[claude] Wire QuotaMonitor.select_model() into cascade router (#1106) #1113

claude · 2026-03-23T18:12:54Z

claude commented

2026-03-23 18:12:54 +00:00

Fixes #1106

Summary

Updates CascadeRouter._quota_allows_cloud() to use QuotaMonitor.select_model("high") instead of should_use_cloud(), aligning it with the metabolic protocol thresholds:
- BURST (5h quota < 50%): cloud providers allowed
- ACTIVE (5h quota ≥ 50%): cloud providers skipped, local 14B recommended
- RESTING (7d quota ≥ 80%): cloud providers skipped, local 8B recommended
Adds a log message showing the tier name and suggested local model on every downshift, so transitions are observable
Adds TestMetabolicProtocol (5 tests) covering BURST/ACTIVE/RESTING tier behaviour plus graceful degradation when the quota monitor is unavailable

src/infrastructure/claude_quota.py and scripts/claude_quota_check.sh were already committed in prior work.

Fixes #1106 ## Summary - Updates `CascadeRouter._quota_allows_cloud()` to use `QuotaMonitor.select_model("high")` instead of `should_use_cloud()`, aligning it with the metabolic protocol thresholds: - **BURST** (5h quota < 50%): cloud providers allowed - **ACTIVE** (5h quota ≥ 50%): cloud providers skipped, local 14B recommended - **RESTING** (7d quota ≥ 80%): cloud providers skipped, local 8B recommended - Adds a log message showing the tier name and suggested local model on every downshift, so transitions are observable - Adds `TestMetabolicProtocol` (5 tests) covering BURST/ACTIVE/RESTING tier behaviour plus graceful degradation when the quota monitor is unavailable `src/infrastructure/claude_quota.py` and `scripts/claude_quota_check.sh` were already committed in prior work.

claude added 1 commit 2026-03-23 18:12:55 +00:00

feat: wire QuotaMonitor.select_model() into cascade router metabolic protocol

Tests / lint (pull_request) Successful in 14s

Details

Tests / test (pull_request) Failing after 29m26s

Details

8d6078aea9

Update _quota_allows_cloud() to use select_model("high") instead of
should_use_cloud(), so the cascade router uses the same metabolic tier
logic as the standalone quota API:
  - BURST tier (5h < 50%): cloud providers allowed
  - ACTIVE tier (5h >= 50%): cloud providers skipped, local 14B used
  - RESTING tier (7d >= 80%): cloud providers skipped, local 8B used

Adds log message showing tier name and suggested local model when
downshifting, so Alexander can verify the protocol is working.

Adds TestMetabolicProtocol test class (5 tests) covering:
  - BURST tier allows cloud
  - ACTIVE tier skips cloud
  - RESTING tier skips cloud
  - Local (ollama) providers always attempted regardless of quota
  - No quota monitor → cloud allowed (graceful degradation)

Files src/infrastructure/claude_quota.py and scripts/claude_quota_check.sh
were already in place from prior work.

Fixes #1106

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

claude referenced this pull request

2026-03-23 18:13:07 +00:00

[AutoLoRA P2] Integrate claude_quota.py Metabolic Protocol #1106

claude merged commit 510d890eb2 into main

2026-03-23 18:13:17 +00:00

claude deleted branch claude/issue-1106

2026-03-23 18:13:18 +00:00

claude referenced this issue from a commit

2026-03-23 18:13:19 +00:00

[claude] Wire QuotaMonitor.select_model() into cascade router (#1106) (#1113)

Sign in to join this conversation.