PyAutoNervesBoard

1# version:2#   python_version_check: False  # uncomment to suppress the Python version warning3#                                # if running on a non-recommended Python (anything other than 3.12 / 3.13).45updates:6  iterations_per_quick_update: 1e99 # Non-linear search iterations between every quick update, which just displays the maximum likelihood model fit.7  iterations_per_full_update: 1e99  # Non-linear search iterations between every full update, which outputs all visuals and result fits (e.g. model.result, search.summary), this exits the search and can be slow.8  quick_update_background: false    # If True, perform_quick_update runs on a background daemon thread so the sampler is never blocked while visuals render. Requires the Analysis subclass to set `supports_background_update = True`.9  live_visual_update: false         # If True, quick-update visuals are pushed to a live surface (a Jupyter cell when in a kernel, a matplotlib viewer subprocess otherwise) in addition to being written to disk.10psf:11  use_fft_default: true              # If True, PSFs are convolved using FFTs by default, which is faster and uses less memory in all cases except for very small PSFs, False uses direct convolution.12grid:13  max_evaluation_grid_size: 1000   # An evaluation grid whose shape is adaptive chosen is used to compute quantities like critical curves, this integer is the max size of the grid ensuring faster run times.14inversion:15  check_reconstruction: true        # If True, the inversion's reconstruction is checked to ensure the solution of a meshs's mapper is not an invalid solution where the values are all the same.16  use_positive_only_solver: true      # If True, inversion's use a positive-only linear algebra solver by default, which is slower but prevents unphysical negative values in the reconstructed solutuion.17  use_edge_zeroed_pixels : true       # If True, the edge pixels of a pixelization are set to zero, which prevents unphysical values in the reconstructed solution at the edge of the pixelization.18  no_regularization_add_to_curvature_diag_value : 1.0e-3 # The default value added to the curvature matrix's diagonal when regularization is not applied to a linear object, which prevents inversion's failing due to the matrix being singular.19  use_border_relocator: true          # If True, by default a pixelization's border is used to relocate all pixels outside its border to the border.20  reconstruction_vmax_factor: 0.5   # Plots of an Inversion's reconstruction use the reconstructed data's bright value multiplied by this factor.21hpc:22  hpc_mode: false                   # If True, use HPC mode, which disables GUI visualization, logging to screen and other settings which are not suited to running on a super computer.23  iterations_per_quick_update: 250000 #  Non-linear search iterations between every quick update, which just displays the maximum likelihood model fit.24  iterations_per_full_update: 1e99  # Non-linear search iterations between every full update, which outputs all visuals and result fits (e.g. model.result, search.summary), this exits the search and can be slow.25  quick_update_background: false    # Keep off on HPC: background rendering pulls matplotlib into the sampler process even when no display is attached.26  live_visual_update: false         # Keep off on HPC: nodes are headless with no GUI and no notebook kernel, so the live display surface has nothing to attach to.27adapt:28  adapt_minimum_percent: 0.0129  adapt_noise_limit: 100000000.030numba:31  use_numba: true32  cache: true33  nopython: true34  parallel: false35output:36  force_pickle_overwrite: false     #   force_pickle_overwrite: false     # If True, pickle files output by a search (e.g. samples.pickle) are recreated when a new model-fit is performed.37  force_visualize_overwrite: false  # If True, visualization images output by a search (e.g. subplots of the fit) are recreated when a new model-fit is performed.38  info_whitespace_length: 80        # Length of whitespace between the parameter names and values in the model.info / result.info39  log_level: INFO                   # The level of information output by logging.40  log_to_file: false                # If True, outputs the non-linear search log to a file (and not printed to screen).41  log_file: output.log              # The name of the file the logged output is written to (in the non-linear search output folder)42  model_results_decimal_places: 3   # Number of decimal places estimated parameter values / errors are output in model.results.43  remove_files: false               # If True, all output files of a non-linear search (e.g. samples, visualization, etc.) are deleted once the model-fit has completed, such that only the .zip file remains.44  samples_to_csv: true              # If True, non-linear search samples are written to a .csv file.45  unconverged_sample_size : 100     # If outputting results of an unconverged search, the number of samples used to estimate the median PDF values and errors.46parallel:47  warn_environment_variables: true  # If True, a warning is displayed when the search's number of CPU > 1 and enviromment variables related to threading are also > 1.48profiling:49  parallel_profile: false           # If True, the parallelization of the fit is profiled outputting a cPython graph.50  repeats: 1                        # The number of repeat function calls used to measure run-times when profiling.51structures:52  native_binned_only: false           # If True, data structures are only stored in their native and binned format. This is used to reduce memory usage in autocti.53test:54  check_likelihood_function: true   # if True, when a search is resumed the likelihood of a previous sample is recalculated to ensure it is consistent with the previous run.55  lh_timeout_seconds:               # If a float is input, the log_likelihood_function call is timed out after this many seconds, to diagnose infinite loops. Default is None, meaning no timeout.56  disable_positions_lh_inversion_check: true57version:58  python_version_check: True        # If False, bypass the PyAutoNerves Python 3.12+ recommendation (3.9/3.10/3.11 technically work but are not officially supported).59  # The compatibility FLOOR: the oldest library release whose API this60  # workspace's scripts require. Preferred over workspace_version61  # (autonerves/workspace.py). Bump DELIBERATELY — only when a script62  # starts needing new API — never per release. Must always name an63  # INSTALLABLE (non-yanked) release.64  minimum_library_version: 2026.7.9.165  workspace_version_check: True     # If False, bypass the workspace/library version check. Set to False on `main`-branch clones — `main` updates faster than releases, so mismatches are expected and not actionable.

