beqforge 0.2.1__tar.gz → 0.2.3__tar.gz

{beqforge-0.2.1 → beqforge-0.2.3}/.github/workflows/build-device-check.yml

@@ -61,7 +61,7 @@ jobs:
             executable: beqforge-device-check
     steps:
       - uses: actions/checkout@v7
-      - uses: astral-sh/setup-uv@v10.2.0
+      - uses: astral-sh/setup-uv@v10.3.0
         with:
           enable-cache: true
       - name: Install Linux audio and USB libraries

{beqforge-0.2.1 → beqforge-0.2.3}/.github/workflows/build-executable.yml

@@ -25,7 +25,7 @@ jobs:
     steps:
       - name: Find successful tests for the tagged commit
         id: check
-        uses: actions/github-script@v8
+        uses: actions/github-script@v9
         with:
           script: |
             core.setOutput('reusable', 'false');
@@ -72,6 +72,8 @@ jobs:
     needs: previous_tests
     if: needs.previous_tests.outputs.reusable != 'true'
     uses: ./.github/workflows/tests.yml
+    # CODECOV_TOKEN, for the coverage upload
+    secrets: inherit
     permissions:
       contents: read
 
@@ -102,7 +104,7 @@ jobs:
       contents: read
     steps:
       - uses: actions/checkout@v7
-      - uses: astral-sh/setup-uv@v10.2.0
+      - uses: astral-sh/setup-uv@v10.3.0
         with:
           enable-cache: true
       - name: Build wheel and source distribution
@@ -191,7 +193,7 @@ jobs:
       - uses: actions/checkout@v7
 
       - name: Install uv
-        uses: astral-sh/setup-uv@v10.2.0
+        uses: astral-sh/setup-uv@v10.3.0
         with:
           enable-cache: true
 

{beqforge-0.2.1 → beqforge-0.2.3}/.github/workflows/tests.yml

@@ -67,7 +67,7 @@ jobs:
         run: choco install ffmpeg -y
 
       - name: Install uv
-        uses: astral-sh/setup-uv@v10.2.0
+        uses: astral-sh/setup-uv@v10.3.0
         with:
           enable-cache: true
 
@@ -76,7 +76,17 @@ jobs:
 
       # the device-check tests run in their own job below, in parallel, with the extra installed
       - name: Run tests
-        run: uv run --extra designer pytest -n auto --durations=25 --ignore-glob='tests/test_device_check_*.py'
+        run: uv run --extra designer pytest -n auto --durations=25 --ignore-glob='tests/test_device_check_*.py' --cov --cov-report=xml
+
+      - name: Upload coverage
+        if: ${{ !cancelled() }}
+        uses: codecov/codecov-action@v7.1.1
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          files: coverage.xml
+          disable_search: true
+          flags: designer,${{ runner.os }}
+          name: designer-${{ runner.os }}
 
   device-check:
     name: Device check (${{ matrix.os }})
@@ -90,7 +100,7 @@ jobs:
       - uses: actions/checkout@v7
 
       - name: Install uv
-        uses: astral-sh/setup-uv@v10.2.0
+        uses: astral-sh/setup-uv@v10.3.0
         with:
           enable-cache: true
 
@@ -100,7 +110,17 @@ jobs:
 
       - name: Run device-check tests
         shell: bash
-        run: uv run --extra device-check pytest -n auto --durations=25 tests/test_device_check_*.py
+        run: uv run --extra device-check pytest -n auto --durations=25 tests/test_device_check_*.py --cov --cov-report=xml
+
+      - name: Upload coverage
+        if: ${{ !cancelled() }}
+        uses: codecov/codecov-action@v7.1.1
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          files: coverage.xml
+          disable_search: true
+          flags: device-check,${{ runner.os }}
+          name: device-check-${{ runner.os }}
 
   package-profiles:
     name: Install profile (${{ matrix.profile }})
@@ -111,7 +131,7 @@ jobs:
         profile: [base, designer, optimiser, device-check, all]
     steps:
       - uses: actions/checkout@v7
-      - uses: astral-sh/setup-uv@v10.2.0
+      - uses: astral-sh/setup-uv@v10.3.0
         with:
           enable-cache: true
       - name: Build the single distribution

{beqforge-0.2.1 → beqforge-0.2.3}/AGENTS.md

@@ -56,7 +56,9 @@ importing them; use an isolated worktree for simultaneous experiments.
 The root build configuration includes every workflow and the optimiser's bundled seed in one
 wheel and source distribution. `.github/workflows/tests.yml` checks isolated dependency
 profiles and runs the suite in parallel (`pytest -n auto`, so tests must not share mutable state
-or fixed ports). Feature branch pushes test on Linux only and a newer push cancels the branch's
+or fixed ports), with coverage (`--cov`, configured in `pyproject.toml`, `multiprocessing`
+children included) uploaded to Codecov by each test job using the `CODECOV_TOKEN` secret;
+release runs inherit it. Feature branch pushes test on Linux only and a newer push cancels the branch's
 previous run; `main`, tags and manual runs test Linux, macOS and Windows and are never cancelled. `.github/workflows/build-executable.yml` builds release executables and distributions;
 tag pushes publish to PyPI after validation and successful executable builds. Tags must match
 the shared version exactly (`vX.Y.Z`). Manual runs build artifacts without publishing.

