversight, and practical voice.

  3. 03

    What does refusal cost?

    Measure lost finance, access, employment, safety, reach, or legal standing.

  4. 04

    What alternatives are real?

    Test switching cost, interoperability, time, capacity, and dependency—not theoretical choice.

  5. 05

    Who carries risk and loss?

    Separate decision authority from financial, social, and political exposure.

  6. 06

    What would disprove the claim?

    Publish the strongest counter-reading and the evidence that could change the finding.

04 / Evidence position

The label describes evidence—not truth by colour.

The public ledger contains only claims that have crossed the publication threshold. Provisional theses and undocumented leads remain in the private research register.

Disputed
Credible evidence or interpretations remain materially incompatible.
Supported
The conclusion is supported within an explicitly limited scope.
Mixed record
The direction or strength differs across cases, periods, or methods.
Not supported
The available record does not support the registered claim.

05 / Source treatment

Primary does not mean neutral.

A programme document may be the strongest evidence for its own conditions and still be insufficient evidence for social outcomes. A critical paper may identify genuine harm and still rely on a weak comparison. Source type and source position are different questions.

  • Record author, publisher, purpose, date, URL, local path, and file hash.
  • Cite the exact page, table, clause, series, or passage supporting the claim.
  • Write what the source establishes and what it does not establish.
  • Preserve material corrections, conflicts, methodological limits, and responses.
  • Never cite an AI response, search snippet, or working note as evidence.

06 / Causal language

Sequence is not cause. Benefit is not authorship.

The record statesDirectly present in a traceable primary document.
The data showA defined measurement with provenance and reproducible calculation.
The evidence suggestsAn inference supported but not uniquely established.
We have not establishedA material extension that the current record cannot support.

A beneficiary did not necessarily design the event. A motive did not necessarily cause it.

07 / Data and evidence files

Keep the original. Show the transformation.

Evidence handling separates preservation from publication. A cleaned or redacted publication copy is not a replacement for the source material.

  1. 01

    Preserve the original

    Store the received bytes unchanged with restricted access.

  2. 02

    Hash the original

    Record its SHA-256 digest, receipt time, source channel, and handler.

  3. 03

    Create a working copy

    Inspect it in an isolated environment chosen for the file type and threat.

  4. 04

    Prepare the publication copy

    Remove unnecessary metadata, redact protected details, or transform the working copy for publication.

  5. 05

    Hash and record again

    Give the publication copy its own SHA-256 digest and document every material transformation.

  • Units, currencies, prices, deflators, and population denominators are explicit.
  • Commitments, disbursements, stocks, flows, guarantees, and realised losses stay separate.
  • Missing values, revisions, definition breaks, and selected baselines remain visible.
  • Published charts retain the underlying table and processing record.

For the complete file-safety workflow, read the NKNW security guide.

08 / Corrections

The archive must show where it changed.

Material factual corrections are dated and explained. Silent edits are limited to typography or accessibility changes that do not alter meaning. A rejected claim stays in the ledger with its rejection reason instead of disappearing.

09 / Research privacy

Public-interest records are not permission to expose private lives.

Internal working notes and unnecessary personal data do not belong in the public build. Publication requires relevance, lawful handling, context, minimisation, and a clear distinction between public authority, public role, and private person.