spectro-kernel 0.9.0__tar.gz → 0.9.1__tar.gz

{spectro_kernel-0.9.0 → spectro_kernel-0.9.1}/CHANGELOG.md

@@ -6,6 +6,163 @@ Until `1.0.0` the public API may change between minor versions.
 
 ## [Unreleased]
 
+## [0.9.1] — 2026-11-09
+
+A correction release driven by a reader's report on the public MCP server
+(broken GitHub links, a `backend` field misread, one description that did
+not match its code, one duplicated constant) and by the full audit it
+prompted: every brick's text was compared with its code. Three kinds of
+changes, kept strictly apart :
+
+- **Kernel fixes** with no numeric effect on any brick ;
+- **Behaviour corrections** on nine bricks that did not do what their
+  description said. None of them is used by a downstream application
+  (checked against NORVIS Redux, LineMill, SpecRelay and the BeSS
+  projects before the change) ;
+- **Documentation and metadata** on forty bricks, including six used by
+  NORVIS Redux and LineMill, where only the text changes.
+
+Two findings that *would* change results of bricks NORVIS Redux and
+LineMill run today are deliberately **not** applied and are listed under
+"Known" below for a separate decision.
+
+### Fixed — kernel
+
+- `get_algorithm_source` (library, `spectro source`, MCP tool) rebuilds the
+  repository path from the package directory, so a wheel install reports
+  `src/spectro_kernel/algorithms/<category>/<file>.py` and a working GitHub
+  link instead of the bare file name. The public server, installed from the
+  wheel, returned `snr_der.py` for every brick. Tests cover the editable,
+  wheel and unknown layouts.
+- `algorithms/_robust.py` defines `MAD_TO_SIGMA` (1.482602) and
+  `DER_SNR_NOISE_FACTOR` (1.482602 / √6, Stoehr et al. 2008) once ; seven
+  copies replaced. The three bricks that carried the rounded literal
+  `0.6052697` — `match_lamp_lines` 1.0.3, `reidentify_arc_features` 1.0.1,
+  `measure_arc_geometry` 2.0.1 — now use the exact value : a 5e-8 relative
+  shift of a peak-detection threshold that is never exported ; on every test
+  fixture the closest peak candidate sits 1.8 % from the threshold.
+- The documented chain *extract → calibrate → `refine_wavelength_zeropoint`
+  / `measure_line_offsets`* failed after a fitted calibration, because only
+  the coefficient mode of `wavelength_calibrate_polynomial` cleared the
+  extractor's `needs_wavelength_calibration` flag (0.9.0 regression).
+  `wavelength_calibrate_polynomial` 1.1.1 (fit mode),
+  `wavelength_calibration_solar` 2.0.1 and `wavelength_calibrate_easyspec`
+  1.0.2 now set `wavelength_calibrated = True` and
+  `needs_wavelength_calibration = False` ; the axes are unchanged.
+- The documentation site builds on every push (the Netlify `ignore` script
+  had skipped the 0.9.0 deploy).
+
+### Changed — behaviour aligned on the description (bricks used by no application)
+
+- `gaia_query` 1.1.0 — `data_release` accepts the release aliases the
+  description promised (`DR3`, `EDR3`, `DR2`, case-insensitive) or a full
+  table name ; anything else is refused instead of being pasted into the
+  ADQL `FROM`, where `'DR3'` made the query fail.
+- `vizier_query` 2.0.0 — cone search of `radius_arcsec` around the resolved
+  object (`Vizier.query_region`) ; 1.x called `query_object` and ignored the
+  radius it documented.
+- `rv_precision_bouchy` 3.0.0 — without an uncertainty array the photon
+  noise σ_F ∝ √|F| is scaled to the spectrum's own DER_SNR noise, as the
+  description always said ; `snr_floor` is only the fallback when that
+  estimate is unavailable (2.x always used `snr_floor`, so σ_v scaled with
+  an assumption, never with the data).
+- `bpt_line_ratios` 3.0.0 — `log_sii_ha` needs both [S II] lines ; 2.x
+  reported (6716 + 6731)/Hα with one member of the doublet.
+- `doppler_tomogram` 3.0.0 — velocities a spectrum does not cover stay NaN
+  and are excluded from the back-projection average, as the NaN-aware
+  description and the code comment said ; 2.x zeroed them and counted them
+  as phases with no excess, diluting the map.
+- `merge_echelle_orders` 2.0.0 — a non-finite flux sample carries no weight ;
+  1.x kept its weight in the denominator and pulled the merge toward 0
+  under a NaN gap.
+- `read_echelle_fits` 2.0.0 — IRAF multispec `WAT2_*` records are decoded
+  (linear and log dispersions, `z` factor) ; a non-linear record
+  (Chebyshev / Legendre / spline, not decoded) leaves the order on a
+  flagged pixel axis (`meta["wcs"]`) instead of the shared linear WCS,
+  which gave a silently wrong axis. 1.x never read `WAT*` although its
+  docstring said so.
+- `embed_pretrained` 2.1.0 — `model_path` may be a directory holding
+  `model.pt`, as documented.
+- `plot_dynamic_spectrum` 1.1.0 — y labels are `DATE-OBS` or the exposure
+  index ; the object-name fallback collapsed every row of a single target
