Just saying thank you

[hugalafutro]

Thanks, it lets me stretch the CC sessions a bit longer and I have not noticed any degradation in output or conversation quality.

Headroom Performance Report
============================================================
Window: last 168h (actual data: 2026-06-19 10:47:11 → 2026-06-26 10:45:47)
Records outside window:  10,178 (filtered out — increase --hours to include them)

Requests:     9105
Tokens:       1,598,087,269 -> 1,455,081,165 (8.9% reduction)
Total saved:  143,006,104 tokens

Per-Model Breakdown
----------------------------------------
  claude-opus-4-8: 8888 reqs, 143,002,200 tokens saved (9%), list price $5.00/MTok  ~$715.01 at list price
  claude-sonnet-4-6: 57 reqs, 3,904 tokens saved (0%), list price $3.00/MTok  ~$0.01 at list price
  passthrough:count_tokens: 160 reqs, 0 tokens saved (0%), list price unknown
  * Actual bill savings depend on provider caching behavior

Cache Performance
----------------------------------------
  Cache read:    2,105,789,319 tokens
  Cache write:   87,888,587 tokens
  Hit rate:      96.0%
  Unstable:      285/8835 requests had cache_write > 2x cache_read
  First 5 avg:   read=276,951 write=543
  Last 5 avg:    read=201,211 write=3,682

Optimization Overhead
----------------------------------------
  Average:  807ms
  Max:      30478ms
  >500ms:   3880 requests

Conversation Size
----------------------------------------
  Min msgs:  1
  Max msgs:  1596
  Avg msgs:  291

Transform Effectiveness
----------------------------------------
  content_router: 0.1% avg reduction, 8945 uses, 2,756,014 saved

Content Router Routing
----------------------------------------
  Compressed:  2247 (0%)
  Excluded:    82311 (18%) — Read/Glob outputs
  Skipped:     310823 (66%) — <50 words
  Unchanged:   72132 (15%) — ratio too high
  ! Excluded tools dominate — consider compressing stale Read outputs

TOIN Learning
----------------------------------------
  Patterns:     207
  Compressions: 832
  Retrievals:   199 (23.9%)

TOIN Highlights (live store)
----------------------------------------
  22/206 patterns have ≥10 samples (eligible for `python -m headroom.cli.toin_publish`)

  Strategy distribution:
     131 pattern(s)  default
      75 pattern(s)  kompress

Recommendations
----------------------------------------
  1. 3880 requests took >500ms for optimization — consider reducing transform pipeline
  2. Read/Glob outputs are majority of messages but excluded — compress stale reads (>10 turns old) for significant savings

Log files: 5 | Lines parsed: 154,341
Log dir: /home/sammael/.headroom/logs