Evidence before assertion
Every material claim starts with a traceable observation and source context.
Methodology / v2.4
A rigorous path from
signal to decision.
Recension’s methodology is a transparent, testable system for turning incomplete information into responsible intelligence. It makes evidence, assumptions, uncertainty, and change visible.
Standard
Methodology v2.4
Reviewed
30 Aug 2026
Cadence
Quarterly + event-driven
Scope
240+ markets / decision support
Our claim: this is decision support, not a promise of certainty, prediction, or universal coverage.
Operating principles
Rigor is a set of behaviors, not a badge.
Uncertainty is information
Known, assessed, unknown, and watch-for states stay distinct.
Challenge is a control
Alternative explanations and red-team review are part of the workflow.
Proportionate collection
Collection and retention match the decision, sensitivity, and legitimate purpose.
Human accountability
Automation accelerates discovery; accountable humans own consequential judgments.
Correction over concealment
When evidence changes, the record changes visibly with the reason and timestamp.
01 / Intelligence lifecycle
Six gates. One accountable chain.
Each gate creates an artifact that can be reviewed, challenged, and improved.
Gate 01 / 06
Frame
Translate a decision into a precise question, scope, owner, and horizon.
Artifact produced
Question brief + collection plan
Failure mode
A broad prompt produces an untestable answer.
Analyst questions
- What decision will this inform?
- What would make the answer useful?
02 / Evidence & provenance
A source is not a fact.
We preserve the difference between where information came from and what it supports. A reliable source can still report a weak claim; repeated reports can share one origin.
Reliability
Historical accuracy in the relevant domain.
Independence
Whether corroborating sources add new evidence.
Relevance
Direct bearing on the intelligence question.
Timeliness
Fit with the decision window and current conditions.
Claim record / illustrative
TraceableObservationReported port closure
Source classPublic + partner
Last verified18 min ago
Independent support2 sources
ConfidenceModerate
Next reviewTrigger-based
03 / Analytical model
Make the reasoning inspectable.
01
Observation / what was seen
02
Fact / validated observation
03
Inference / reasoned interpretation
04
Assessment / best-supported explanation
05
Implication / meaning for the decision
06
Trigger / indicator for action
Alternative hypotheses
Indicators & warnings
Assumption checks
Premortem analysis
Red-team challenge
Scenario analysis
04 / Confidence & calibration
Confidence carries conditions.
Confidence reflects evidence quality, convergence, and fit. It is not probability unless explicitly labeled as one.
HighEVIDENCE STATE
Multiple independent, relevant sources; few material assumptions.
ModerateEVIDENCE STATE
Good evidence with meaningful gaps or assumptions.
LowEVIDENCE STATE
Limited or conflicting evidence; plausible alternatives remain.
UnknownEVIDENCE STATE
Insufficient evidence to responsibly assess the claim.
05 / Quality controls
Review is part of the product.
Trace the evidence.
Review provenance, freshness, independence, permissions, and constraints.
Stress the reasoning.
Separate fact from inference, inspect assumptions, and test alternatives.
Keep change visible.
Version assessments, record corrections, and preserve decision context.
06 / Responsible intelligence
Capability requires restraint.
Use the least sensitive information necessary, respect lawful boundaries, and keep consequential interpretation under human accountability.
Proportionality
Purpose-bound, reviewable, and accountable by default.
Human oversight
Purpose-bound, reviewable, and accountable by default.
Privacy by design
Purpose-bound, reviewable, and accountable by default.
Governance
Purpose-bound, reviewable, and accountable by default.
Not legal advice: product controls do not replace legal, compliance, privacy, or security review.
07 / Limitations
Know where the method can break.
Incomplete coverage
No system sees everything.
MITIGATION
Disclose unknowns and use alternate sources.
Manipulated sources
Evidence can be manufactured.
MITIGATION
Check origin, incentives, and corroboration.
Context loss
Translation can change meaning.
MITIGATION
Retain original context and review ambiguity.
Entity ambiguity
Similar names create false matches.
MITIGATION
Keep candidate matches and resolution history.
Correlated evidence
Reports may share one origin.
MITIGATION
Model source dependence.
Changing conditions
Valid assessments can go stale.
MITIGATION
Use triggers and visible reassessment.
08 / Living standard
The method updates with the world.
We review quarterly and after material events, customer challenges, incident learnings, or capability changes.
Monitor
Signals surface a change.
Review
Owners test the standard.
Publish
Approved changes are versioned.
09 / Field guide
The language we use.
Signal
A change that may matter.
Provenance
The trace from claim to source.
Indicator
A sign that supports or challenges an assessment.
Scenario
A coherent possible future, not a forecast.
Coverage
The observable portion of a question.
Assessment
A reasoned judgment with confidence.
Trigger
A condition prompting action or review.
Unknown
A material question evidence cannot answer.
Questions we welcome
Inspect the method.
A strong methodology should survive scrutiny. Ask us to show the chain.
Who owns the methodology?
Recension maintains the methodology as a living standard. Changes are reviewed, versioned, and communicated when they affect interpretation.
Does confidence mean probability?
Not by default. Confidence describes the strength and convergence of evidence. A modeled probability is labeled separately with its assumptions.
How transparent are sources?
Each material conclusion carries provenance appropriate to the source and customer permissions. Reliability, independence, freshness, and confidence remain distinct.
What happens when a conclusion is wrong?
Users can challenge a claim or source. The issue is triaged, re-reviewed, and corrected or qualified with the decision trail intact.
Where does automation stop?
Automation helps discover, normalize, compare, and monitor signals. Human review remains accountable for consequential judgments.
What are the limits?
Coverage can be incomplete, sources can be manipulated, entities can be ambiguous, and conditions can change faster than collection.