Addendum v0.7 · Universal Analytics & Human Review Layer
If Trades has the authorized data, a human can interrogate it.
Compare it, model it, visualize it, and generate a reviewable report — without engineering work for every new question. Analytics informs humans. It does not convert probability into fact or automatically authorize high-consequence action.
Four analytical levels
| Level | Question answered |
|---|---|
| Descriptive | What happened? |
| Diagnostic | Why did it happen? |
| Predictive | What is likely to happen next, with what probability / uncertainty? |
| Prescriptive | What actions should humans consider, what are the tradeoffs, and what is predicted under each? |
Human-review outputs
Reports, graphs, heat maps / matrices, scales, scorecards, probability reports, forecasts, models, performance reports, and solution reports. Probability must never be presented as certainty. Distinguish prediction confidence, evidence strength, source quality, agreement, verification, sample size, and calibration history.
Mandatory report metadata
| Field | Required disclosure |
|---|---|
| Scope / filters | Exactly what records, branches, roles, modes, and dates were included or excluded. |
| Data sources | ServiceTitan / ProBooks / Trades-native / connector / manual or other authorized sources. |
| Sample size | Jobs / events / observations supporting the result. |
| Formula / methodology | How the metric, scale, savings estimate, or probability was calculated. |
| Assumptions | Operational / model assumptions that materially affect interpretation. |
| Confidence / uncertainty | Model confidence, evidence strength, and interval / range. |
| Status | Observed, estimated, attributed, reconciled/realized, simulated, or unresolved. |
| Version lineage | Model, ruleset, policy, pricebook, and report-definition version. |
| Receipt references | Traceable lineage to underlying decision / evidence records. |
Drill-down
Aggregate → cohort → event/job → decision receipt → evidence packet → source/version.
A visualization is not an opaque truth source.
Anti-misleading rules
Flag tiny samples, missing data, stale sources, selection bias, incomparable cohorts, estimated values, wide uncertainty, and model drift. Estimated savings must never display as realized. Correlation is not causation without evaluation. Rankings show sample size and context.
Frozen requirement
Any authorized metric must be capable of human-readable reporting, visualization, comparison, modeling, and evidence-backed review at the appropriate scope. No important model output should exist only as an internal machine value that management cannot inspect and challenge.