Feature/noisepy eval

.github/workflows/publish.yml

@@ -0,0 +1,39 @@
+name: publish
+
+# Publishes to PyPI when a version tag is pushed (v*). Uses PyPI Trusted
+# Publishing (OIDC) — configure once at pypi.org: project seisfetch →
+# Publishing → add GitHub publisher (owner Denolle-Lab, repo seisfetch,
+# workflow publish.yml, environment pypi). No token secrets needed.
+
+on:
+  push:
+    tags: ["v*"]
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v5
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - run: pip install build twine
+      - run: python -m build
+      - run: twine check dist/*
+      - uses: actions/upload-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+  publish:
+    needs: build
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      id-token: write  # OIDC for trusted publishing
+    steps:
+      - uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+      - uses: pypa/gh-action-pypi-publish@release/v1

.github/workflows/test.yml

@@ -0,0 +1,62 @@
+name: test
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+  workflow_dispatch:
+
+jobs:
+  lint:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v5
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      # pinned to the same ruff as pre-commit/pixi — unpinned CI ruff
+      # drifted ahead and failed format --check on style changes
+      - run: pip install ruff==0.15.9
+      - run: ruff check seisfetch tests benchmarks
+      - run: ruff format --check seisfetch tests benchmarks
+
+  pytest:
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - { os: ubuntu-latest, python: "3.9" }
+          - { os: ubuntu-latest, python: "3.10" }
+          - { os: ubuntu-latest, python: "3.12" }
+          - { os: ubuntu-latest, python: "3.13" }
+          - { os: macos-latest, python: "3.12" }
+    runs-on: ${{ matrix.os }}
+    steps:
+      - uses: actions/checkout@v5
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python }}
+      - name: Install with dev extras
+        run: pip install -e ".[dev]"
+      - name: Unit tests (offline)
+        run: pytest tests -m "not integration" -q
+
+  # informational: does the NEXT pymseed break us? Never blocks merges.
+  pymseed-next:
+    runs-on: ubuntu-latest
+    continue-on-error: true
+    steps:
+      - uses: actions/checkout@v5
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - name: Install with latest pymseed (ignore our pin)
+        run: |
+          pip install -e ".[dev]"
+          pip install --upgrade pymseed
+          python -c "import pymseed; print('pymseed', pymseed.__version__)"
+      - name: Parse-path tests against latest pymseed
+        run: >
+          pytest tests/test_convert.py tests/test_segments_api.py
+          tests/test_correctness_majors.py
+          tests/precision/test_parse_identity.py -q

.pre-commit-config.yaml

@@ -1,6 +1,6 @@
 repos:
   - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.8.6
+    rev: v0.15.9
     hooks:
       - id: ruff
         args: [--fix, --exit-non-zero-on-fix]

