mirror of
https://github.com/headroomlabs-ai/headroom.git
synced 2026-08-27 14:17:10 -04:00
a99dc61424
## Description
Adds the first levers that reduce the tokens the model **writes back**
(output), complementing Headroom's existing input compression. Output
costs 5× input on Opus-class models and is full of waste (ceremony,
restated code, deep "thinking" on routine steps). Two phases in one
self-contained PR off `main`: the request-side output shaper, then
per-user verbosity learning plus an honest counterfactual savings
estimator and dashboard surfacing.
## Type of Change
- [ ] Bug fix (non-breaking change that fixes an issue)
- [x] New feature (non-breaking change that adds functionality)
- [ ] Breaking change (fix or feature that would cause existing
functionality to change)
- [x] Documentation update
- [ ] Performance improvement
- [ ] Code refactoring (no functional changes)
## Changes Made
- **Output shaper** (`output_shaper.py`, opt-in
`HEADROOM_OUTPUT_SHAPER=1`): cache-safe verbosity steering appended to
the system-prompt tail (5 levels); effort routing that lowers
`output_config.effort` on mechanical tool-result continuations; legacy
`thinking.budget_tokens` clamp. Never injects effort where absent, never
toggles `thinking.type`.
- **`headroom learn --verbosity`**: mines Claude Code transcripts for
behavioral signals (interrupts, length-adaptive fast-skips, echo ratio),
recommends a verbosity level (heuristic + optional `--llm-judge`), and
seeds the savings baseline.
- **Counterfactual estimator** (`output_savings.py`): per-stratum
synthetic-control (estimated) + A/B holdout (measured) with a propagated
95% CI; conversation-stable arm assignment for A/B validity and
prefix-cache safety.
- **AIMD verbosity controller** (`verbosity_controller.py`):
additive-increase / fast-back-off state machine; live signal emission
gated off by default.
- **Wiring + surfaces**: shaper resolves the learned level; recording
rides the existing `transforms_applied` channel through the outcome
funnel (no `RequestOutcome` changes); `headroom output-savings` CLI;
dashboard "Output Tokens Saved" card.
- **Docs**: simple-words user guide + design doc with the counterfactual
methodology.
## Testing
- [x] Unit tests pass (`pytest`)
- [x] Linting passes (`ruff check .`)
- [x] Type checking passes (`mypy headroom`)
- [x] New tests added for new functionality
- [x] Manual testing performed
### Test Output
```text
$ pytest tests/test_output_savings.py tests/test_output_savings_cli.py \
tests/test_verbosity_learn.py tests/test_verbosity_controller.py \
tests/test_output_shaper.py -q
94 passed in 0.54s
$ pytest tests/test_request_outcome.py tests/test_handler_outcome_tag_invariant.py \
tests/test_proxy_dashboard_stats_cache.py -q
44 passed
$ ruff format --check .
831 files already formatted
$ mypy headroom --ignore-missing-imports
Success: no issues found in 361 source files
```
## Real Behavior Proof
- Environment: macOS, Python 3.12 (`.venv`), `anthropic` 0.76, live API
model `claude-opus-4-8`.
- Exact command / steps: `HEADROOM_OUTPUT_SHAPER=1`; `headroom learn
--verbosity --apply` (seeds level + baseline); `python
scripts/eval_output_shaper.py A` (live before/after); simulate holdout
traffic then `headroom output-savings`.
- Observed result: code-review ask — baseline 1,750 output tokens → L2
1,354 (−22.7%) → L3 599 (−65.8%), same bugs found. `learn --verbosity`
on 24 real sessions → 11% interrupt / 26% fast-skip → L3 (high
confidence). Measured A/B path → 31.7% reduction (95% CI 27.7%–35.7%).
94 new tests + 44 existing outcome/dashboard tests green; ruff + mypy
clean.
- Not tested: live streaming-path recording exercised only via unit
tests (the `transforms_applied` funnel is shared across paths); runtime
AIMD signal emission is gated off by default and not exercised live.