{beqforge-0.2.1 → beqforge-0.2.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: beqforge
-Version: 0.2.1
+Version: 0.2.3
 Summary: Audio-derived BEQ design, published filter optimisation and DSP validation
 License-File: LICENCE.md
 Requires-Python: <3.15,>=3.13
@@ -270,8 +270,9 @@ The image's defaults, all overridable with the environment variables above:
 
 It runs as uid/gid 1000 (`--build-arg PUID=… PGID=…` to change), which must be able to write
 the shared root and the cache: the server checks both with a write at startup. Its
-`HEALTHCHECK` asks `/health`. The server answers one request at a time (its fitter forks), so a
-caller sending several at once queues them; size the caller's timeout for that.
+`HEALTHCHECK` asks `/health`, which answers while a design runs: designs run in a separate
+worker process, one at a time, so a caller sending several at once queues them; size the
+caller's timeout for that. `docker stop` stops a design in progress.
 
 Build and check it locally from the repository root:
 

{beqforge-0.2.1 → beqforge-0.2.3}/README.md

@@ -244,8 +244,9 @@ The image's defaults, all overridable with the environment variables above:
 
 It runs as uid/gid 1000 (`--build-arg PUID=… PGID=…` to change), which must be able to write
 the shared root and the cache: the server checks both with a write at startup. Its
-`HEALTHCHECK` asks `/health`. The server answers one request at a time (its fitter forks), so a
-caller sending several at once queues them; size the caller's timeout for that.
+`HEALTHCHECK` asks `/health`, which answers while a design runs: designs run in a separate
+worker process, one at a time, so a caller sending several at once queues them; size the
+caller's timeout for that. `docker stop` stops a design in progress.
 
 Build and check it locally from the repository root:
 