CHANGELOG.md

@@ -0,0 +1,95 @@
+# Changelog
+
+All notable changes to seisfetch are documented here. The format follows
+[Keep a Changelog](https://keepachangelog.com/en/1.1.0/); versions follow
+[Semantic Versioning](https://semver.org).
+
+## 0.3.0 — 2026-08-04
+
+The obspy-replacement evaluation release: parse rewrite, gap-aware API,
+typed failure contract, instrument-response removal, and the full
+three-persona external critique with every blocker, correctness major, and
+operations finding resolved (`docs/reviews/2026-08-external-critique.md`).
+
+### Changed (breaking or behavior-visible)
+
+- **License** is now the compound expression `MIT AND LGPL-3.0-only`:
+  ObsPy-derived numerical translations live in
+  `seisfetch/contrib/obspy_ports.py` under LGPL-3.0-only; everything else
+  stays MIT. See THIRD_PARTY_NOTICES.md.
+- **Failure contract**: fetch paths no longer swallow errors into empty
+  bytes. Clean 404s are tolerated per key; any other failure raises
+  `FetchError` (`on_error="warn"` opts out); nothing-found raises
+  `NoDataError` unless `missing_ok=True`. FDSN maps 204/404 to no-data and
+  raises `FDSNError` on real HTTP errors.
+- **`location="*"` (the default) now works**: resolved by paginated LIST
+  discovery on SCEDC/NCEDC (finds location-coded channels instead of
+  guessing blank-location keys); FDSN passes `*` through and spells blank
+  as `--`.
+- **Day windows are half-open** `[start, end)`: a request ending at
+  midnight no longer fetches the following day's objects.
+- **`get_numpy`/`get_xarray` trim to the requested window by default**
+  (`trim=False` restores whole-object behavior) and filter station-day
+  objects to the requested channel/location after parse.
+- **`bundle_to_obspy`/`get_waveforms` return one Trace per segment**
+  (obspy-read parity; no masked arrays). `merge=1` restores the old
+  force-merge.
+- **`to_dict()` warns on gappy channels** (the legacy concatenation has a
+  wrong time axis after the first gap); `to_dict(fill_value=)` places
+  segments at true offsets with obspy `merge(method=1)` overlap semantics.
+- **Mixed sampling rates under one NSLC raise `MixedSamplingRateError`**
+  from `to_dict`/`metadata` (`segments()` is the per-rate escape hatch).
+- **`FDSNMultiClient` defaults to failover** (first non-empty provider
+  wins); the old broadcast is `strategy="broadcast"`.
+- **`fetch_bulk_numpy` drops raw bytes after parsing** (`keep_raw=True`
+  restores; byte accounting preserved).
+- **`parse_mseed(collect_flags=False)`**: per-record quality flags are now
+  opt-in; `TraceBundle.traces` holds true continuous segments (one per
+  contiguous run, not one per miniSEED record) and `num_segments` counts
+  real segments.
+- pymseed pinned `>=0.6,<0.10` (verified against 0.9.3); the private-API
+  sid fallback is isolated behind a guard.
+
+### Added
+
+- **Parse fast path** on libmseed's `MS3TraceList` with decode into
+  numpy-owned memory: 11 MB Steim2 channel-day 217 ms (v0.2.0) → ~21 ms,
+  faster than `obspy.read` in every measured environment, at ~1/3 the
+  parse memory.
+- **Gap-aware API**: `TraceBundle.segments()`, `trim()`, `overlaps()`,
+  `to_dict(fill_value=)`.
+- **`seisfetch.exceptions`**: `SeisfetchError`, `FetchError`,
+  `NoDataError`, `FDSNError`, `MixedSamplingRateError`.
+- **Instrument response removal without obspy**
+  (`seisfetch/contrib/response.py`): evalresp-equivalent evaluator
+  (1.6e-10 vs compiled evalresp, including its conditional A0 rule),
+  obspy-`remove_response` port (machine precision on real data, ~2x
+  faster), SeisIO.jl-style translation, stdlib StationXML parsing, loud
+  failures on defective metadata.
+- **NoisePy adapter** (`seisfetch/contrib/noisepy_adapter.py`): numpy
+  ports of NoisePy's `rm_resp=NO` preprocessing chain, validated exact
+  against obspy; end-to-end CCFs through real NoisePy are bit-identical.
+- **Operations hardening**: adaptive retries, connect/read timeouts,
+  pool sized to fan-out, one shared executor per client, paginated
+  listings, EarthScope credential refresh, `iter_bulk_raw` streaming.
+- **Benchmarks with committed results** (`benchmarks/results/*.json`,
+  `RESULTS.md`), docker machine matrix (Fargate/Lambda-class), CI
+  (offline pytest matrix + lint), tutorial notebook
+  `notebooks/05_response_removal.ipynb`.
+
+### Fixed
+
+- Silent gap concatenation in `to_dict`/`bundle_to_xarray` time axes.
+- Non-deterministic multi-day/multi-provider byte order (`as_completed`
+  joins → submission order).
+- Contained-segment crash in `to_dict(fill_value=)`.
+- Out-of-order contiguous records now heal identically on the fast and
+  fallback parse paths; truncated buffers warn about unparsed bytes.
+- BG (The Geysers) routed to NCEDC (was SCEDC).
+- Timezone-offset timestamps in response epoch selection.
+
+## 0.2.0 — 2026-04-27
+
+Initial public state: S3 clients for EarthScope/SCEDC/NCEDC, FDSN HTTP
+client, pymseed-based `parse_mseed`, bulk engine, xarray/zarr/obspy
+exporters, Earth2Studio adapters, CLI.

