fix(usage): stop extra-usage credit pools from marking a provider out of limit

[patroza]

Problem

With Claude at 5h 18% / weekly 78%, the provider dot and usage tooltip still showed red / "limit reached" — because the ai-usage daemon reports the extra-usage credit buffer as a monthly window sitting at 100% (Extra usage (off), €47.89 of €45).

That buffer is only drawn from once the plan windows are exhausted, so draining it (or having it switched off) says nothing about whether the provider is usable right now. usageMarkerForItem treated any window at 100% as a hard block. The DMS widget on the same feed renders that row as neutral context, which is the behaviour we want here.

Change

• isEnforcedUsageWindow() in the shared client-runtime presentation module: a window is context-only when the daemon flags it informational: true, when its label is a spend buffer (extra usage, credit(s), overage, on-demand, pay as you go), or when it is switched off ((off)).
• Marker fill, the outlook ring, the immediate-window pick, and worstUsagePercent now consider enforced windows only. Money-denominated plan windows still enforce (opencode-go monthly $, cursor monthly billing cycle).
• Web tooltip renders non-enforced rows with a neutral bar and muted value instead of a red bar; the red "limit reached" line no longer fires for them. Mobile picks this up automatically via the shared marker.
• Contract picks up the daemon's existing informational and limit window fields.

Tests

apps/web/src/aiUsageState.test.ts — the exact live claude/max feed shape (5h 18 / weekly 78 / Extra usage (off) 100) is no longer critical, grok's informational legacy $ pool no longer arms the outlook ring, a maxed real window still goes critical alongside extras, cursor's monthly billing window still enforces, plus direct isEnforcedUsageWindow coverage.

[patroza]

Problem

With Claude at 5h 18% / weekly 78%, the provider dot and usage tooltip still showed red / "limit reached" — because the ai-usage daemon reports the extra-usage credit buffer as a monthly window sitting at 100% (Extra usage (off), €47.89 of €45).

That buffer is only drawn from once the plan windows are exhausted, so draining it (or having it switched off) says nothing about whether the provider is usable right now. usageMarkerForItem treated any window at 100% as a hard block. The DMS widget on the same feed renders that row as neutral context, which is the behaviour we want here.

Change

• isEnforcedUsageWindow() in the shared client-runtime presentation module: a window is context-only when the daemon flags it informational: true, when its label is a spend buffer (extra usage, credit(s), overage, on-demand, pay as you go), or when it is switched off ((off)).
• Marker fill, the outlook ring, the immediate-window pick, and worstUsagePercent now consider enforced windows only. Money-denominated plan windows still enforce (opencode-go monthly $, cursor monthly billing cycle).
• Web tooltip renders non-enforced rows with a neutral bar and muted value instead of a red bar; the red "limit reached" line no longer fires for them. Mobile picks this up automatically via the shared marker.
• Contract picks up the daemon's existing informational and limit window fields.

Tests

apps/web/src/aiUsageState.test.ts — the exact live claude/max feed shape (5h 18 / weekly 78 / Extra usage (off) 100) is no longer critical, grok's informational legacy $ pool no longer arms the outlook ring, a maxed real window still goes critical alongside extras, cursor's monthly billing window still enforces, plus direct isEnforcedUsageWindow coverage.

[patroza]

Problem

With Claude at 5h 18% / weekly 78%, the provider dot and usage tooltip still showed red / "limit reached" — because the ai-usage daemon reports the extra-usage credit buffer as a monthly window sitting at 100% (Extra usage (off), €47.89 of €45).

That buffer is only drawn from once the plan windows are exhausted, so draining it (or having it switched off) says nothing about whether the provider is usable right now. usageMarkerForItem treated any window at 100% as a hard block. The DMS widget on the same feed renders that row as neutral context, which is the behaviour we want here.

Change

• isEnforcedUsageWindow() in the shared client-runtime presentation module: a window is context-only when the daemon flags it informational: true, when its label is a spend buffer (extra usage, credit(s), overage, on-demand, pay as you go), or when it is switched off ((off)).
• Marker fill, the outlook ring, the immediate-window pick, and worstUsagePercent now consider enforced windows only. Money-denominated plan windows still enforce (opencode-go monthly $, cursor monthly billing cycle).
• Web tooltip renders non-enforced rows with a neutral bar and muted value instead of a red bar; the red "limit reached" line no longer fires for them. Mobile picks this up automatically via the shared marker.
• Contract picks up the daemon's existing informational and limit window fields.

