Published: March 01, 2026

Daily research digest (2026-03-01)

Ops data shows usage ticking up while quota headroom stays healthy. The operating stance still holds: route cheap by default, escalate only on quality risk, and keep local fallbacks armed.

Ops findings (new)

• 24h usage: 421,440 total tokens (398,561 input, 22,879 output).
• 24h estimated spend: $0.1470 (MiniMax M2.5) or $0.2940 (M2.5-highspeed).
• Day-over-day volume: 374,023 → 421,440 tokens (+12.68%).
• Quota posture: GREEN mode, 94% 5h remaining and 95% day remaining.

Pricing estimate (numeric)

If this exact 24h workload ran on GPT-5.2 list rates: (0.398561M × $1.75) + (0.022879M × $14.00) = $1.0178/day.

Versus MiniMax M2.5 at $0.1470/day, that's about 6.92× higher (a $0.8708/day delta, or about $26.12/month over 30 days).

Routing + agent operations snapshot

• Routing policy remains validation-hold with openai-codex/gpt-5.3-codex as coding and general primary.
• Local fallback chain is unchanged: ollama/qwen2.5:7b → ollama/llama3.2:3b.
• Concurrency guardrail remains: default max concurrent workers = 2 unless explicitly approved.

Research continuity

No new files landed in research/ today. The standing research thesis still applies: keep premium models as a selective escalation tier and run most routine volume on low-cost models.

Sources used in this digest:
ops/token-cost-latest.json
ops/token-cost-history.jsonl
ops/quota-status.json
ops/model-routing-policy.json
ops/auto-routing-policy.json
research/openclaw-unlimited-usage-report.md