{beqforge-0.2.1 → beqforge-0.2.3}/TODO.md

@@ -18,29 +18,31 @@ are retained so records and code comments remain traceable. Completed items are
 | Priority | ID | Status | Value, work and blocker or reason for parking |
 | --- | --- | --- | --- |
 | 1 | Playback contract | open | Reconcile the supplied bass-management model with the model actually used for targets, verification and clipping. Only `lpf_fs` is consumed today; check the ignored fields before claiming the listener's clipping cost. Caller propagation is an external subtask. |
-| 2 | E10 | open | Predeclare fresh content-goal validation before further tuning or a default change. Existing seeds and corpus have repeatedly informed development; writing the protocol needs no new audio. |
-| 3 | E1 | open | Add record-only scene-held-out verification: evidence independent of the material that generated the target. A gate needs a separate demonstrated benefit. |
-| 4 | Steep-filter scoring | open | Implement coloured/codec-like floors and recipe-specific truth bands under E10; inspect shape failures and published-ceiling excess. Existing titles can supply programme texture. |
-| 5 | Steep-filter default | blocked | Recovery improves substantially with `--content-edge`, but changing the default needs E10, the scoring above, listening judgements and more independent real steep titles. Safe House already works with the default; keep the option off meanwhile. |
-| 6 | F2 / device behaviour | in progress; first hardware pilot measured | Coefficient quantisation does not establish hardware accuracy. Limit changes need measured internal arithmetic/noise/limit cycles on a named device and a capture setup; the [measurement plan](plans/F2-device-precision-validation.md) specifies Python sweep/capture, optional REW cross-checks, software controls, single-entry/full-catalogue analysis and a standalone cross-platform executable. The `beqforge-device-check` preview includes frozen manifests, capture/recovery, CLI-only miniDSP controls (minidsp-rs 0.1.12, batched through `-f`) with float32 transport attribution, staged qualification/convergence, uncertainty-aware filter checks, entry/unique-cascade frequency distributions and matched-device deltas. First verified pilot on a miniDSP 2x4 HD (USB loopback, 96 kHz): the device plays its float32 coefficients as predicted (worst 0.0074 dB, 2-200 Hz), while float32 coefficients themselves put 18.8 dB of error into a 5 Hz Q 6 peak; see the plan, section 12. First catalogue sample (50 cascades): 49 play as predicted on every bin (worst 0.0082 dB); the 50th was clipped by the device and is to be remeasured (F2a, plan section 13). Next: the `grid` and `boundary` suites on hardware to map where float32 coefficients stop representing low, high-Q filters. No Q-limit change yet. |
-| 7 | Validation coverage (former TODO 10) | blocked | Broad real-programme generalisation needs the missing shapes and independently characterised paired releases listed below. Sparse titles exist; their absolute truth and wider sparse coverage remain missing. |
-| 8 | Identification | blocked | Generalising `N(f)` and its fit band needs independent, varied source material, especially paired releases. Sensitivity experiments can proceed; programme-only uncertainty cannot identify the cause of a droop. |
-| 9 | E4 | parked | No measured failure has been attributed to isolated ceiling holes. Reopen if unevenness or E9 excess is traced to such bins; smoothing may only reduce the raw licence. |
-| 10 | F4 follow-up | parked | Both fitter-objective selection spikes lost passes/true positives. Judge several fits only with a demonstrated usable alternative discarded before `_judge`; a smaller residual is insufficient evidence. |
-| 11 | F3 follow-up | parked | A fifth section moved no verdict when tried and costs more fitting. Reopen if an evidence-supported target has a demonstrated usable fit beyond four sections and no usable fit within four. |
-| 12 | T2 | parked | The steep-variant audit found no swept cap binding a weak winner. Derive caps only when a binding cap demonstrably prevents a better supported result. |
-| 13 | Residual (former TODO 5) | open | Correct the reported target error and its actual measurement band without changing fit selection. Primarily audit accuracy; any escalation/objective change is a separate, higher-impact decision. Non-parametric reporting is already allowed upstream. |
-| 14 | Authored features | parked | Manual `--exclude` works; no validated automatic detector exists. Reopen with real labelled features and negatives showing that detection preserves legitimate bass. |
-| 15 | Dynamic processing (former TODO 11) | parked | No suspected real case has scoped a useful tripwire. Reopen when compression/limiting demonstrably compromises this correction; non-LTI processing alone does not prove every BEQ is invalid. |
-| 16 | Tilted passband reference (goal/corpus follow-up) | parked | No real in-sub-band case needs a sloping reference yet; Dossier's abstention was decided. Reopen with one. Goal-aware corpus labels belong to the open E10 protocol work. Nearest so far: 13 Going on 30's mix tilts about −7 dB/oct above 50 Hz (bass is coherent across channels, ~200 Hz is not) and finds its plateau only through the one-sided fallback. That reference reads a 36 dB deficit below 25 Hz; each channel's own plateau reads 30-34 dB. A reference built from the channels would be 2-5 dB less generous there (see plans/done-design-and-pipeline.md, 11 October 2026). |