## Review Readiness
- [x] I have performed a self-review
- [x] This PR is ready for human review
## Checklist
- [x] My code follows the project's style guidelines
- [x] I have performed a self-review of my code
- [x] I have commented my code, particularly in hard-to-understand areas
- [x] I have made corresponding changes to the documentation
- [x] My changes generate no new warnings
- [x] I have added tests that prove my fix is effective or that my
feature works
- [x] New and existing unit tests pass locally with my changes
- [ ] I have updated the CHANGELOG.md if applicable
## Additional Notes
Output savings are counterfactual (we never observe what the model
*would* have written), so the estimator separates **estimated** (vs a
learned baseline) from **measured** (A/B holdout via
`HEADROOM_OUTPUT_HOLDOUT`) and always reports a confidence band — never
a single made-up number. CHANGELOG left unchecked (release-please
manages it). Runtime AIMD self-tuning is intentionally a TODO
(controller built/tested; live signal emission gated behind
`HEADROOM_VERBOSITY_AUTOTUNE`).
|
||
|---|---|---|
| .. | ||
| fixtures | ||
| tests | ||
| audit_wheel_glibc_symbols.py | ||
| build_rust_extension.sh | ||
| changelog-gen.py | ||
| eval_output_shaper.py | ||
| export_kompress_v2_onnx.py | ||
| install-git-hooks.sh | ||
| install.ps1 | ||
| install.sh | ||
| pr-governance.py | ||
| README.md | ||
| record_fixtures.py | ||
| refresh_model_limits.sh | ||
| replay_codex_ws_load.py | ||
| repro_codex_replay.py | ||
| smoke_issue_327.py | ||
| sync-plugin-versions.py | ||
| validate-workflows.sh | ||
| verify-versions.py | ||
| version-sync.py | ||
scripts/
Utility scripts bundled with the Headroom repo. Most are one-off operator tools; a few are runnable as part of development workflows.
Reproducing the reconnect storm
repro_codex_replay.py reproduces the multi-agent Codex reconnect/retry storm
against a local Headroom proxy (default http://127.0.0.1:8787), as described
in wiki/plans/2026-04-17-codex-proxy-runtime-analysis.md under "Latest
Correction". Use it to:
- Regression-check that
/livezstays responsive under a cold-start storm. - Empirically tune the Unit 4 pre-upstream semaphore default
(
HEADROOM_ANTHROPIC_PRE_UPSTREAM_CONCURRENCY). - Exercise the Codex WS lifecycle + Anthropic HTTP path simultaneously without needing to replay captured production traffic.
Run
# Default: 8 WS + 4 HTTP clients, 30s storm, p99 /livez must stay <= 500ms.
python scripts/repro_codex_replay.py
# Tighter budget, shorter run:
python scripts/repro_codex_replay.py \
--url http://127.0.0.1:8787 \
--ws-clients 16 \
--anthropic-clients 8 \
--duration 60 \
--livez-threshold-ms 100
# Dump the full summary as JSON for downstream tooling:
python scripts/repro_codex_replay.py --json
Exit code:
0— warmup succeeded (or was skipped), storm ran for the requested duration, and/livezp99 stayed under--livez-threshold-ms.1— soft assertion failed, proxy unreachable, or unhandled exception. Proxy-unreachable is detected and reported within ~5 seconds.
Fixtures
The script loads two hand-crafted, fully synthetic JSON fixtures:
scripts/fixtures/anthropic_replay_body.json— shape of a large agent reconnect replay/v1/messages?beta=truePOST body.scripts/fixtures/codex_response_create_frame.json— first Codex WS frame with the{"type": "response.create", "response": {...}}envelope.
Override via --ws-frame-fixture / --anthropic-body-fixture if you have
captured traffic to replay instead.
Interpretation
/livez p99under threshold means the event loop is not starved during the storm. If it rises with the semaphore unbounded (HEADROOM_ANTHROPIC_PRE_UPSTREAM_CONCURRENCY=10000) and drops back under the default, Unit 4's backpressure is working.Codex WS: openedshould equal--ws-clients.response.completedtypically stays low when upstream auth isn't configured locally — the goal is handshake + relay wiring, not real upstream traffic.Anthropic HTTP: ok_2xx + non_2xx + timed_out + errorsshould roughly equalattempted. Sustained non-zerotimed_outduring the storm is the failure signal the plan targets.
A smoke test at tests/test_scripts/test_repro_codex_replay_smoke.py
exercises the script against a mock FastAPI server on every PR.
Install scripts
install.sh— POSIX installer.install.ps1— Windows PowerShell installer.
These are generated by the release pipeline; edit with care.