Source
Show
Content
| Time | Code | Status | Cost |
|---|
No calls match this filter.
Source
—
Spend by source
Spend by provider
Spend by code
Shadow enforcement — what it would have saved
— calls
—
Failure groups
| Source | Code | Class | Count | Last seen | Last error |
|---|
Shadow would-block events are not in this table — those calls succeeded. See Cost → Shadow enforcement for that picture.
Suppressed by kill-switch
A suppressed call costs nothing and reaches no provider — distinct from both failures and shadow blocks.
Degraded call-sites — fallback in use
A fallback means the primary provider failed upstream — the call succeeded, but on plan B. Filter the Feed with the fallbacks chip to see each one.
| Code | 24 h volume · hourly | State |
|---|
Hover a code name for what it actually does. NO CALL SITE = configured but nothing sends it traffic (calls fall through to OBDQUERY) — arming or re-routing those changes nothing until a call site exists. Click a row to edit its route. An amber dashed line on a sparkline is that code's alert cap (alert_calls_per_hour in ai-routes.json); buckets above it render amber.
Provider price reference
List prices per 1M tokens, USD · read from /etc/fulcrum/ai-providers.json · loading…
Reference · manual refresh · not live telemetry
Last updated
—
| Model | Input $ / 1M | Output $ / 1M | Price status |
|---|
This tab is a spec sheet, not a feed — a human re-runs the ai-pricing-refresh skill once in a while and it rewrites the JSON; nothing here polls. UNVERIFIED = the price could not be confirmed on the provider's official pricing page: an ≈ amber figure is a best-effort number to double-check, “—” means no number was published at all — never treat either as billing truth. Routed codes is derived from the live Codes tab state, so it follows route edits. Actual per-call cost in Feed / Cost is computed at call time by the proxy, not from this table.