-| 17 | Mixed-channel policy | parked | `channel_scope` is currently an unpopulated optional response field, not a computed `mixed` finding awaiting a gate. Reopen with a case where existing per-channel restoration and common pricing produce an unsuitable mix correction. |
-| 18 | Fraction dial (former TODO 7) | parked | No user/title has needed partial licensed restoration as a preference; tilt and tolerance already ship. Reopen with a concrete requested outcome, then price and validate it like every target. |
-| 19 | Protective corner (former TODO 9) | parked | The corner is already derived from boost cap and implied order, floored by `lowest_frequency_hz`. No case requires an independent override or different fallback alignment; reopen with one. |
-| 20 | Device configuration on the wire | parked | The current contract deliberately omits device preference; server flags already configure realisation. Reopen if one server must serve different device formats/rates per request; then agree an upstream extension. Hardware validity remains F2. |
-| 21 | Catalogue comparison (former TODO 8) | parked | A disagreement detector has no demonstrated accuracy benefit. Reopen with an independent discrepancy to investigate; authored filters remain comparisons, never calibration targets. |
-| 22 | Performance | parked | Further reuse changes cost, not filter accuracy. Reopen after measuring repeated numerical identities or material remaining analysis cost; missed-fit search belongs to F3/F4 instead. |
-| 23 | Chart peak bias | parked | Deliberate catalogue-compatible per-bin maximum; it affects presentation only. Reopen if a reader needs an alternative percentile view; keep it distinct from the 95th-percentile evidence envelope. |
-| 24 | Documentation citations | open | Remove or repoint orphan references and stale descriptions of shipped behaviour. Low filter impact, but prevents future changes being based on obsolete assumptions. |
+| 2 | Texture yardstick band | open (E10 only) | Implemented: the mix's ripple is measured about a cubic over the unchanged band, so the LFE knee is trend. 9 (2009), Beauty and the Beast, Crazy, Stupid, Love and Despicable Me 3 now accept; Carrie stays declined. Probe and protocol unchanged; corpus 0/45 gate pass. Neither width-normalised form separated `broadband/5`. The degree was chosen on corpus and library data, so it needs the fresh E10 validation before it counts as settled; see [the outcome](plans/texture-yardstick-sub-band.md#outcome-17-october-2026). |
+| 3 | E10 | open | Predeclare fresh content-goal validation before further tuning or a default change. Existing seeds and corpus have repeatedly informed development; writing the protocol needs no new audio. |
+| 4 | E1 | open | Add record-only scene-held-out verification: evidence independent of the material that generated the target. A gate needs a separate demonstrated benefit. |
+| 5 | Steep-filter scoring | open | Implement coloured/codec-like floors and recipe-specific truth bands under E10; inspect shape failures and published-ceiling excess. Existing titles can supply programme texture. |
+| 6 | Steep-filter default | blocked | Recovery improves substantially with `--content-edge`, but changing the default needs E10, the scoring above, listening judgements and more independent real steep titles. Safe House already works with the default; keep the option off meanwhile. |
+| 7 | F2 / device behaviour | in progress; first hardware pilot measured | Coefficient quantisation does not establish hardware accuracy. Limit changes need measured internal arithmetic/noise/limit cycles on a named device and a capture setup; the [measurement plan](plans/F2-device-precision-validation.md) specifies Python sweep/capture, optional REW cross-checks, software controls, single-entry/full-catalogue analysis and a standalone cross-platform executable. The `beqforge-device-check` preview includes frozen manifests, capture/recovery, CLI-only miniDSP controls (minidsp-rs 0.1.12, batched through `-f`) with float32 transport attribution, staged qualification/convergence, uncertainty-aware filter checks, entry/unique-cascade frequency distributions and matched-device deltas. First verified pilot on a miniDSP 2x4 HD (USB loopback, 96 kHz): the device plays its float32 coefficients as predicted (worst 0.0074 dB, 2-200 Hz), while float32 coefficients themselves put 18.8 dB of error into a 5 Hz Q 6 peak; see the plan, section 12. First catalogue sample (50 cascades): 49 play as predicted on every bin (worst 0.0082 dB); the 50th was clipped by the device and is to be remeasured (F2a, plan section 13). Next: the `grid` and `boundary` suites on hardware to map where float32 coefficients stop representing low, high-Q filters. No Q-limit change yet. |
+| 8 | Validation coverage (former TODO 10) | blocked | Broad real-programme generalisation needs the missing shapes and independently characterised paired releases listed below. Sparse titles exist; their absolute truth and wider sparse coverage remain missing. |
