This release includes 1 breaking change for platform teams planning a safe upgrade.
✓ No known CVEs patched in this version
Topics
+5 more
ReleasePort's take
Light signalForecast model v2.1 eliminates the CI inversion cap, exposing raw p10/p90 intervals without limits.
Why it matters: Affects forecast calculations; developers must adjust any code that relied on capped interval values in version v0.14.0.
Summary
AI summaryForecast model v2.1 removes CI inversion by reporting raw Monte Carlo p10/p90 intervals.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Breaking | High |
Forecast model v2.1 removes CI inversion cap; interval now raw p10/p90 uncapped. Forecast model v2.1 removes CI inversion cap; interval now raw p10/p90 uncapped. Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Feature | Medium |
Dashboard now shows update pill when a newer claumon release is available. Dashboard now shows update pill when a newer claumon release is available. Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Feature | Low |
Forecast popups resized from 640px to 880px wide; chart SVG scales accordingly. Forecast popups resized from 640px to 880px wide; chart SVG scales accordingly. Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Feature | Low |
Claude Fable 5 pricing added ($10/MTok input, $50/MTok output, cache conventions). Claude Fable 5 pricing added ($10/MTok input, $50/MTok output, cache conventions). Source: llm_adapter@2026-06-11 Confidence: high |
— |
Full changelog
Forecast model v2.1, update badge, Fable 5 pricing
-
Forecast model v2.1: the 80% CI can no longer invert. The reported
interval is now the raw Monte Carlo terminal p10/p90, uncapped. v2.0 capped
only the upper edge at 100%, so when projected demand blew past the limit
the popup could show nonsense like "80% CI 134%-100%" (lower above upper).
The forecast measures demand, which legitimately exceeds the window limit;
overshoot magnitude is signal and the gauge ring still saturates at 100%.
Reporting-convention change only: the generative model, calibration, MC,
and ETA are untouched. Spec updated (MODEL.tex/pdf), v2.0 archived under
internal/forecast/archive/v2.0/, regression test added. -
The dashboard tells you when a newer claumon is out. A background check
polls GitHub for the latest release shortly after startup and then daily,
reusing the existing updater plumbing. When the running build is behind, an
accent-colored pill appears next to the version in the top bar, linking to
the releases page; live pages light up via anupdate_availableSSE event
without a reload./api/infonow reportsupdate_available,
latest_version, andreleases_url. Best-effort and silent on failure;
dev builds never trigger it. -
Claude Fable 5 pricing. Cost estimates now cover
claude-fable-5
($10/$50 per MTok input/output, cache prices following the usual convention:
0.1x read, 1.25x 5-minute write, 2x 1-hour write). Synced to the embedded
fallback; pricing date bumped to 2026-06-10. -
Bigger forecast popups. The session and weekly trajectory modals grew
from 640px to 880px wide (still capped at 94vw on small screens); the chart
SVG scales with them, so the simulated paths and axis labels are easier to
read. -
README and landing page repositioned. The pitch now leads with the gap
claumon fills: Anthropic's usage analytics dashboard is for Team/Enterprise
org admins, not individual Pro/Max plans. New "How it compares" table vs
ccusage, Claude-Code-Usage-Monitor, and claude-usage, and the forecast
model spec (MODEL.pdf) is surfaced on both pages. No functional changes.
No forecast-model or storage changes.
Breaking Changes
- Forecast model v2.1 no longer inverts the 80% CI; it now reports raw Monte Carlo terminal p10/p90 values instead of capping at 100%.
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About Claumon
All releases →Related context
Beta — feedback welcome: [email protected]