+  onto one label.
+
+### Documentation and metadata (PATCH bumps, no numeric change)
+
+- `measure_arc_geometry` : one joint least-squares fit of every (row, dx)
+  sample, no per-row or per-line averaging ; the `smile_order` 4 / 6 terms
+  are fitted but only dy² enters R ; implemented here on scipy / numpy
+  primitives (the `backend = "scipy"` convention : *implemented on
+  numerical primitives*, not *wrapped*).
+- `correct_tilt_affine` / `correct_slant_affine` / module docstring : the two
+  physical descriptions were swapped — tilt levels the trace (vertical
+  shear), slant straightens the monochromatic lines (horizontal shear) ; the
+  shear uses the angle in radians, ≈ tan θ for the small angles involved.
+- `clip_cosmic_rays` : astroscrappy's default clean is a 5×5 masked mean,
+  not a local median ; `backend = "astroscrappy"`.
+- `classify_template_chi2` : `top_n` includes the best match ;
+  `min_overlap` gates the observed spectrum (templates are dropped below
+  100 shared samples) ; `teff_unstable` needs templates with a `T_eff`
+  (Pickles route), so it stays False and `best_teff_k` is absent on the
+  `atlas_dir` route.
+- `extract_spectrum_easyspec` : `trace_method` is argmax / moments / multi,
+  `extraction_weights` gaussian / tophat, `shift_y_pixels` is the sky-window
+  offset raised to 3 × `trace_half_width` by easyspec ; the Monte-Carlo
+  errors are not kept (no uncertainty on the output).
+- `cosmic_ray_remove_easyspec` : the gain correction is not optional, the
+  output is always in electrons.
+- `wavelength_calibration_in_situ` : the uniform grid is built from the two
+  linear coefficients only and out-of-axis samples are filled with 0 ;
+  `fit_emission_lines_gaussian` writes `extras["emission_line_centroids"]`.
+- `match_lamp_lines` : `wave_min` / `wave_max` declared required (the brick
+  already refused their absence ; `spectro preset validate` now catches it) ;
+  the hand-off to `wavelength_calibrate_polynomial` goes through its
+  parameters, as `examples/wavelength_calibration.py` shows ; peaks are the
+  most prominent, noise is DER_SNR ; `lamp_atlas` intensities are
+  informative only.
+- `vr_ratio` : the most prominent peak per side, not the highest sample.
+- `detect_trace` : the FWHM helper returns the span of samples above
+  half-maximum, no interpolation, no restriction to the centre.
+- `wavelength_calibration_solar` : prominence as a fraction of the range,
+  matching in wavelength space.
+- `subtract_sky_2d` `sky_half_width` is the window height ;
+  `extract_spectrum_optimal` flanking sky fit is unweighted and its lower
+  window starts one row further than `subtract_sky_2d`'s.
+- `simbad_query`, Plotly bricks : they register without the optional
+  library and fail at run time with an install hint.
+- `flat_combine`, `flat_normalize`, `flat_combine_easyspec` : point at
+  `flat_normalize_response` and `flat_normalize_easyspec`, which exist.
+- `read_sdss_spectrum`, `compare_line_fits` (reports `best_profile`),
+  `compare_snr_methods`, `fit_telluric_scaling` (both extras keys),
+  `extract_spectrum_sum` (row, not column), `bias_combine` (feeds
+  `subtract_bias`), `_line_flux.continuum` (fitted offset),
+  `sii_electron_density` (fails / NaN), `export_fits_bess` (sentinel for
+  masked pixels only), `embed_band_power` (log10 of |power|),
+  `denoise_median_2d` border, `atmospheric_extinction_correct` (geometric
+  zenith angle), `activity_index_caii_hk` (`s_index_calibrated` is
+  α·S + β), `combine_spectra_arithmetic` (pass-through inside reference
+  NaN gaps too), `doppler_tomogram` (untimed spectra skipped, odd
+  `n_velocity` optional), `normalize_spline` (knots include the boundary),
+  embedding bricks (how to deal with masked samples),
+  `wavelength_calibrate_easyspec` (the applied polynomial is
+  `numpy.polyfit` on the identifications).
+- Backend labels : `export_hdf5` h5py, `doppler_shift` numpy,
+  `vsini_fourier` numpy, `correct_smile_xcorr` scipy.
+
+### Known — not changed, for a separate decision (bricks in production)
+
+- `wavelength_calibration_in_situ` (NORVIS Redux, LineMill) : the uniform
+  grid should span the polynomial over the whole axis and fill with NaN,
+  and the brick should clear `needs_wavelength_calibration` like the other
+  calibrators (a NORVIS `WAVCAL` header would flip to True).
+- `correct_tilt_affine` / `correct_slant_affine` (NORVIS Redux, LineMill) :
+  the shear should be tan θ rather than θ (0.1 % at 3°).
+- `detect_trace` (LineMill) : the FWHM should be interpolated at half-maximum
+  around the central row ; a second source in the slit currently widens it
+  and the SNR exclusion zone with it.
+- `response_from_standard` (LineMill) : `n_fit_samples` and `response_rms`
+  reported from the post-clip mask ; to be checked against the final fit.
+
 ## [0.9.0] — 2026-10-23
 