+| 9 | Identification | blocked | Generalising `N(f)` and its fit band needs independent, varied source material, especially paired releases. Sensitivity experiments can proceed; programme-only uncertainty cannot identify the cause of a droop. |
+| 10 | E4 | parked | No measured failure has been attributed to isolated ceiling holes. Reopen if unevenness or E9 excess is traced to such bins; smoothing may only reduce the raw licence. |
+| 11 | F4 follow-up | parked | Both fitter-objective selection spikes lost passes/true positives. Judge several fits only with a demonstrated usable alternative discarded before `_judge`; a smaller residual is insufficient evidence. |
+| 12 | F3 follow-up | parked | A fifth section moved no verdict when tried and costs more fitting. Reopen if an evidence-supported target has a demonstrated usable fit beyond four sections and no usable fit within four. |
+| 13 | T2 | parked | The steep-variant audit found no swept cap binding a weak winner. Derive caps only when a binding cap demonstrably prevents a better supported result. |
+| 14 | Residual (former TODO 5) | open | Correct the reported target error and its actual measurement band without changing fit selection. Primarily audit accuracy; any escalation/objective change is a separate, higher-impact decision. Non-parametric reporting is already allowed upstream. |
+| 15 | Authored features | parked | Manual `--exclude` works; no validated automatic detector exists. Reopen with real labelled features and negatives showing that detection preserves legitimate bass. |
+| 16 | Dynamic processing (former TODO 11) | parked | No suspected real case has scoped a useful tripwire. Reopen when compression/limiting demonstrably compromises this correction; non-LTI processing alone does not prove every BEQ is invalid. |
+| 17 | Tilted passband reference (goal/corpus follow-up) | parked | No real in-sub-band case needs a sloping reference yet; Dossier's abstention was decided. Reopen with one. Goal-aware corpus labels belong to the open E10 protocol work. Nearest so far: 13 Going on 30's mix tilts about −7 dB/oct above 50 Hz (bass is coherent across channels, ~200 Hz is not) and finds its plateau only through the one-sided fallback. That reference reads a 36 dB deficit below 25 Hz; each channel's own plateau reads 30-34 dB. A reference built from the channels would be 2-5 dB less generous there (see plans/done-design-and-pipeline.md, 11 October 2026). |
+| 18 | Mixed-channel policy | parked | `channel_scope` is currently an unpopulated optional response field, not a computed `mixed` finding awaiting a gate. Reopen with a case where existing per-channel restoration and common pricing produce an unsuitable mix correction. |
+| 19 | Fraction dial (former TODO 7) | parked | No user/title has needed partial licensed restoration as a preference; tilt and tolerance already ship. Reopen with a concrete requested outcome, then price and validate it like every target. |
+| 20 | Protective corner (former TODO 9) | parked | The corner is already derived from boost cap and implied order, floored by `lowest_frequency_hz`. No case requires an independent override or different fallback alignment; reopen with one. |
+| 21 | Device configuration on the wire | parked | The current contract deliberately omits device preference; server flags already configure realisation. Reopen if one server must serve different device formats/rates per request; then agree an upstream extension. Hardware validity remains F2. |
+| 22 | Catalogue comparison (former TODO 8) | parked | A disagreement detector has no demonstrated accuracy benefit. Reopen with an independent discrepancy to investigate; authored filters remain comparisons, never calibration targets. |
+| 23 | Performance | parked | Further reuse changes cost, not filter accuracy. Reopen after measuring repeated numerical identities or material remaining analysis cost; missed-fit search belongs to F3/F4 instead. |
+| 24 | Chart peak bias | parked | Deliberate catalogue-compatible per-bin maximum; it affects presentation only. Reopen if a reader needs an alternative percentile view; keep it distinct from the 95th-percentile evidence envelope. |
+| 25 | Documentation citations | open | Remove or repoint orphan references and stale descriptions of shipped behaviour. Low filter impact, but prevents future changes being based on obsolete assumptions. |
+| 26 | Texture yardstick retune (library review) | parked | The cubic (row 2) keeps the 200 Hz band only because the sub-band cap admitted one synthetic negative, `broadband/5`: chance unevenness in noise bursts, at a 2.6 dB boost. With 45 gated cases a single case moves the gate's upper bound from 0.079 to 0.118. Carrie is still declined (0.85x about a cubic, 1.33x with the cap). Reopen when the beqdesigner library review (~1,290 titles) is complete. Then retune against its declines: the band edge, the trend degree, and whether the gate weights a false acceptance by its boost. The library holds no known negative, so it can expose wrong declines but not measure false acceptances. Any retune still needs the corpus or a fresh E10 set for that, ideally with negatives built from real programme texture (row 5). See [the outcome](plans/texture-yardstick-sub-band.md#outcome-17-october-2026). |
 
