Wire stats

How much, what kinds, how reliable.

Public counterpart to the transparency ledger. Refreshed every 10 minutes.

Wires published

357

all severity, all sources

Retraction rate

0.00%

90-day · 0/357 all-time

Right-of-reply median

24.0h

n=2 · sent → received

Cross-model verified

0%

0 of 357 have ≥2 models

Severity distribution

  1. SEV-1
    18 5.0%
  2. SEV-2
    18 5.0%
  3. SEV-3
    321 89.9%
  4. SEV-4
    0 0.0%

Volume by month

11251104 25/0425/0526/0526/06

Last 4 months. Bar height = wires published. Counts incident publication dates, not first-seen dates — so launch-archive backfill stacks on a single month.

Top providers

  1. OpenAI
    256
  2. Anthropic
    16
  3. Google
    15
  4. Meta
    6
  5. Mistral
    2

Provider names extracted from the wire's providers_named array. A wire can be tagged with multiple providers; counts double-count by design.

Source mix

  1. Provider blog
    259
  2. Reddit / HN
    44
  3. Model drift cron
    33
  4. Court / regulator
    17

Where each wire's first signal came from. extension = LiveScope reader telemetry; tip_line = reader-submitted via /wire/tip; the rest are automated cron sources.

House-style lint — top findings

  1. warn house_style.provider_name_drift 382
  2. info house_style.model_name_check 99
  3. warn house_style.word_count_low 6
  4. warn house_style.marketing_language 5
  5. warn house_style.editorialising_framing 3
  6. error house_style.provider_attribution_missing 3
  7. warn house_style.raw_url 3
  8. error house_style.editorialising_adjective 1

Aggregated from every wire's auto-pipeline lint pass. Errors gate publish; warnings surface in the editor's checklist for judgement. Rule definitions: source on GitHub.