1# Workspace overrides for the library lensing latent toggles. The2# PyAutoLens library defaults the three raw-flux keys to `true` (they3# need no instrument inputs) and the µJy / dimensionless variants to4# `false`. Enabling the µJy ones here means a workspace fit produces5# real microjansky output as long as the user also passes `magzero` to6# `al.AnalysisImaging(...)` — without `magzero` they return NaN and emit7# one warning per process.8#9# Run `scripts/guides/results/latent_variables.py` for a tutorial on10# what each key means and `scripts/guides/units/flux.py` for how to11# convert a raw-flux latent to microjanskies in post.1213# total_lens_flux — integrated lens-galaxy flux in the fit's raw image14# units. No instrument inputs required. NaN when lens has no light profile.15total_lens_flux: true1617# total_lensed_source_flux — image-plane source flux after lensing, in18# raw image units. No instrument inputs required.19total_lensed_source_flux: true2021# total_source_flux — source-plane intrinsic source flux in raw image22# units. Uses fit.tracer_linear_light_profiles_to_light_profiles so MGE /23# linear light profiles work correctly. No instrument inputs required.24total_source_flux: true2526# total_lens_flux_mujy — same as total_lens_flux but converted to27# microjanskies. Requires magzero on AnalysisImaging; NaN + one warning28# per process if missing.29total_lens_flux_mujy: true3031# total_lensed_source_flux_mujy — image-plane source flux after lensing,32# in microjanskies. Requires magzero.33total_lensed_source_flux_mujy: true3435# total_source_flux_mujy — source-plane intrinsic source flux in microjanskies.36# Uses fit.tracer_linear_light_profiles_to_light_profiles so MGE / linear37# light profiles work correctly. Requires magzero.38total_source_flux_mujy: true3940# magnification — dimensionless ratio of image-plane to source-plane flux.41# magzero is not required.42magnification: true4344# effective_einstein_radius — Einstein radius in arcseconds via the45# zero-contour of the tangential eigenvalue field.46# magzero is not required.47effective_einstein_radius: true4849# autogalaxy library default (loaded by autonerves into the same `latent`50# conf node) — explicitly disabled here to silence the cross-library51# "unknown latent" warning that would otherwise fire on every fit.52total_galaxy_0_flux: false

1version: 12disable_existing_loggers: false34handlers:5  console:6    class: logging.StreamHandler7    level: INFO8    stream: ext://sys.stdout9    formatter: formatter1011root:12  level: INFO13  handlers: [ console ]1415formatters:16  formatter:17    format: '%(asctime)s - %(name)s - %(levelname)s - %(message)s'

1# The notation configs define the labels of every model parameter and its derived quantities, which are used when2# visualizing results (for example labeling the axis of the PDF triangle plots output by a non-linear search).345# label: The label given to the each parameter, for plots like PDF corner plots.67# For example, if `centre=x`, the plot axis will be labeled 'x'.8910# superscript: the superscript used on certain plots that show the results of different model-components.1112# For example, if `Gaussian=g`, plots where the parameters of the Gaussian model-component have superscript `g`.1314label:15  label:16    sigma: \sigma17    alpha: \alpha18    angle_binary: \theta19    beta: \beta20    break_radius: \theta_{\rm B}21    centre_0: y22    centre_1: x23    coefficient: \lambda24    contribution_factor: \omega_{\rm 0}25    core_radius: C_{\rm r}26    core_radius_0: C_{rm r0}27    core_radius_1: C_{\rm r1}28    effective_radius: R_{\rm eff}29    einstein_radius: \theta_{\rm Ein}30    ell_comps_0: \epsilon_{\rm 1}31    ell_comps_1: \epsilon_{\rm 2}32    multipole_comps_0: M_{\rm 1}33    multipole_comps_1: M_{\rm 2}34    flux: F35    gamma: \gamma36    gamma_1: \gamma37    gamma_2: \gamma38    inner_coefficient: \lambda_{\rm 1}39    inner_slope: t_{\rm 1}40    intensity: I_{\rm b}41    kappa: \kappa42    kappa_s: \kappa_{\rm s}43    log10m_vir: log_{\rm 10}(m_{vir})44    m: m45    mass: M46    mass_at_200: M_{\rm 200}47    mass_ratio: M_{\rm ratio}48    mass_to_light_gradient: \Gamma49    mass_to_light_ratio: \Psi50    mass_to_light_ratio_base: \Psi_{\rm base}51    mass_to_light_radius: R_{\rm ref}52    noise_factor: \omega_{\rm 1}53    noise_power: \omega{\rm 2}54    noise_scale: \sigma_{\rm 1}55    normalization_scale: n56    outer_coefficient: \lambda_{\rm 2}57    outer_slope: t_{\rm 2}58    overdens: \Delta_{\rm vir}59    pixels: N_{\rm pix}60    radius_break: R_{\rm b}61    redshift: z62    redshift_object: z_{\rm obj}63    redshift_source: z_{\rm src}64    rs: r_{\rm s}65    ra: r_{\rm a}66    scale_radius: R_{\rm s}67    scatter: \sigma68    sigma_scale: \sigma69    separation: s70    sersic_index: n71    shape_0: y_{\rm pix}72    shape_1: x_{\rm pix}73    signal_scale: V74    sky_scale: \sigma_{\rm 0}75    slope: \gamma76    truncation_radius: R_{\rm t}77    weight_floor: W_{\rm f}78    weight_power: W_{\rm p}79  superscript:80    ExternalShear: ext81    Mesh: mesh82    Point: point83    SMBH: smbh84    Redshift: z85    Regularization: reg86    HyperBackgroundNoise: hyper87    HyperGalaxy: hyper88    HyperImageSky: hyper89    InputDeflections: defl9091# label_format: The format certain parameters are output as in output files like the `model.results` file.9293# For example, if `centre={:.2d}`, the format of the centre parameter in results files will use this Python format.9495label_format:96  format:97    sigma: '{:.4f}'98    alpha: '{:.4f}'99    angle_binary: '{:.4f}'100    angular_diameter_distance_to_earth: '{:.4f}'101    beta: '{:.4f}'102    c_2: '{:.4f}'103    centre_0: '{:.4f}'104    centre_1: '{:.4f}'105    coefficient: '{:.4f}'106    concentration: '{:.4f}'107    contribution_factor: '{:.4f}'108    core_radius: '{:.4f}'109    core_radius_0: '{:.4f}'110    core_radius_1: '{:.4f}'111    effective_radius: '{:.4f}'112    einstein_mass: '{:.4e}'113    einstein_radius: '{:.4f}'114    ell_comps_0: '{:.4f}'115    ell_comps_1: '{:.4f}'116    multipole_comps_0: '{:.4f}'117    multipole_comps_1: '{:.4f}'118    flux: '{:.4e}'119    gamma: '{:.4f}'120    inner_coefficient: '{:.4f}'121    inner_slope: '{:.4f}'122    intensity: '{:.4f}'123    kappa: '{:.4f}'124    kappa_s: '{:.4f}'125    kpc_per_arcsec: '{:.4f}'126    log10m_vir: '{:.4f}'127    luminosity: '{:.4e}'128    m: '{:.1f}'129    mass: '{:.4e}'130    mass_at_200: '{:.4e}'131    mass_at_truncation_radius: '{:.4e}'132    mass_ratio: '{:.4f}'133    mass_to_light_gradient: '{:.4f}'134    mass_to_light_ratio: '{:.4f}'135    n_x: '{:.1d}'136    n_y: '{:.1d}'137    noise_factor: '{:.3f}'138    noise_power: '{:.3f}'139    noise_scale: '{:.3f}'140    normalization_scale: '{:.4f}'141    outer_coefficient: '{:.4f}'142    outer_slope: '{:.4f}'143    overdens: '{:.4f}'144    pixels: '{:.4f}'145    ra : '{:.4f}'146    radius: '{:.4f}'147    radius_break: '{:.4f}'148    redshift: '{:.4f}'149    redshift_object: '{:.4f}'150    redshift_source: '{:.4f}'151    rho: '{:.4f}'152    rs: '{:.4f}'153    scale_radius: '{:.4f}'154    separation: '{:.4f}'155    sersic_index: '{:.4f}'156    shape_0: '{:.4f}'157    shape_1: '{:.4f}'158    signal_scale: '{:.4f}'159    sky_scale: '{:.4f}'160    slope: '{:.4f}'161    truncation_radius: '{:.4f}'162    weight_floor: '{:.4f}'163    weight_power: '{:.4f}'

1# Determines whether files saved by the search are output to the hard-disk. This is true both when saving to the2# directory structure and when saving to database.34default: true # If true then files which are not explicitly listed here are output anyway. If false then they are not.56### Samples ###78# The `samples.csv`file contains every sampled value of every free parameter with its log likelihood and weight.910# This file is often large, therefore disabling it can significantly reduce hard-disk space use.1112# `samples.csv` is used to perform marginalization, infer model parameter errors and do other analysis of the search13# chains. Even if output of `samples.csv` is disabled, these tasks are still performed by the fit and output to14# the `samples_summary.json` file. However, without a `samples.csv` file these types of tasks cannot be performed15# after the fit is complete, for example via the database.1617samples: true1819# The `samples.csv` file contains every accepted sampled value of every free parameter with its log likelihood and20# weight. For certain searches, the