 ## Review evidence
 

{beqforge-0.2.1 → beqforge-0.2.3}/beq_common/__init__.py

@@ -1,6 +1,6 @@
 """Shared filter primitives; no workflow package dependencies."""
 
-__version__ = "0.2.1"
+__version__ = "0.2.3"
 
 from .types import BiquadSpec as BiquadSpec
 from .types import BiquadType as BiquadType

{beqforge-0.2.1 → beqforge-0.2.3}/beqforge/AGENTS.md

@@ -43,7 +43,7 @@ still holds: no other API, no other service, and the server is a thin transport
 | `beqforge/explain.py` | The plain-language account a designer response leads with — `found` (reference plateau, which channels carry it and which are silent, down to where there is real bass content, judged band, goal), `correction` (the cascade, then frequency by frequency what was wanted, what the content supports and before → after), `clipping` (the sub-feed peak and how far to turn the sub down, on the request's or the assumed bass management) and `alternatives` (why each other candidate lost). Reads the `Report`; measures and decides nothing. |
 | `beqforge/designer.py` | The `design(request) -> response` adapter between beqdesigner's `designer-interface.md` v1.1 and `beqforge.pipeline.run` — request/response dataclasses (1.1: every candidate the judge failed goes back in `rejected` with its failures, review only, beside the answer or the decline), wire-format (de)serialisation and the field-by-field mapping onto `Report`/`Candidate`/`Verdict`. Also `validate_response`, an independent re-implementation of beqdesigner's own §3-§5 rules (that repo isn't importable — PyQt6), called by the server before a response ever goes on the wire. Pure and independently testable; no socket. |
 | `tools/design_beq.py` | **The entry point.** One command, a filter and its reasoning. `--charts DIR` for the pictures; writes a `.run.json.gz` record beside the material unless `--no-record`. |
-| `tools/designer_server.py` | `beqforge serve-designer` — the HTTP transport (`designer-interface.md` §7.1) around `beqforge/designer.py`. Server-wide flags for device realisation/strategies/exclusions; everything per-title comes from the request. Single-threaded (`http.server.HTTPServer`, not `ThreadingHTTPServer`) on purpose: the fitter forks worker processes (`beqforge.filters.PARALLEL_FITS`) when a fit escalates past one section count, and forking a multi-threaded process risks a deadlock. Costs nothing real — the contract is one synchronous POST per `design()` call. Every response names its build (`beqforge_revision`); `--record-dir DIR` also writes each request's run record, named by a digest of the request audio, so a review-queue entry can be replayed. `--cache-dir DIR` keeps the stage cache there (`cache.DirStore`, one file per entry, atomic): a repeat request for a title skips the analysis — 110 s cold, 50 s warm on a 2-hour 8-channel title. Each request logs a timing line (read, parse, decode, design, respond); `tools/experiments/request_timing.py` reports them. `--shared-root DIR` accepts audio by reference (contract 1.2, beqdesigner's `design/designer-by-reference.md` §3): an array names a WAV relative to DIR plus the SHA-256 of its decoded column, and `beqforge/reference.py` resolves, decodes and checks it; one that cannot be honoured is a 422 naming the array, never a decline. `GET /health` reports `contract_version` and `shared_root`. |
+| `tools/designer_server.py` | `beqforge serve-designer` — the HTTP transport (`designer-interface.md` §7.1) around `beqforge/designer.py`. Server-wide flags for device realisation/strategies/exclusions; everything per-title comes from the request. A `ThreadingHTTPServer`, so `/health` answers during a design, with every design run in one long-lived worker process started with "spawn" (`_DesignWorker`): the fitter forks worker processes (`beqforge.filters.PARALLEL_FITS`) when a fit escalates past one section count, and forking a multi-threaded process risks a deadlock, so the process that forks must stay single-threaded — the worker's log records come back over a `SimpleQueue`, not a `Queue`, for that reason. One worker keeps designs serial (the contract is one synchronous POST per `design()` call); a worker that dies answers 500 and is replaced. SIGTERM shuts down like Ctrl-C, stopping the worker's process group, fit pool included. Every response names its build (`beqforge_revision`); `--record-dir DIR` also writes each request's run record, named by a digest of the request audio, so a review-queue entry can be replayed. `--cache-dir DIR` keeps the stage cache there (`cache.DirStore`, one file per entry, atomic): a repeat request for a title skips the analysis — 110 s cold, 50 s warm on a 2-hour 8-channel title. Each request logs a timing line (read, parse, decode, design, respond, queued); `tools/experiments/request_timing.py` reports them. `--shared-root DIR` accepts audio by reference (contract 1.2, beqdesigner's `design/designer-by-reference.md` §3): an array names a WAV relative to DIR plus the SHA-256 of its decoded column, and `beqforge/reference.py` resolves, decodes and checks it; one that cannot be honoured is a 422 naming the array, never a decline. `GET /health` reports `contract_version` and `shared_root`. |
 | `tools/replay.py` | Redraw charts and export to beqdesigner from a record — no rerun, no extraction. Refuses on a stale record unless `--force`. |
 | `tools/extract.py` | ffmpeg → 1 kHz per-channel `.npz`. Requires an explicit supported channel layout; preserves layout provenance. Needs no beqdesigner. |
 | `tools/summarise.py` | Sanity-check an extraction before using it. |
@@ -333,7 +333,11 @@ least `min_judge_octaves` above the floor — no fixed top (historical review T6
   (`pipeline.passband_ripple_db`) is left alone: it abstains with `within_programme_ripple`
   before any strategy runs. That ripple is the title's own yardstick, so there is no constant.
   Crest to trough rather than a one-sided dip, because the reference sits on the crests. This
-  is what took the negative corpus from 4 false acceptances in 45 to 0.
+  is what took the negative corpus from 4 false acceptances in 45 to 0. The mix's ripple is
+  taken about a cubic in log frequency, so an LFE-led mix's knee above the crossover is trend,
+  not texture; each channel's is taken about a line. Don't shorten the band instead:
+  crest to trough shrinks with bin count, and the 1x boundary was calibrated on the full band
+  ([plans/texture-yardstick-sub-band.md](../plans/texture-yardstick-sub-band.md)).
 * **Steep filters: the default declines them; `judge_from_content_edge` is an opt-in that
   recovers them and lifts some noise.** Below a steep filter's edge the priced target already
   falls into the quiet floor with the programme — contrast bounds it — and the shape clauses

{beqforge-0.2.1 → beqforge-0.2.3}/beqforge/diagnose.py

@@ -618,15 +618,21 @@ def diagnose(
     constrain channel restoration and the band over which a filter is judged.
     """
     params = params or DiagnoseParams()
-    freqs, mix_db = mean_spectrum(material.mono_mix, material.fs)
-    band = (freqs >= params.band_hz[0]) & (freqs <= params.band_hz[1])
-    freqs, mix_db = freqs[band], mix_db[band]
+    spectrum_freqs, mix_spectrum = mean_spectrum(material.mono_mix, material.fs)
+    band = (spectrum_freqs >= params.band_hz[0]) & (spectrum_freqs <= params.band_hz[1])
+    freqs, mix_db = spectrum_freqs[band], mix_spectrum[band]
     spectra = {
         name: mean_spectrum(samples, material.fs)[1]
         for name, samples in material.channels.items()
     }
     shares, share_errors = coherent_shares(material)
-    _, mix_plateau = plateau_reference(mix_db, freqs, params)
+    # The mix plateau is found on the whole spectrum, exactly as `pipeline` finds it, so tracking
+    # is measured from the plateau the run judges against. The reference grid ends exactly at the
+    # band top: cut off at the last bin below it, a plateau reaching that edge loses its top point
+    # and can vanish. Diagnosis then lost the plateau `pipeline` had, measured no tracking at all
+    # and reported the band top as where tracking stopped (A Few Good Men: 150.5-200 Hz found by
+    # one, none by the other).
+    _, mix_plateau = plateau_reference(mix_spectrum, spectrum_freqs, params)
     from beqforge.extraction import ExtractionParams, extract
 
     extraction_params = extraction_params or ExtractionParams()

