Wire stats

How much, what kinds, how reliable.

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

Wires published

109

all severity, all sources

Retraction rate

0.00%

90-day · 0/109 all-time

Right-of-reply median

5.3h

n=3 · sent → received

Cross-model verified

0%

0 of 109 have ≥2 models

Severity distribution

  1. SEV-1
    2 1.8%
  2. SEV-2
    7 6.4%
  3. SEV-3
    79 72.5%
  4. SEV-4
    21 19.3%

Volume by month

1155745 25/0425/0526/0526/0626/11

Last 5 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
    33
  2. Anthropic
    18
  3. Microsoft
    7
  4. Google
    7
  5. Zhipu AI
    4
  6. Mistral
    2
  7. Meta
    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
    46
  2. Model drift cron
    30
  3. Reddit / HN
    24
  4. Court / regulator
    4
  5. Red team
    1

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 592
  2. warn house_style.word_count_low 344
  3. info house_style.model_name_check 109
  4. warn house_style.raw_url 9
  5. warn house_style.marketing_language 6
  6. warn house_style.editorialising_framing 4
  7. error house_style.provider_attribution_missing 3
  8. error house_style.editorialising_adjective 1

Aggregated from every wire's auto-pipeline lint pass. Errors stop a story from publishing; warnings are recorded but do not stop it. Rule definitions: the house style rules.