# Confidence and conclusion framework

**Document ID:** NW-FA-CF-001
**Version:** 1.0
**Adopted:** 2026-08-04

This framework was adopted after the initial public analysis. Existing speech
conclusions are therefore provisional until they are rescored under this
framework and independently reviewed.

## Separate technical findings from speech interpretations

Technical conclusions may be based on repeatable measurement. Speech
interpretations require additional perceptual and phonetic support.

## Evidence dimensions

Each speech interpretation must be assessed across five dimensions:

1. **Raw observability** — perceptible in raw stereo or raw isolated channel.
2. **Treatment stability** — remains substantially consistent across
   conservative disclosed treatments.
3. **Independent agreement** — converges in free responses from listeners who
   were not shown the proposed phrase.
4. **Phonetic compatibility** — observable timing, voicing, syllabic, burst,
   or frication features are compatible with the proposed wording.
5. **Alternative explanations** — competing interpretations and
   codec/noise artifacts have been considered.

## Categories

### Confirmed technical finding

Use only for an objective property reproduced by measurement or direct file
inspection, such as channel count, sample rate, duration, or a repeatable
channel-routing result.

### Strongly supported speech

All of the following are required:

- audible in raw material;
- stable across conservative treatments;
- substantial independent free-response convergence;
- compatible with observable phonetic structure;
- no comparably supported alternative.

This category is unavailable until independent listening data exist.

### Probable speech

The essential wording is audible in raw or raw isolated audio, remains stable
under conservative treatment, and is supported by either independent
agreement or clear phonetic compatibility. Limited unresolved segments or
alternatives may remain.

Until independent review is completed, this must be labeled
**provisional probable**.

### Plausible or possible speech

The proposed wording is compatible with some audible features, but substantial
alternatives, masking, prompting effects, or unresolved segments remain.

### Unintelligible

No stable wording survives raw comparison, conservative processing, and
unprompted review. Fragments may be audible without supporting a complete
phrase.

### Not confirmed

Use for a specific public claim that cannot be reproduced as a stable
utterance. This does not prove the word was never spoken elsewhere in the
recording.

### Unsupported

Use only when the proposed wording depends principally on prompting,
aggressive processing, generated material, or isolated model output and lacks
repeatable support in the examined audio.

## Required wording

Reports must identify:

- the category;
- whether the result is provisional or independently reviewed;
- the exact region;
- the evidence dimensions satisfied;
- major alternatives;
- the source limitation.

Terms such as “definitely,” “proves,” and “clearly says” are prohibited unless
the evidence meets the applicable criteria and the source limitation does not
undermine the claim.