Provider health
Latency, error rate, and cost per provider over the last 24 hours.
See also: /insights/calibration — weekly Brier + CORP reliability per classifier_version.
NLP fallback rate (last 24h)
0.0%
L'm-fallback share of (finbert-hf + lm-fallback) successful NLP calls. Alert at > 5%.
Active alerts (4) — per D-19
- reddit-xpoz — 53.5% error rate (53/99)[UPSTREAM_5XX]23h ago
- polygon — 40.8% error rate (29/71)[AUTH_FAILED]43d ago
- anthropic-search — 30.8% error rate (24/78)[UNKNOWN]46d ago
- twitter-xpoz — 76.6% error rate (49/64)[UNKNOWN]53d ago
Fallback heatmap (last 24h) — per D-10
- hackernews
- 0.0% (12)
- reddit-xpoz
- 0.0% (132)
- stocktwits
- 0.0% (24)
- twitter-xpoz
- 0.0% (12)
Calibration (Plan 20-B-03)
hackernews
12 calls / 24h
- p50
- 274 ms
- p95
- 554 ms
- p99
- 650 ms
- error
- 0.0%
- cache hit
- 0.0%
- fallback
- 0.0%
- cost / call
- $0.0000
- cost / 24h
- $0.0000
reddit-xpoz
132 calls / 24h
- p50
- 5285 ms
- p95
- 12003 ms
- p99
- 12004 ms
- error
- 9.8%
- cache hit
- 0.0%
- fallback
- 0.0%
- cost / call
- $0.0009
- cost / 24h
- $0.1190
stocktwits
24 calls / 24h
- p50
- 134 ms
- p95
- 244 ms
- p99
- 492 ms
- error
- 0.0%
- cache hit