 Bricks for self-calibrating long-slit instruments, requested by NORVIS Redux,

{spectro_kernel-0.9.0 → spectro_kernel-0.9.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: spectro-kernel
-Version: 0.9.0
+Version: 0.9.1
 Summary: A shared catalogue of astronomical spectroscopy algorithms, composable into reproducible pipelines, usable as a Python library, a CLI, or an MCP server.
 Project-URL: Homepage, https://github.com/matthieulel/spectro-kernel
 Project-URL: Documentation, https://docs.spectrokernel.io/

spectro_kernel-0.9.1/src/spectro_kernel/algorithms/_robust.py

@@ -0,0 +1,29 @@
+"""Robust noise constants shared by the bricks (one definition, seven former copies).
+
+``MAD_TO_SIGMA`` converts a median absolute deviation to a Gaussian σ
+(1 / Φ⁻¹(3/4) ≈ 1.4826). ``DER_SNR_NOISE_FACTOR`` is the DER_SNR noise scale
+of Stoehr et al. 2008 (ASP Conf. Ser. 394, 505): σ = 1.482602 / √6 ·
+median(|2 F[i] − F[i−2] − F[i+2]|).
+"""
+
+from __future__ import annotations
+
+import math
+
+import numpy as np
+
+MAD_TO_SIGMA = 1.482602
+DER_SNR_NOISE_FACTOR = MAD_TO_SIGMA / math.sqrt(6.0)
+
+
+def der_snr_noise(flux: np.ndarray) -> float:
+    """DER_SNR noise estimate of a 1-D flux array (NaN-aware, 0.0 below 5 samples)."""
+    flux = np.asarray(flux, dtype=np.float64)
+    if flux.size < 5:
+        return 0.0
+    diff = np.abs(2.0 * flux[2:-2] - flux[:-4] - flux[4:])
+    diff = diff[np.isfinite(diff)]
+    return DER_SNR_NOISE_FACTOR * float(np.median(diff)) if diff.size else 0.0
+
+
+__all__ = ["DER_SNR_NOISE_FACTOR", "MAD_TO_SIGMA", "der_snr_noise"]

{spectro_kernel-0.9.0 → spectro_kernel-0.9.1}/src/spectro_kernel/algorithms/activity/activity_index_caii_hk.py

@@ -116,7 +116,7 @@ def log_rhk_noyes(s_mw: float, b_v: float) -> float | None:
     category=AlgorithmCategory.ACTIVITY,
     # 2.0.0: masked samples (Spectrum1D.mask) are dropped from the passband
     # integrals like non-finite ones.
-    version="2.0.0",
+    version="2.0.1",
 )
 class ActivityIndexCaiiHk(BaseAlgorithm):
     """Mount Wilson S index from the Ca II H & K cores, plus R'_HK when B−V is given.
@@ -162,7 +162,8 @@ class ActivityIndexCaiiHk(BaseAlgorithm):
         "calibrated on 0.44 < B−V < 0.82 : outside that range log_rhk is "
         "still written but flagged in the message. Rejects (no log_rhk) when "
         "R_HK ≤ R_phot. Metrics are prefixed for provenance: "
-        "s_index_alpha and s_index_calibrated echo the inputs. v2.0.0: masked "
+        "s_index_alpha echoes the input, s_index_calibrated is the computed "
+        "alpha · S + beta. v2.0.0: masked "
         "samples (Spectrum1D.mask) are ignored — dropped from the passband "
         "integrals (and their transmission from the weight sum) like "
         "non-finite ones."

