Estimative Language
What PIE's numbers and qualifiers actually mean. Every flag, prediction, and brief judgment uses the fixed lexicon on this page — modeled on the U.S. Intelligence Community's analytic standards (ICD 203) — so a term means the same thing every day, on every surface.
Each flag carries a confidence score (0–100%).
It grades the evidence chain, not the probability of an outcome.
| Score | Tier | What it means |
|---|---|---|
| 90–100% | PRIMARY SOURCE | Verified from government records, teardown reports, court filings, or direct documentation. |
| 80–89% | STRONG INFERENCE | Multiple corroborating sources; the conclusion is well-supported. |
| 60–79% | ANALYTICAL | Derived from parts-db and supply-chain analysis; reflects PIE's methodology and its limits. |
| <60% | SIGNAL | Early indicator. Treat as a lead to investigate, not a conclusion. |
Predictions carry a numeric probability.
When the brief or a flag uses a likelihood word, it maps to these bands — the
standard ICD 203 spectrum:
| Probability | Term |
|---|---|
| <5% | remote / almost no chance |
| 5–20% | very unlikely |
| 20–45% | unlikely |
| 45–55% | roughly even odds |
| 55–80% | likely |
| 80–95% | very likely |
| >95% | almost certain |
Prediction confidence is a separate field:
it reflects inter-model agreement in the ensemble (and cross-validation across LLM
providers), not the probability of the event. How well-calibrated these numbers have
been is tracked publicly on
Forecast Accountability.
| Level | Basis |
|---|---|
| HIGH confidence | Multiple independent sources, at least one primary; minimal reliance on assumptions. |
| MODERATE confidence | Credibly sourced and plausible, but single-threaded or partly inferred. |
| LOW confidence | Fragmentary or uncorroborated sourcing; significant analytic assumptions. |
Major judgments state their basis inline — e.g. "moderate confidence; derived from parts-db concentration plus one reporting source."
| Type | Meaning |
|---|---|
| primary | Government record, filing, official document, or physical teardown. |
| reporting | Journalism or trade-press coverage of the event. |
| aggregated | Compiled from multiple feeds (e.g. distributor stock APIs, contract databases). |
| derived | Computed by PIE's own analysis (concentration ratios, correlations, graph traversal). |
| estimated | Modeled estimate where direct data is unavailable. |
Risk labels are analytic signals, not allegations. Full methodology — flag taxonomy, gray-zone scoring dimensions, and known data gaps — is published at uas-handbook.com. Machine access to all underlying data: data API reference.