Scoring Formula
How evidence becomes a score — the IndexTheory penalty model.
The Formula
score = max(floor, 100 − Σ min(c(τ) · p, c(τ)))
where:
- τ = evidence tier (T1–T5)
- c(τ) = tier coefficient — the maximum penalty a single piece of evidence can contribute at tier τ
- p = penalty signal from 0.0 (no violation) to 1.0 (full violation), determined by the evidence's confidence and claim type
- floor = per-index minimum — prevents a score of zero from low-tier evidence alone
The formula is penalty-from-ceiling: every index starts at 100. Each piece of approved evidence subtracts a bounded penalty. Higher-tier evidence (T4, T5) produces larger penalties per piece. The total cannot exceed the maximum contribution of each piece — evidence of the same violation from multiple low-tier sources cannot compound past T5-equivalent weight.
Tier Coefficients
| Tier | Source type | Max penalty per evidence |
|---|---|---|
| T1 | Screenshot / personal account | ≤ 5 pts |
| T2 | Journalism / industry survey | ≤ 10 pts |
| T3 | Academic research paper | ≤ 20 pts |
| T4 | Platform transparency report | ≤ 35 pts |
| T5 | Regulatory filing / court order / consent decree | ≤ 60 pts |
Actual coefficients are subject to calibration revision. See calibration status →
Score Floors
Each index has a floor value determined by the highest evidence tier present. The floor prevents a score from collapsing solely on the basis of weak or anecdotal evidence.
| Highest tier in scan | Score floor |
|---|---|
| T1 or T2 only | 60 — score cannot drop below 60 |
| T3 present | 40 — research-level evidence allows deeper descent |
| T4 present | 20 — platform admissions allow significant descent |
| T5 present | 0 — legally binding findings impose no floor |
Composite Scores
The Platform Signal Integrity Index (PSII) and Platform Confidence Index (PCI) are composites of multiple sub-indexes. Composite scores are weighted averages of their component indexes. The weights applied are defined in the index specification documents and may be revised during calibration.
If any component index lacks sufficient evidence to score, it does not contribute to the composite with a zero — it is excluded from the weighted average entirely. This prevents a platform with no coverage on one dimension from being artificially penalized for the absence of evidence.
Calibration Status
Calibrated — Protocol-6
Coefficient values, floor thresholds, and composite weights have been empirically calibrated in TheoryLab via Protocol-6. Constants governing tier penalty ceilings, score floors, evidence trust, and component weights are canon-locked.
Scores should be read as research intelligence — relative indicators of documented violation patterns, not absolute measurements. Not suitable for legal, hiring, or consequential decisions.