{spectro_kernel-0.9.0 → spectro_kernel-0.9.1}/src/spectro_kernel/algorithms/advanced/doppler_tomogram.py

@@ -32,8 +32,9 @@ that suppresses the streak artefacts of pure back-projection. MEM is
 intentionally left out of v1.0.0 to keep the surface tight.
 
 Phase computation : ``φ = ((JD_mid − epoch_HJD) / period) mod 1``.
-Each input ``Spectrum1D`` must carry its **mid-exposure JD** in
-``spec.meta["jd_mid"]`` ; if absent, the algorithm falls back to
+Each input ``Spectrum1D`` should carry its **mid-exposure JD** in
+``spec.meta["jd_mid"]`` (a spectrum with no usable timestamp is skipped
+and counted in the message) ; if absent, the algorithm falls back to
 deriving it from ``spec.meta["dateobs"]`` and ``spec.meta["exptime_s"]``
 (``jd_mid = Time(dateobs).jd + exptime_s / 2 / 86400``), which requires
 ``astropy.time`` (imported lazily).
@@ -85,7 +86,9 @@ def _resolve_jd_mid(spec: Spectrum1D) -> float | None:
 
 
 def _line_excess_on_velocity_grid(
-    spec: Spectrum1D, line_centre_aa: float, velocity_grid: np.ndarray,
+    spec: Spectrum1D,
+    line_centre_aa: float,
+    velocity_grid: np.ndarray,
 ) -> np.ndarray | None:
     """Resample the spectrum onto a velocity grid and subtract the continuum.
 
@@ -118,9 +121,9 @@ def _line_excess_on_velocity_grid(
 
 
 def _back_project(
-    trailed: np.ndarray,             # (n_phases, n_velocity)
-    phases: np.ndarray,              # (n_phases,)
-    velocity_grid: np.ndarray,       # (n_velocity,)
+    trailed: np.ndarray,  # (n_phases, n_velocity)
+    phases: np.ndarray,  # (n_phases,)
+    velocity_grid: np.ndarray,  # (n_velocity,)
     gamma_kms: float,
     n_velocity: int,
 ) -> np.ndarray:
@@ -157,7 +160,9 @@ def _back_project(
     # 2.0.0: back-projection locus follows Marsh & Horne 1988 Eq. 1
     # (v = γ − Vx·cos 2πφ + Vy·sin 2πφ); v1 had sin/cos swapped and
     # produced maps reflected with respect to the literature convention.
-    version="2.0.0",
+    # 3.0.0: velocities a spectrum does not cover stay NaN and are excluded from
+    # the back-projection average (2.x set them to 0 and counted them).
+    version="3.0.0",
 )
 class DopplerTomogram(BaseAlgorithm):
     """Doppler tomogram of a binary from N phase-resolved spectra.
@@ -190,8 +195,7 @@ class DopplerTomogram(BaseAlgorithm):
 
     backend = "numpy"
     references = [
-        "Marsh & Horne 1988, MNRAS 235, 269 — Doppler tomography by "
-        "back-projection (§2).",
+        "Marsh & Horne 1988, MNRAS 235, 269 — Doppler tomography by back-projection (§2).",
         "Horne 1985, MNRAS 213, 129 — MEM tomographic reconstruction "
         "(not implemented; cited for completeness).",
         "Marsh 2001, Lecture Notes in Physics 573 — astrotomography review.",
@@ -225,7 +229,9 @@ class DopplerTomogram(BaseAlgorithm):
         "gamma_kms": "Systemic velocity (km/s) added to the back-projection.",
         "line_center_aa": "Rest wavelength (Å) of the line to tomogram.",
         "velocity_window_kms": "Half-extent (km/s) of the V_x/V_y axes.",
-        "n_velocity": "Number of velocity samples per axis (odd, 11–401).",
+        "n_velocity": (
+            "Number of velocity samples per axis (11–401 ; an odd value puts a sample at 0)."
+        ),
     }
     input_requirements: list[str] = []  # validated dynamically
     output_produces = [
@@ -240,9 +246,7 @@ class DopplerTomogram(BaseAlgorithm):
     def run(self, ctx: WorkContext, params: dict[str, Any]) -> AlgorithmOutput:
         spectra = list(ctx.spectra or [])
         if len(spectra) < 3:
-            return AlgorithmOutput.fail(
-                f"Need ≥ 3 spectra in ctx.spectra; got {len(spectra)}."
-            )
+            return AlgorithmOutput.fail(f"Need ≥ 3 spectra in ctx.spectra; got {len(spectra)}.")
 
