Naming Convention
Section titled “Naming Convention”The problem
Section titled “The problem”The framework has five kinds of named entities: criteria, metrics, calibrations, protocols, and synthesis protocols. Legacy IDs use short codes — C5, A01, M4 — that are ambiguous across entity types. “C5” could refer to criterion C5 (convergent validity), calibration C5 (mediation analysis), or a metric with the C5 prefix. “A01” could be protocol A01 (Pearl SCM) or a metric in the causal family. When cross-referencing between pages or writing code that registers entities, these collisions produce confusion.
The convention
Section titled “The convention”Each entity type has a typed prefix that makes the ID unambiguous:
CRIT — Criteria
Section titled “CRIT — Criteria”CRIT-C1 through CRIT-V5 (27 total). The sub-ID preserves the validity-type letter:
- C = Construct validity (C1 Falsifiability, C2 Structural plausibility, C3 Task specificity, C4 Minimality, C5 Convergent validity)
- M = Measurement validity (M1 Reliability, M2 Invariance, M3 Baseline separation, M4 Sensitivity, M5 Calibration, M6 Construct coverage)
- I = Internal validity (I1 Necessity, I2 Sufficiency, I3 Specificity, I4 Consistency, I5 Confound control)
- E = External validity (E1 Intervention reach, E2 Graded response, E3 Selectivity, E4 Effect magnitude, E5 Robustness, E6 Cross-architecture)
- V = Interpretive validity (V1 Level declaration, V2 Level-evidence match, V3 Narrative coherence, V4 Alternative exclusion, V5 Scope honesty)
MET — Metrics
Section titled “MET — Metrics”MET-activation-patching, MET-das-iia, MET-mutual-information, etc. Full kebab-case slug, never abbreviated. The slug is the canonical name used in code and frontmatter. Examples:
| Typed ID | What it refers to |
|---|---|
MET-activation-patching | Activation patching metric (C2 on the MI Causal page) |
MET-das-iia | DAS interchange intervention accuracy (C1 on MI Causal) |
MET-rsa | Representational similarity analysis (E03 on MI Representational) |
MET-caa-steering | Contrastive activation addition (C09 on MI Steering) |
MET-axbench | AxBench concept detection and steering (B20 on MI Benchmarks) |
CAL — Calibrations
Section titled “CAL — Calibrations”CAL-01 through CAL-16 (16 total). Sequential numbers corresponding to the calibration checklist. Examples:
| Typed ID | Name | Legacy ID |
|---|---|---|
CAL-01 | Bootstrap Stability | C11 |
CAL-02 | Convergent Validity | C12 |
CAL-03 | Measurement Invariance | C13 |
CAL-04 | Derived Metrics (Sensitivity) | C14 |
CAL-05 | Reliability Suite | C16 |
PROT — Protocols
Section titled “PROT — Protocols”PROT-{letter}{number} format, where the letter indicates the evidence family. Examples:
| Typed ID | Name | Legacy ID |
|---|---|---|
PROT-A01 | Pearl SCM | A01 |
PROT-A02 | Counterfactual DAS/IIA | A02 |
PROT-B01 | Spectral/SVD | B01 |
PROT-D01 | Faithfulness | D01 |
SYN — Synthesis protocols
Section titled “SYN — Synthesis protocols”SYN-01 through SYN-09 (9 total). Sequential numbers. Examples:
| Typed ID | Name |
|---|---|
SYN-01 | Functional Parcellation |
SYN-02 | Dawid-Skene Consensus |
SYN-03 | Robust Rank Aggregation |
In prose
Section titled “In prose”Informal references are fine in running text. You can write “I3 Specificity” or “criterion C5” or “protocol A01” without the typed prefix — context makes the type clear. The typed prefix is the canonical ID used in:
- Code (metric registry keys, config files, result schemas)
- Frontmatter (
id:fields in page metadata) - Formal cross-references (links between pages, summary tables)
- Machine-readable outputs (JSON results, scoring pipelines)
When writing prose, use whichever form reads most naturally. When writing code or structured data, use the typed prefix.
Legacy IDs
Section titled “Legacy IDs”Existing pages use untyped IDs (C1, M4, A01, EX12). These will be migrated to typed IDs over time. When encountering an untyped ID, context determines the type:
- If it appears on a validity-type page (construct, measurement, internal, external, interpretive), it is a criterion.
- If it appears on a metric page (mi-causal-metrics, mi-structural-metrics, etc.), it is a metric.
- If it appears on the calibrations page, it is a calibration.
- If it appears on the protocols page, it is a protocol.
- IDs starting with EX are evaluation metrics (a subset of MET).
- IDs starting with S followed by a number on the synthesis protocols page are synthesis protocols.
During the migration period, both forms are valid. The typed form is preferred for new content.
Quick reference
Section titled “Quick reference”| Prefix | Entity type | ID format | Count | Example |
|---|---|---|---|---|
CRIT | Criterion | CRIT-{letter}{number} | 27 | CRIT-I3 (Specificity) |
MET | Metric | MET-{kebab-slug} | 130+ | MET-activation-patching |
CAL | Calibration | CAL-{number} | 16 | CAL-01 (Bootstrap Stability) |
PROT | Protocol | PROT-{letter}{number} | 81 | PROT-A01 (Pearl SCM) |
SYN | Synthesis protocol | SYN-{number} | 9 | SYN-02 (Dawid-Skene) |