Tests

apps/web/src/aiUsageState.test.ts — the exact live claude/max feed shape (5h 18 / weekly 78 / Extra usage (off) 100) is no longer critical, grok's informational legacy $ pool no longer arms the outlook ring, a maxed real window still goes critical alongside extras, cursor's monthly billing window still enforces, plus direct isEnforcedUsageWindow coverage.

[patroza]

Problem

With Claude at 5h 18% / weekly 78%, the provider dot and usage tooltip still showed red / "limit reached" — because the ai-usage daemon reports the extra-usage credit buffer as a monthly window sitting at 100% (Extra usage (off), €47.89 of €45).

That buffer is only drawn from once the plan windows are exhausted, so draining it (or having it switched off) says nothing about whether the provider is usable right now. usageMarkerForItem treated any window at 100% as a hard block. The DMS widget on the same feed renders that row as neutral context, which is the behaviour we want here.

Change

• isEnforcedUsageWindow() in the shared client-runtime presentation module: a window is context-only when the daemon flags it informational: true, when its label is a spend buffer (extra usage, credit(s), overage, on-demand, pay as you go), or when it is switched off ((off)).
• Marker fill, the outlook ring, the immediate-window pick, and worstUsagePercent now consider enforced windows only. Money-denominated plan windows still enforce (opencode-go monthly $, cursor monthly billing cycle).
• Web tooltip renders non-enforced rows with a neutral bar and muted value instead of a red bar; the red "limit reached" line no longer fires for them. Mobile picks this up automatically via the shared marker.
• Contract picks up the daemon's existing informational and limit window fields.

Tests

apps/web/src/aiUsageState.test.ts — the exact live claude/max feed shape (5h 18 / weekly 78 / Extra usage (off) 100) is no longer critical, grok's informational legacy $ pool no longer arms the outlook ring, a maxed real window still goes critical alongside extras, cursor's monthly billing window still enforces, plus direct isEnforcedUsageWindow coverage.

[patroza]

Problem

With Claude at 5h 18% / weekly 78%, the provider dot and usage tooltip still showed red / "limit reached" — because the ai-usage daemon reports the extra-usage credit buffer as a monthly window sitting at 100% (Extra usage (off), €47.89 of €45).

That buffer is only drawn from once the plan windows are exhausted, so draining it (or having it switched off) says nothing about whether the provider is usable right now. usageMarkerForItem treated any window at 100% as a hard block. The DMS widget on the same feed renders that row as neutral context, which is the behaviour we want here.

Change

• isEnforcedUsageWindow() in the shared client-runtime presentation module: a window is context-only when the daemon flags it informational: true, when its label is a spend buffer (extra usage, credit(s), overage, on-demand, pay as you go), or when it is switched off ((off)).
• Marker fill, the outlook ring, the immediate-window pick, and worstUsagePercent now consider enforced windows only. Money-denominated plan windows still enforce (opencode-go monthly $, cursor monthly billing cycle).
• Web tooltip renders non-enforced rows with a neutral bar and muted value instead of a red bar; the red "limit reached" line no longer fires for them. Mobile picks this up automatically via the shared marker.
• Contract picks up the daemon's existing informational and limit window fields.

Tests

apps/web/src/aiUsageState.test.ts — the exact live claude/max feed shape (5h 18 / weekly 78 / Extra usage (off) 100) is no longer critical, grok's informational legacy $ pool no longer arms the outlook ring, a maxed real window still goes critical alongside extras, cursor's monthly billing window still enforces, plus direct isEnforcedUsageWindow coverage.

[patroza]

Problem

With Claude at 5h 18% / weekly 78%, the provider dot and usage tooltip still showed red / "limit reached" — because the ai-usage daemon reports the extra-usage credit buffer as a monthly window sitting at 100% (Extra usage (off), €47.89 of €45).

That buffer is only drawn from once the plan windows are exhausted, so draining it (or having it switched off) says nothing about whether the provider is usable right now. usageMarkerForItem treated any window at 100% as a hard block. The DMS widget on the same feed renders that row as neutral context, which is the behaviour we want here.