         period = float(params["period_days"])
         if period <= 0.0:
@@ -284,11 +288,15 @@ class DopplerTomogram(BaseAlgorithm):
         rows.sort(key=lambda row: row[0])
         phases = np.asarray([row[0] for row in rows], dtype=np.float64)
         trailed = np.stack([row[1] for row in rows], axis=0)
-        # NaN-aware: replace residual NaNs from np.interp by 0 before
-        # accumulation (they don't contribute, and "no data" ≠ "0 signal").
-        trailed_clean = np.where(np.isfinite(trailed), trailed, 0.0)
+        # NaN-aware: velocities a spectrum does not cover stay NaN; np.interp
+        # propagates them and _back_project excludes them from the average
+        # ("no data" ≠ "0 signal" — 2.x zeroed them and counted them).
         doppler_map = _back_project(
-            trailed_clean, phases, velocity_grid, gamma_kms, n_velocity,
+            trailed,
+            phases,
+            velocity_grid,
+            gamma_kms,
+            n_velocity,
         )
 
         result = {

{spectro_kernel-0.9.0 → spectro_kernel-0.9.1}/src/spectro_kernel/algorithms/catalogs/gaia.py

@@ -5,11 +5,35 @@ from __future__ import annotations
 from typing import Any
 
 from ...base import AlgorithmOutput, BaseAlgorithm
+from ...errors import InvalidParameterError
 from ...registry import register_algorithm
 from ...types import AlgorithmCategory, CatalogResult, WorkContext
 
+_RELEASE_TABLES = {
+    "dr3": "gaiadr3.gaia_source",
+    "edr3": "gaiaedr3.gaia_source",
+    "dr2": "gaiadr2.gaia_source",
+}
 
-@register_algorithm("gaia_query", category=AlgorithmCategory.CATALOG, version="1.0.0")
+
+def _release_table(value: object) -> str:
+    """ADQL table of a release alias ('DR3', 'EDR3', 'DR2') or a full table name."""
+    if value is None or str(value).strip() == "":
+        return _RELEASE_TABLES["dr3"]
+    key = str(value).strip().lower().replace("gaia", "").replace(" ", "")
+    if key in _RELEASE_TABLES:
+        return _RELEASE_TABLES[key]
+    if "." in str(value):
+        return str(value).strip()
+    raise InvalidParameterError(
+        f"data_release {value!r} is neither a release alias (DR3, EDR3, DR2) nor a "
+        "table name such as 'gaiadr3.gaia_source'."
+    )
+
+
+# 1.1.0: data_release accepts the release aliases the description always
+# promised (DR3, DR2, EDR3, case-insensitive) next to a full table name.
+@register_algorithm("gaia_query", category=AlgorithmCategory.CATALOG, version="1.1.0")
 class GaiaQuery(BaseAlgorithm):
     """Cone-search the Gaia archive around ICRS coordinates.
 
@@ -28,9 +52,10 @@ class GaiaQuery(BaseAlgorithm):
         "Ginsburg et al. 2019, AJ 157, 98 — astroquery",
     ]
     long_description = (
-        "Requires network access and the optional 'catalogs' extra. By default "
-        "queries the latest release available via astroquery.gaia. Use `data_release` "
-        "to pin a specific release such as 'DR3' or 'DR2'."
+        "Requires network access and the optional 'catalogs' extra. Queries "
+        "gaiadr3.gaia_source by default (the latest full release). `data_release` "
+        "accepts a release alias ('DR3', 'EDR3', 'DR2', case-insensitive) or a full "
+        "table name such as 'gaiadr2.gaia_source'."
     )
     default_params = {
         "ra_deg": None,
@@ -43,7 +68,10 @@ class GaiaQuery(BaseAlgorithm):
         "ra_deg": "Right ascension (ICRS) in degrees.",
         "dec_deg": "Declination (ICRS) in degrees.",
         "radius_arcsec": "Cone-search radius around the position.",
-        "data_release": "Gaia release table to query (e.g. 'gaiadr3.gaia_source').",
+        "data_release": (
+            "Gaia release: an alias ('DR3', 'EDR3', 'DR2') or a full table name "
+            "('gaiadr3.gaia_source'); null means DR3."
+        ),
     }
     input_requirements: list[str] = []
     output_produces = ["catalog_lookups"]
@@ -53,14 +81,13 @@ class GaiaQuery(BaseAlgorithm):
             from astroquery.gaia import Gaia as _Gaia
         except ImportError:
             return AlgorithmOutput.fail(
-                "gaia_query needs astroquery. "
-                "Install with: pip install 'spectro-kernel[catalogs]'."
+                "gaia_query needs astroquery. Install with: pip install 'spectro-kernel[catalogs]'."
             )
 
