Version: 1.0-Provisional, Implementation Status: Active
1. Purpose
The Parameter Governance Methodology (PGM) defines the governance rules for documenting, classifying, maintaining, and revising parameters used in parameterized models.
PGM does not determine whether a model is scientifically correct. Instead, it governs how parameter values are recorded, traced, revised, and evaluated over time so that model evolution remains transparent, reproducible, and auditable.
PGM is intentionally model-agnostic and may be applied to any parameterized analytical or computational model.
Design Maxim: Every undocumented decision eventually becomes a bug.
2. Scope
PGM applies to any parameter that influences model behavior through a defined quantitative value.
Examples include:
- constants
- coefficients
- thresholds
- decay rates
- lookup values
- weighting factors
- scaling functions
- probability values
PGM does not define:
- model architecture
- assessment methodologies
- validation methodologies
- mathematical algorithms
Those are governed by separate artifacts.
3. Definitions
3.1 Parameter
A parameter is a documented quantitative value that influences the behavior of a model.
3.2 Provenance
The documented origin of a parameter value.
Examples include:
- derived mathematically
- adopted from published research
- engineering estimate
- AI-generated estimate
- expert judgment
- arbitrary convention
3.3 Evidence Ledger
The structured record maintained for every governed parameter.
The Evidence Ledger records parameter metadata such as provenance, confidence, evidence status, revision history, and governance classification. It documents the current state of knowledge about a parameter but does not itself validate that parameter.
3.4 Confidence
Confidence represents how strongly the current parameter value is believed to approximate reality.
Confidence is distinct from evidence.
High confidence without supporting evidence SHOULD be explicitly justified in ledger notes.
3.5 Sensitivity
Sensitivity describes how strongly changes to a parameter influence model outputs.
Sensitivity is normally determined through model execution and sensitivity analysis rather than documentation review.
3.6 Independence
Evidence is independent when it originates from sources outside the model being evaluated.
Evidence derived directly or indirectly from a model’s own outputs SHALL NOT be considered independent validation.
4. General Principles
4.1 Documentation Before Authority
Parameters SHALL be documented before they acquire operational authority.
4.2 Unknown Is Acceptable
Unknown provenance, confidence, or evidence status SHALL be explicitly recorded rather than inferred.
4.3 Documentation Is Not Validation
Recording a parameter does not validate it.
The Evidence Ledger records the current governance state of a parameter.
Scientific validation is governed separately.
4.4 Provenance Is Mandatory
Every governed parameter SHALL include documented provenance.
5. Parameter Classes
Type I — Derived Parameter
Completely determined from existing definitions or mathematics.
Examples:
- unit conversions
- mathematical constants
- formula outputs
Type II — Conventional Parameter
Established by documented convention rather than empirical claim.
Examples:
- encoding values
- categorical mappings
- enumerations
- display conventions
Type III — Engineering Prior
A reasoned initial estimate used until stronger evidence becomes available.
Engineering Priors are expected to evolve.
Type IV — Empirical Parameter
Supported by independent evidence documenting observed behavior.
Promotion from Type III to Type IV REQUIRES documented independent evidence satisfying Section 12 and SHALL be recorded in the Evidence Ledger.
6. Evidence Levels
- None
- Proposed
- Supported
- Corroborated
- Validated
These evidence levels describe the current evidentiary status of a parameter and are independent of parameter class.
7. Provenance Categories
Examples include:
- Mathematical derivation
- Published research
- Published framework specification
- Engineering estimate
- AI-assisted estimate
- Expert judgment
- Historical calibration
- Arbitrary convention
8. Anchorability
Anchorability describes whether a parameter is believed capable of future external evidentiary support.
Possible states include:
- Directly measurable
- Latent but externally constrainable
- Structural modeling choice
- Unknown
A parameter believed to have no possible independent external constraint SHALL be classified as a Structural Modeling Choice rather than as Latent but Externally Constrainable. Unknown remains an acceptable temporary classification where future anchorability has not yet been determined.
PGM records the declared anchorability status but does not determine whether external anchoring has been scientifically achieved.
9. Evidence Ledger Fields
Each governed parameter SHOULD include:
- Name
- Description
- Current Value
- Units
- Parameter Class
- Evidence Level
- Provenance
- Confidence
- Sensitivity (if known)
- Anchorability
- Revision Authority
- Independence Status
- None
- Model-internal only
- Independent evidence available
- Revision History
- Supporting Notes
Additional implementation-specific fields MAY be added.
10. Revision Authority
Different parameter classes possess different revision authority.
Type I
May change only to correct mathematical or transcription errors.
Type II
May change through documented governance revision.
Type III
May evolve as new evidence accumulates.
Type IV
Requires documented independent evidence supporting revision.
11. AI-Generated Parameters
Parameters proposed by AI SHALL disclose:
- model used
- generation date
- reasoning summary
- provenance classification
AI-generated estimates SHALL NOT be represented as empirical evidence.
12. Independence Requirement
Evidence used for validation SHALL originate independently of the model under evaluation.
Evidence derived from the model’s own outputs SHALL NOT constitute independent validation.
13. Relationship to Validation
PGM governs parameter documentation and governance.
Scientific validation—including external corroboration, over-determination testing, held-out case evaluation, and model performance—is governed by the separate Validation Methodology (VM).
14. Relationship to Assessment
PGM does not govern the assessment of real-world events into model inputs.
That function belongs to the separate Event Assessment Rubric (EAR), which standardizes event interpretation and input generation.
15. Compliance
A model complies with PGM when:
- governed parameters are documented
- provenance is recorded
- parameter class is declared
- evidence status is recorded
- revision authority is defined
- revisions are traceable
- governance requirements are satisfied
Compliance with PGM does not imply scientific validity.
| Version | Date | Summary |
|---|---|---|
| 1.0-Provisional | 2026-06-27 | Initial publication. Establishes parameter classification, provenance, evidence levels, anchorability, revision authority, independence requirements, and governance principles. |