Change

• isEnforcedUsageWindow() in the shared client-runtime presentation module: a window is context-only when the daemon flags it informational: true, when its label is a spend buffer (extra usage, credit(s), overage, on-demand, pay as you go), or when it is switched off ((off)).
• Marker fill, the outlook ring, the immediate-window pick, and worstUsagePercent now consider enforced windows only. Money-denominated plan windows still enforce (opencode-go monthly $, cursor monthly billing cycle).
• Web tooltip renders non-enforced rows with a neutral bar and muted value instead of a red bar; the red "limit reached" line no longer fires for them. Mobile picks this up automatically via the shared marker.
• Contract picks up the daemon's existing informational and limit window fields.

Tests

apps/web/src/aiUsageState.test.ts — the exact live claude/max feed shape (5h 18 / weekly 78 / Extra usage (off) 100) is no longer critical, grok's informational legacy $ pool no longer arms the outlook ring, a maxed real window still goes critical alongside extras, cursor's monthly billing window still enforces, plus direct isEnforcedUsageWindow coverage.

[patroza]

Problem

With Claude at 5h 18% / weekly 78%, the provider dot and usage tooltip still showed red / "limit reached" — because the ai-usage daemon reports the extra-usage credit buffer as a monthly window sitting at 100% (Extra usage (off), €47.89 of €45).

That buffer is only drawn from once the plan windows are exhausted, so draining it (or having it switched off) says nothing about whether the provider is usable right now. usageMarkerForItem treated any window at 100% as a hard block. The DMS widget on the same feed renders that row as neutral context, which is the behaviour we want here.

Change

• isEnforcedUsageWindow() in the shared client-runtime presentation module: a window is context-only when the daemon flags it informational: true, when its label is a spend buffer (extra usage, credit(s), overage, on-demand, pay as you go), or when it is switched off ((off)).
• Marker fill, the outlook ring, the immediate-window pick, and worstUsagePercent now consider enforced windows only. Money-denominated plan windows still enforce (opencode-go monthly $, cursor monthly billing cycle).
• Web tooltip renders non-enforced rows with a neutral bar and muted value instead of a red bar; the red "limit reached" line no longer fires for them. Mobile picks this up automatically via the shared marker.
• Contract picks up the daemon's existing informational and limit window fields.

Tests

apps/web/src/aiUsageState.test.ts — the exact live claude/max feed shape (5h 18 / weekly 78 / Extra usage (off) 100) is no longer critical, grok's informational legacy $ pool no longer arms the outlook ring, a maxed real window still goes critical alongside extras, cursor's monthly billing window still enforces, plus direct isEnforcedUsageWindow coverage.

[patroza]

Problem

With Claude at 5h 18% / weekly 78%, the provider dot and usage tooltip still showed red / "limit reached" — because the ai-usage daemon reports the extra-usage credit buffer as a monthly window sitting at 100% (Extra usage (off), €47.89 of €45).

That buffer is only drawn from once the plan windows are exhausted, so draining it (or having it switched off) says nothing about whether the provider is usable right now. usageMarkerForItem treated any window at 100% as a hard block. The DMS widget on the same feed renders that row as neutral context, which is the behaviour we want here.

Change

• isEnforcedUsageWindow() in the shared client-runtime presentation module: a window is context-only when the daemon flags it informational: true, when its label is a spend buffer (extra usage, credit(s), overage, on-demand, pay as you go), or when it is switched off ((off)).
• Marker fill, the outlook ring, the immediate-window pick, and worstUsagePercent now consider enforced windows only. Money-denominated plan windows still enforce (opencode-go monthly $, cursor monthly billing cycle).
• Web tooltip renders non-enforced rows with a neutral bar and muted value instead of a red bar; the red "limit reached" line no longer fires for them. Mobile picks this up automatically via the shared marker.
• Contract picks up the daemon's existing informational and limit window fields.

Tests

apps/web/src/aiUsageState.test.ts — the exact live claude/max feed shape (5h 18 / weekly 78 / Extra usage (off) 100) is no longer critical, grok's informational legacy $ pool no longer arms the outlook ring, a maxed real window still goes critical alongside extras, cursor's monthly billing window still enforces, plus direct isEnforcedUsageWindow coverage.