         ra = float(params["ra_deg"])
         dec = float(params["dec_deg"])
         radius_deg = float(params["radius_arcsec"]) / 3600.0
-        table_name = params.get("data_release") or "gaiadr3.gaia_source"
+        table_name = _release_table(params.get("data_release"))
         adql = (
             f"SELECT TOP 1 source_id, ra, dec, parallax, phot_g_mean_mag, "
             f"DISTANCE(POINT('ICRS', ra, dec), POINT('ICRS', {ra}, {dec})) AS sep "
@@ -81,7 +108,7 @@ class GaiaQuery(BaseAlgorithm):
             ctx.catalog_lookups[key] = result
             return AlgorithmOutput.ok(
                 artifacts={"result": result.to_dict()},
-                message=f"Gaia returned no source within {params['radius_arcsec']}\".",
+                message=f'Gaia returned no source within {params["radius_arcsec"]}".',
             )
 
         row = table[0]
@@ -95,9 +122,7 @@ class GaiaQuery(BaseAlgorithm):
             fields={
                 "parallax_mas": float(row["parallax"]) if row["parallax"] is not None else None,
                 "G_mag": (
-                    float(row["phot_g_mean_mag"])
-                    if row["phot_g_mean_mag"] is not None
-                    else None
+                    float(row["phot_g_mean_mag"]) if row["phot_g_mean_mag"] is not None else None
                 ),
                 "separation_deg": float(row["sep"]),
             },
@@ -105,5 +130,5 @@ class GaiaQuery(BaseAlgorithm):
         ctx.catalog_lookups[source_id] = result
         return AlgorithmOutput.ok(
             artifacts={"result": result.to_dict()},
-            message=f"Gaia source {source_id} at {row['sep']*3600:.2f}\".",
+            message=f'Gaia source {source_id} at {row["sep"] * 3600:.2f}".',
         )

{spectro_kernel-0.9.0 → spectro_kernel-0.9.1}/src/spectro_kernel/algorithms/catalogs/simbad.py

@@ -1,8 +1,8 @@
 """Algorithm: resolve an object against the SIMBAD database.
 
 Requires the optional ``catalogs`` extra (``pip install spectro-kernel[catalogs]``).
-If astroquery is not installed the algorithm simply does not register, and the rest of
-the catalogue loads normally.
+The algorithm registers whether or not astroquery is installed; without it, running
+the brick fails with an install hint and the rest of the catalogue is unaffected.
 """
 
 from __future__ import annotations
@@ -25,7 +25,7 @@ def _cell(row: Any, *names: str) -> Any:
     return None
 
 
-@register_algorithm("simbad_query", category=AlgorithmCategory.CATALOG, version="1.0.0")
+@register_algorithm("simbad_query", category=AlgorithmCategory.CATALOG, version="1.0.1")
 class SimbadQuery(BaseAlgorithm):
     """Resolve an object name against SIMBAD and store the record in the context.
 

{spectro_kernel-0.9.0 → spectro_kernel-0.9.1}/src/spectro_kernel/algorithms/catalogs/vizier.py

@@ -5,11 +5,14 @@ from __future__ import annotations
 from typing import Any
 
 from ...base import AlgorithmOutput, BaseAlgorithm
+from ...errors import InvalidParameterError
 from ...registry import register_algorithm
 from ...types import AlgorithmCategory, CatalogResult, WorkContext
 
 
-@register_algorithm("vizier_query", category=AlgorithmCategory.CATALOG, version="1.0.0")
+# 2.0.0: cone search of radius_arcsec around the resolved object (query_region);
+# 1.x called query_object and ignored radius_arcsec.
+@register_algorithm("vizier_query", category=AlgorithmCategory.CATALOG, version="2.0.0")
 class VizierQuery(BaseAlgorithm):
     """Look up an object in VizieR — the CDS table service.
 
@@ -54,12 +57,17 @@ class VizierQuery(BaseAlgorithm):
             )
 
         name = str(params["object_name"])
+        if float(params["radius_arcsec"]) <= 0.0:
+            raise InvalidParameterError("radius_arcsec must be > 0.")
         try:
+            from astropy.coordinates import Angle  # noqa: PLC0415
+
+            radius = Angle(float(params["radius_arcsec"]), unit="arcsec")
             v = _Vizier(row_limit=int(params["row_limit"]))
             tables = (
-                v.query_object(name, catalog=params["catalog"])
+                v.query_region(name, radius=radius, catalog=params["catalog"])
                 if params["catalog"]
-                else v.query_object(name)
+                else v.query_region(name, radius=radius)
             )
         except Exception as exc:  # noqa: BLE001 - network/parse errors
             return AlgorithmOutput.fail(f"VizieR query failed: {exc}")

{spectro_kernel-0.9.0 → spectro_kernel-0.9.1}/src/spectro_kernel/algorithms/classification/classify_template_chi2.py

@@ -15,7 +15,9 @@ top-N matches and a couple of confidence flags :
   continua).
 - ``teff_unstable`` — True when ``max − min`` of the top-3 ``T_eff``
   exceeds 500 K (the top matches span more than one effective
-  temperature bin).
+  temperature bin). It needs templates with a ``T_eff``: the Pickles
+  atlas route provides one, templates loaded from ``atlas_dir`` do not,
+  so the flag stays False and ``best_teff_k`` is absent there.
 
