Result Schema

Result Schema arachne.result/1.0

Field	Meaning
status, messages[]	ok / insufficient_evidence / error; messages have level, code, text
config_hash, rubric_version, engine_version, config	Exactly what produced this result
reference, participant	Ids, durations, alignment_method; participant quality (SNR estimate, voiced coverage, speech seconds, clipping)
words[]	Per word: reference and participant each with start, end, pause_after (seconds). Use for transcript highlighting and audio seeking
regions[]	Candidate flaw regions (below)
scores	overall; dimensions (pacing, pauses, pitch_contour, energy, null if withheld); weights; formula. null if evidence is insufficient
global	e.g. global_pace_ratio (above 1 = participant faster overall)
features	Per-word raw and speaker-normalised values for both recordings, plus units
timeseries	rate_hz (50) and per recording: energy_db, f0_semitones (null = unvoiced), speech_mask. Time = index / rate_hz
rules, limitations	Rule metadata and the caveats to display

Region object

Field	Meaning
region_id, rule_id, dimension, direction	Identity and what kind of finding
start_s, end_s	Span on the participant's clock
word_range, words	Word indices and text
measured	Rule-specific measured values vs reference (e.g. part_wpm, ref_wpm)
delta, delta_unit	Size of deviation (s, dB, or log_ratio)
z	Robust z-score
severity, severity_label	0 to 1, and mild / moderate / strong
explanation	Template text naming feature, comparison, span and rule

Notes for UI authors

• Reference and participant have different clocks. Map between them through words[].
• Pause regions of type "missing" are drawn as the junction plus or minus 0.2 s, by convention.
• Show limitations somewhere visible. Regions are measurements, not verdicts.