{beqforge-0.2.1 → beqforge-0.2.3}/beqforge/pipeline.py

@@ -772,6 +772,9 @@ MAINS_HZ = (50.0, 60.0)
 # inside the 50 Hz spacing so it never reaches the next harmonic
 HUM_NOTCH_HALF_WIDTH_HZ = 2.5
 HUM_SURROUND_HALF_WIDTH_HZ = 10.0
+# the mix's texture is measured about a cubic in log frequency, so an LFE-led mix's knee above
+# the crossover is trend, not ripple (plans/texture-yardstick-sub-band.md)
+MIX_TEXTURE_TREND_DEGREE = 3
 
 
 def mains_hum_notches(
@@ -815,7 +818,11 @@ def mains_hum_notches(
 
 
 def passband_ripple_db(
-    material: Material, params: "PipelineParams", noise_floor_hz: float = math.nan
+    material: Material,
+    params: "PipelineParams",
+    noise_floor_hz: float = math.nan,
+    *,
+    trend_degree: int = 1,
 ) -> tuple[float, tuple[float, float]] | None:
     """The programme's own texture where nothing is missing: crest-to-trough, dB, and where.
 
@@ -823,6 +830,17 @@ def passband_ripple_db(
     correction — the top of the judged band to the top of the analysis band — as the spread of
     the curve about its own log-frequency trend, so a tilted passband is not read as ripple.
 
+    The trend is a polynomial of `trend_degree` in log frequency. The mix's texture blocker
+    asks for a cubic (`MIX_TEXTURE_TREND_DEGREE`). In an LFE-led mix the level falls away
+    where the LFE channel's content ends (60-100 Hz), a knee a straight line cannot remove, so
+    it read as ripple and withheld a real deficit: 9 (2009) measured 8.8 dB of "ripple" against
+    a 6.2 dB shortfall, 2.2 dB about a cubic. The band is not shortened instead. Crest to
+    trough shrinks with the number of bins even on noise, so ending it at the crossover lowered
+    every yardstick and admitted a broadband negative on the corpus. A cubic keeps the band the
+    1x boundary was calibrated on and leaves the negatives where they were: at most 0.76x about
+    a cubic against 0.77x about a line, with every real and filtered title at 1.08x and up.
+    Each channel's own rule (`channels_missing_low_end`) keeps the line.
+
     Crest to trough, not a one-sided dip, because the reference sits near the 90th percentile
     of the spectrum, on the crests: a shortfall measured from it spans the whole swing. On the
     negative corpus (IMPROVEMENT_PLAN E2) every unfiltered title's low-end deficit came in
@@ -847,7 +865,7 @@ def passband_ripple_db(
         return None
     smooth = smooth_unexcluded(response, freqs, bands, 15, levels=True)[band]
     octaves = np.log2(freqs[band])
-    about = smooth - np.polyval(np.polyfit(octaves, smooth, 1), octaves)
+    about = smooth - np.polyval(np.polyfit(octaves, smooth, trend_degree), octaves)
     return float(np.max(about) - np.min(about)), (low, high)
 
 
@@ -1550,7 +1568,12 @@ def analyse(
                 f"programme (tracking stops at {diagnosis.noise_floor_hz:.1f} Hz), so there "
                 "is no bass passband to restore towards; restoration withheld"
             )
-        texture = passband_ripple_db(material, params, diagnosis.noise_floor_hz)
+        texture = passband_ripple_db(
+            material,
+            params,
+            diagnosis.noise_floor_hz,
+            trend_degree=MIX_TEXTURE_TREND_DEGREE,
+        )
         low_end = low_end_deficit_db(material, params, diagnosis.noise_floor_hz)
         if texture is not None and low_end is not None:
             ripple, (edge_low, edge_high) = texture

{beqforge-0.2.1 → beqforge-0.2.3}/plans/done-design-and-pipeline.md

@@ -748,6 +748,76 @@ beside the answer or the decline (probe unchanged; no decision touched).
     caller's; `beqforge_revision` is already in every response for it.
 