MANIFEST.in

@@ -0,0 +1,13 @@
+# Keep the sdist lean: source + licenses + docs metadata only.
+# Tests carry an 11 MB real-data fixture and notebooks carry executed
+# outputs — they live in the repo, not the distribution.
+include LICENSE THIRD_PARTY_NOTICES.md CHANGELOG.md README.md
+recursive-include seisfetch *.py
+prune tests
+prune benchmarks
+prune notebooks
+prune tools
+prune docs
+prune .github
+prune .jupyter-kernels
+exclude .pre-commit-config.yaml pixi.lock environment.yml .gitignore

README.md

@@ -592,6 +592,11 @@ pixi run test-int
 
 See [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md) for attribution and licenses.
 
+Benchmarks with plots: [benchmarks/RESULTS.md](benchmarks/RESULTS.md) (or the
+self-contained [RESULTS.html](benchmarks/RESULTS.html)). Changes:
+[CHANGELOG.md](CHANGELOG.md). Response-removal tutorial:
+[notebooks/05_response_removal.ipynb](notebooks/05_response_removal.ipynb).
+
 ## Citation
 
 When using data accessed through `seisfetch`:
@@ -609,4 +614,10 @@ Software references:
 
 ## License
 
-MIT. See [LICENSE](LICENSE).
+`MIT AND LGPL-3.0-only`. The package is MIT ([LICENSE](LICENSE)) with one
+exception: [`seisfetch/contrib/obspy_ports.py`](seisfetch/contrib/obspy_ports.py)
+contains numerically exact translations of ObsPy routines and is
+LGPL-3.0-only (ObsPy is © The ObsPy Development Team, LGPL v3). Using
+seisfetch as a library is unaffected; redistributors of modified versions of
+that one file must honor LGPL terms. Details in
+[THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md).

THIRD_PARTY_NOTICES.md

@@ -133,3 +133,30 @@ decoding) draws on the architecture established by:
 
 - **Clements & Denolle (2019)** — *Cactus to Clouds: Processing The SCEDC
   Open Data Set on AWS.* 2019 SCEC Annual Meeting.
+
+---
+
+## Derived code (license-bearing)
+
+- **`seisfetch/contrib/obspy_ports.py` — LGPL-3.0-only.** Contains Python
+  translations of ObsPy routines (`Trace.resample`, `Trace.taper`,
+  `obspy.signal.invsim.cosine_taper` / `cosine_sac_taper` /
+  `invert_spectrum`, `obspy.signal.util._npts2nfft`), preserving their exact
+  numerical behavior including float-operation order. ObsPy is
+  Copyright (C) The ObsPy Development Team, licensed under the GNU Lesser
+  General Public License v3 (https://github.com/obspy/obspy). These
+  translations are works based on the Library and are distributed under
+  LGPL-3.0-only; the file carries an SPDX header. The project license
+  expression is therefore `MIT AND LGPL-3.0-only`.
+- **Parts of `seisfetch/contrib/noisepy_adapter.py` — MIT.** The
+  `check_sample_gaps_np` / `segment_interpolate_np` / `preprocess_raw_np`
+  chain reimplements the preprocessing semantics of NoisePy
+  (`noisepy-seis`, MIT, Copyright (c) Marine Denolle & Chengxin Jiang;
+  https://github.com/noisepy/NoisePy).
+- **Parts of `seisfetch/contrib/response.py` — MIT.**
+  `translate_resp_np` and `damped_oscillator_response` reimplement the
+  response-translation approach of SeisIO.jl (`translate_resp!`,
+  `fctoresp`; MIT, https://github.com/jpjones76/SeisIO.jl). The
+  evalresp-equivalent evaluator in the same file was built by black-box
+  empirical verification against evalresp's observed behavior (clean-room;
+  no evalresp or ObsPy source translated).