 This is intentionally the unweighted χ² ; the kernel does not yet
 carry a per-pixel uncertainty for normalised flux. Modern alternatives
@@ -126,7 +128,7 @@ def _normalised_template(atlas: Atlas, template_id: str, window: int) -> np.ndar
     # 1.1.0: additive `atlas_dir` (path) alternative to the Python `atlas`
     # object so the brick is CLI/MCP-drivable ; per-template normalisation
     # cached on the Atlas ; descending wavelength axes sorted. Numbers unchanged.
-    version="1.1.0",
+    version="1.1.1",
 )
 class ClassifyTemplateChi2(BaseAlgorithm):
     """Spectral classification by χ² against a Pickles-style template atlas.
@@ -144,7 +146,7 @@ class ClassifyTemplateChi2(BaseAlgorithm):
     Output (under ``ctx.extras["classification_report"]``) :
 
     - ``best`` : top match (``spec_type``, ``T_eff``, χ², coverage).
-    - ``top_n`` : the next ``top_n - 1`` matches.
+    - ``top_n`` : the ``top_n`` best matches, the best one included.
     - ``ambiguous`` : χ²[1] / χ²[0] < 1.20.
     - ``teff_unstable`` : ``T_eff`` span across the top 3 > 500 K.
 
@@ -164,8 +166,9 @@ class ClassifyTemplateChi2(BaseAlgorithm):
         "Common wavelength grid (default 3800–7500 Å at 2 Å step, "
         "matching pickles_atlas.canonical_grid()), running-median "
         "continuum normalisation (window 80 samples), unweighted χ² over "
-        "the samples shared by observed + template. Templates with "
-        "useable overlap < min_overlap are skipped. Confidence flags : "
+        "the samples shared by observed + template. The run fails when the "
+        "observed spectrum covers less than min_overlap of the atlas grid ; a "
+        "template is dropped when fewer than 100 samples overlap. Confidence flags : "
         "ambiguous (χ²[1]/χ²[0] < 1.20) and teff_unstable (top-3 ΔT_eff "
         "> 500 K). Supply either `atlas` (an Atlas object, Python API) or "
         "`atlas_dir` (path to the Pickles .dat files, loaded with "
@@ -201,7 +204,10 @@ class ClassifyTemplateChi2(BaseAlgorithm):
             "atlas is given."
         ),
         "top_n": "Number of best-fit matches to report (≥ 1).",
-        "min_overlap": "Reject the run if observed-vs-grid overlap drops below this fraction.",
+        "min_overlap": (
+            "Fail the run when the observed spectrum covers less than this fraction "
+            "of the atlas grid (templates themselves are dropped below 100 shared samples)."
+        ),
         "continuum_window": "Running-median window (samples) for continuum normalisation.",
         "grid_lo": "Reserved (currently unused): the atlas's own grid_aa is used.",
         "grid_hi": "Reserved (currently unused): the atlas's own grid_aa is used.",

{spectro_kernel-0.9.0 → spectro_kernel-0.9.1}/src/spectro_kernel/algorithms/continuum/normalize_spline.py

@@ -85,7 +85,8 @@ def fit_spline_continuum(
 ) -> tuple[np.ndarray, np.ndarray, np.ndarray, int]:
     """IRAF-style iteratively clipped cubic-spline continuum.
 
-    Returns ``(continuum, kept_mask, knots, n_iterations)`` with the
+    Returns ``(continuum, kept_mask, knots, n_iterations)`` — ``knots`` are
+    the knots actually used, boundary knots included — with the
     continuum evaluated on every ``x``. ``low_reject`` / ``high_reject``
     ≤ 0 disable rejection on that side, as in IRAF.
     """
@@ -137,7 +138,7 @@ def fit_spline_continuum(
     # 2.0.0: masked samples (Spectrum1D.mask) are excluded from the spline fit
     # and from continuum_rms / n_rejected ; the output keeps their flux,
     # divided like every other sample, and carries the mask through.
-    version="2.0.0",
+    version="2.0.1",
 )
 class NormalizeSpline(BaseAlgorithm):
     """Normalise the continuum with an IRAF-style iteratively clipped cubic spline.