 
+## The designer server answers `/health` during a design (17 October 2026)
+
+A deployed image went unhealthy while it designed correctly. `serve-designer` was a plain
+`HTTPServer` running `design()` on its only thread, so a `GET /health` waited behind a 50-110 s
+design; the image's `HEALTHCHECK` gives up after 3 s and three failures mark the container
+unhealthy. It had been single-threaded on purpose: the fitter forks its pool, and forking a
+multi-threaded process risks a deadlock.
+
+The server is now a `ThreadingHTTPServer`, and every design runs in one long-lived worker
+process started with "spawn" (`_DesignWorker` in `tools/designer_server.py`). The worker's
+only thread runs the design, so the fitter forks from a single-threaded process as before. Its
+log records come back over a `SimpleQueue` (a `Queue` would start a feeder thread in it) and
+go through the parent's loggers, and the parent waits for a request's records before
+answering. One worker keeps designs serial. A worker that dies answers 500 and is replaced.
+SIGTERM now shuts down as Ctrl-C does (as a container's PID 1 the server used to ignore it),
+and closing stops the worker's process group, fit pool included, rather than wait for a
+design in progress. The timing line gains `queued`. Nothing under `beqforge/` changed, so the
+decision regression probe does not apply.
+
+* **Checks.** New tests: `/health` answers within 3 s while a real design runs, and a killed
+  worker gives a 500 and then a working replacement. The existing server, image and startup
+  tests now build `DesignServer` with `DesignSettings` (51 passed). Against the installed
+  `beqforge serve-designer`: `/health` took at most 16 ms during a design, and SIGTERM in the
+  middle of one exited in 0.3 s with no worker or fit-pool process left. Before the
+  process-group fix, the fit pool was orphaned and kept computing. The image built and
+  `packaging/designer/smoke.py` passed, inline and by reference. Full designer suite with
+  coverage: 741 passed, 1 skipped. The fork-from-a-threaded-process warnings in it come from
+  xdist's own worker threads and appear without these changes.
+* **Coverage.** CI now measures coverage (`pytest-cov`; `[tool.coverage.run]` in
+  `pyproject.toml`, with `concurrency = multiprocessing` so the design worker and fit pools
+  count on every platform; `patch = fork` is unsupported on Windows). Both test jobs upload
+  `coverage.xml` to Codecov with `CODECOV_TOKEN`, and the release workflow passes its secrets
+  to the tests it calls.
+
+
+## The mix plateau diagnosis tracks from is the pipeline's (17 October 2026)
+
+`diagnose` found the mix plateau on the spectrum cut off at the analysis band's top, while
+`pipeline` finds it on the whole spectrum. `plateau_reference`'s grid ends exactly at the band
+top with `right=nan`, so the cut-off copy always lost the top grid point, and a plateau reaching
+that edge could vanish. Diagnosis then took `_temporal_evidence`'s early return (no strata, the
+tracking floor set to the band top), and the pipeline's plateau blocker read that as "nothing
+below it tracks the programme (tracking stops at 200.0 Hz)". The mix plateau is now found on
+the whole spectrum, as the pipeline finds it. Channel plateaus are unchanged.
+
+* **Found in the full-library run (beqdesigner).** A Few Good Men (5.1) declined
+  `no_usable_plateau`: the pipeline's plateau was 150.5-200 Hz, diagnosis had none, and its
+  record showed no strata and a tracking floor of exactly 200.0. `band_tracking` against the
+  pipeline's plateau passes every octave band to ~13 Hz (0.53-0.62). Patched, it accepts two
+  low shelves (16.8 Hz +5.8 dB, 26.8 Hz +6.5 dB, confidence 0.91). Before the Devil Knows
+  You're Dead and Blink Twice show the same record signature.
+* **First version: every plateau on the whole spectrum, rejected.** Handing channels the whole
+  spectrum too moved channel plateaus on eight baseline titles (an extra finite top point moves
+  the 90th-percentile threshold), changed counterfactual targets on four, and in real runs
+  against HEAD flipped counterfactual verdicts on Alto Knights and Obsession. As with the
+  one-sided fallback above, the channel calls were left alone.
+* **Validation of the mix-only version**, against 7a1702f. Probe: 28 Years Later's tracking
+  floor 20.50 -> 20.30 Hz (judged band and targets by up to 0.17 dB); Dossier 137's filter
+  floor 59.3 -> 79.8 Hz and Send Help's 20.0 -> 20.8 Hz, nothing else; the other eight titles
+  unchanged. 28 Years Later's real run against a HEAD run: the winner changes `parametric` ->
+  `flatten`, both passing either way; the accepted corrections agree within 0.6 dB in every band
+  from 5 to 80 Hz (+5.8/+6.0/+6.6/+3.3/+0.6 -> +5.9/+6.6/+6.4/+3.1/+0.7 dB), the same answer.
+  Protocol, seeds 101 and 947: identical false acceptances, abstentions, selections and recovery.
+  Corpus, seeds 1-9: gated 0/45, gate pass 0.079; positives 7/9 (median recovery 1.27 dB RMS);
+  `natural_droop` 4/9 against the committed file's 5/9 (an older revision; reported, not gated).
+  Full suite: 804 passed, 2 skipped. New test in `tests/test_design_diagnose.py`: a mix
+  high-passed at 144-146 Hz, whose only plateau reaches the band top, is tracked from that
+  plateau (fails on the old code: no strata, floor at 200 Hz).
+
+
 ## Levels beside an omission, and mains-hum notches as texture (11 October 2026)
 
 Three titles from a beqdesigner run against `/media/beq` were declined in ways that should not