{spectro_kernel-0.9.0 → spectro_kernel-0.9.1}/src/spectro_kernel/algorithms/corrections/atmospheric_extinction.py

@@ -22,8 +22,8 @@ user table) is always preferable.
 The airmass comes from, in order : the ``airmass`` parameter ; the
 ``AIRMASS`` FITS keyword ; or the geometry (target RA / DEC, DATE-OBS at
 mid-exposure, observer site) through astropy ``AltAz`` and the
-Kasten & Young 1989 formula (their Eq. 3, z the apparent zenith
-angle in degrees) :
+Kasten & Young 1989 formula (their Eq. 3, z the zenith angle in
+degrees — geometric here, ``AltAz`` being evaluated without refraction) :
 
     X(z) = 1 / [cos z + 0.50572 · (96.07995 − z)^{−1.6364}]         (Eq. 2)
 
@@ -287,7 +287,7 @@ def _airmass_from_geometry(
 @register_algorithm(
     "atmospheric_extinction_correct",
     category=AlgorithmCategory.CORRECTION,
-    version="1.0.0",
+    version="1.0.1",
 )
 class AtmosphericExtinctionCorrect(BaseAlgorithm):
     """Remove atmospheric extinction : F_0 = F · 10^(0.4 · k(λ) · X) with a mean site curve.

{spectro_kernel-0.9.0 → spectro_kernel-0.9.1}/src/spectro_kernel/algorithms/corrections/doppler_shift.py

@@ -14,7 +14,7 @@ from ...types import AlgorithmCategory, WorkContext
 _C_KMS = c.to(u.km / u.s).value
 
 
-@register_algorithm("doppler_shift", category=AlgorithmCategory.CORRECTION, version="1.0.0")
+@register_algorithm("doppler_shift", category=AlgorithmCategory.CORRECTION, version="1.0.1")
 class DopplerShift(BaseAlgorithm):
     """Doppler-shift the wavelength axis by a radial velocity.
 
@@ -23,7 +23,7 @@ class DopplerShift(BaseAlgorithm):
     ``-v``. The flux array is unchanged.
     """
 
-    backend = "astropy"
+    backend = "numpy"
     references = [
         "Classical (non-relativistic) Doppler relation: lambda' = lambda * (1 + v/c)",
         "Speed of light from astropy.constants",

{spectro_kernel-0.9.0 → spectro_kernel-0.9.1}/src/spectro_kernel/algorithms/corrections/fit_telluric_scaling.py

@@ -16,7 +16,7 @@ from ...types import AlgorithmCategory, Spectrum1D, WorkContext
 @register_algorithm(
     "fit_telluric_scaling",
     category=AlgorithmCategory.CORRECTION,
-    version="1.0.1",
+    version="1.0.2",
 )
 class FitTelluricScaling(BaseAlgorithm):
     """Fit the airmass that best matches a telluric template to ``ctx.spectrum``.
@@ -25,7 +25,8 @@ class FitTelluricScaling(BaseAlgorithm):
     directly) and an observed spectrum, this finds the airmass exponent
     ``a*`` minimising the residual depth of telluric bands after dividing —
     Beer-Lambert says ``T_obs = T_template^a``. The fitted airmass plus the
-    rescaled template land in ``ctx.extras['telluric_fit']``; chain into
+    rescaled template land in ``ctx.extras['telluric_fit']`` and
+    ``ctx.extras['telluric_template_scaled']``; chain into
     ``remove_telluric_division`` to actually apply the correction.
     """
 

{spectro_kernel-0.9.0 → spectro_kernel-0.9.1}/src/spectro_kernel/algorithms/embedding/embed_band_power.py

@@ -24,7 +24,7 @@ from ...types import AlgorithmCategory, WorkContext
     # 2.0.0: half-open band edges (no double-counted samples)
     # 2.0.1: non-finite samples dropped ; descending axes sorted before binning.
     # 3.0.0: masked samples (Spectrum1D.mask) are dropped like non-finite ones.
-    version="3.0.0",
+    version="3.0.1",
 )
 class EmbedBandPower(BaseAlgorithm):
     """Embed a spectrum as the integrated flux in N adjacent wavelength bands.
@@ -68,7 +68,7 @@ class EmbedBandPower(BaseAlgorithm):
     param_descriptions = {
         "n_bands": "Number of bands; also the output dim. Typically 8–32.",
         "spacing": "Band edges in wavelength: 'linear' or 'log'.",
-        "log_flux": "Take log10(integrated_flux + epsilon) before L2 norm.",
+        "log_flux": "Take log10(|integrated_flux| + epsilon) before L2 norm.",
         "epsilon": "Small floor added before log_flux to keep zero bands finite.",
     }
     input_requirements = ["spectrum"]