majority of samples have a very low weight and have no numerical impact on the21# results of the model-fit. However, these samples are still output to the `samples.csv` file, taking up hard-disk22# space and slowing down analysis of the samples (e.g. via the database).2324# The `samples_weight_threshold` below specifies the threshold value of the weight such that samples with a weight25# below this value are not output to the `samples.csv` file. This can be used to reduce the size of the `samples.csv`26# file and speed up analysis of the samples.2728# For many searches (e.g. MCMC) all samples have an equal weight of 1.0, and this threshold therefore has no impact.29# For these searches, there is no simple way to save hard-disk space. This input is more suited to nested sampling,30# where the majority of samples have a very low weight..3132# Set value to empty (e.g. delete 1.0e-10 below) to disable this feature.3334samples_weight_threshold: 1.0e-103536### Search Internal ###3738# The search internal folder which contains a saved state of the non-linear search in its internal reprsenetation,39# as a .pickle or .dill file.4041# For example, for the nested sampling dynesty, this .dill file is the `DynestySampler` object which is used to42# perform sampling, and it therefore contains all internal dynesty representations of the results, samples, weights, etc.4344# If the entry below is false, the folder is still output during the model-fit, as it is required to resume the fit45# from where it left off. Therefore, settings `false` below does not impact model-fitting checkpointing and resumption.46# Instead, the search internal folder is deleted once the fit is completed.4748# The search internal folder file is often large, therefore deleting it after a fit is complete can significantly49# reduce hard-disk space use.5051# The search internal representation that can be loaded from the .dill file has many additional quantities specific to52# the non-linear search that the standardized autofit forms do not. For example, for emcee, it contains information on53# every walker. This information is required to do certain analyes and make certain plots, therefore deleting the54# folder means this information is list.5556search_internal: false5758### Start Point ###5960# If an Initalizer is used to provide a start point for the non-linear search, visualization of that start point can be61# output to hard-disk to show the user the initial model-fit that is used to start the search. This visualization is62# the visualizer wrapped in the Analysis class, and therefore should show things like the quality of the fit63# to the data and the residuals at the start point.6465start_point: true6667### Latent Variables ###6869# A latent variable is not a model parameter but can be derived from the model. Its value and errors may be of interest70# and aid in the interpretation of a model-fit.7172# For example, for the simple 1D Gaussian example, it could be the full-width half maximum (FWHM) of the Gaussian. This73# is not included in the model but can be easily derived from the Gaussian's sigma value.7475# By overwriting an Analysis class's `compute_latent_variables` method we can manually specify latent variables that76# are calculated and output to a `latent.csv` file, which mirrors the `samples.csv` file. The `latent.csv` file has77# the same weight resampling performed on the `samples.csv` file, controlled via the `samples_weight_threshold` above.7879# There may also be a `latent.results` and `latent_summary.json` files output, which the inputs below control whether80# they are output and how often.8182# Outputting latent variables manually after a fit is complete is simple, just call83# the `analysis.compute_latent_variables()` function.8485# For many use cases, the best set up may be to disable autofit latent variable output during the fit and perform it86# manually after completing a successful model-fit. This will save computational run time by not computing latent87# variables during a any model-fit which is unsuccessful.8889latent_during_fit: false # Whether to output the `latent.csv`, `latent.results` and `latent_summary.json` files during the fit when it performs on-the-fly output.90latent_after_fit: true # If `latent_during_fit` is False, whether to output the `latent.csv`, `latent.results` and `latent_summary.json` files after the fit is complete.91latent_draw_via_pdf : true # Whether to draw latent variable values via the PDF of every sample, which uses fewer samples to estimate latent variable errors. If False, latent variable values are drawn from every sample.92latent_draw_via_pdf_size : 100 # The number of samples drawn to estimate latent variable errors if `latent_draw_via_pdf` is True.93latent_csv: true # Whether to ouptut the `latent.csv` file.94latent_results: true # Whether to output the `latent.results` file.9596# Other Files:9798search_log: true # `search.log`: logging produced whilst running the fit method99100model_graph: false # `model.graph`: graphical-model factor graph summary. Only meaningful for graphical models (autofit.graphical); empty for ordinary PriorModel/Collection fits, so disabled by default.101102# `model.png`: the model figure drawn beside `model.info` (af.ModelPlotter). Opt-in until the103# model-figures epic's phase-3 acceptance renders pass. Unlike other keys an absent key means off.104model_figure: false

1# Curated examples rendered to executed markdown pages (markdown/) with their2# real output images, so they can be read on GitHub. Built manually by3# PyAutoHands's generate_markdown.py — see that module's docstring for the4# rules (never TEST_MODE; features/ scripts never rendered; list order is5# execution order and index order).6- script: start_here.py7  max_minutes: 458- script: scripts/imaging/start_here.py9  max_minutes: 24010- script: scripts/imaging/simulator.py11  max_minutes: 4512- script: scripts/imaging/fit.py13  max_minutes: 4514- script: scripts/imaging/likelihood_function.py15  max_minutes: 4516- script: scripts/imaging/modeling.py17  max_minutes: 24018- script: scripts/guides/tracer.py19  max_minutes: 4520- script: scripts/guides/galaxies.py21  max_minutes: 4522- script: scripts/guides/lens_calc.py23  max_minutes: 4524# --- batch 2a: remaining dataset types (cluster excluded pending runtime call) ---25- script: scripts/interferometer/start_here.py26  max_minutes: 24027- script: scripts/interferometer/simulator.py28  max_minutes: 4529- script: scripts/interferometer/likelihood_function.py30  max_minutes: 4531- script: scripts/interferometer/fit.py32  max_minutes: 4533- script: scripts/interferometer/modeling.py34  max_minutes: 30035- script: scripts/point_source/start_here.py36  max_minutes: 12037- script: scripts/point_source/simulator.py38  max_minutes: 4539- script: scripts/point_source/fit.py40  max_minutes: 4541- script: scripts/point_source/modeling.py42  max_minutes: 12043- script: scripts/multi_dataset/start_here.py44  max_minutes: 12045- script: scripts/multi_dataset/simulator.py46  max_minutes: 4547- script: scripts/multi_dataset/modeling.py48  max_minutes: 24049- script: scripts/group/start_here.py50  max_minutes: 12051- script: scripts/group/simulator.py52  max_minutes: 4553- script: scripts/group/likelihood_function.py54  max_minutes: 4555- script: scripts/group/fit.py56  max_minutes: 4557- script: scripts/group/modeling.py58  max_minutes: 30059- script: scripts/weak/simulator.py60  max_minutes: 4561- script: scripts/weak/likelihood_function.py62  max_minutes: 4563- script: scripts/weak/fit.py64  max_minutes: 4565- script: scripts/weak/modeling.py66  max_minutes: 240

1# Scripts to skip during automated runs (smoke tests, pre-release checks, CI).2# Each entry is matched against script paths:3#   - Entries with '/' do a substring match against the file path4#   - Entries without '/' match the file stem exactly5# Add an inline # comment to document the reason for skipping.6#7# THIS LIST IS REPO-LOCAL. Do not copy entries in from HowToLens (or any other8# repo), and do not copy entries out. Tutorial stems such as `tutorial_searches`9# belong to HowToLens and can never match a file here; a pattern that matches10# nothing is silently inert, not a skip. Every entry below must match a file in11# THIS repo — verify with `should_skip` in PyAutoHands/autohands/build_util.py12# before adding one, and re-check the pattern whenever a script is moved.13#14# SLOW-skip convention:15#   Entries tagged `# SLOW <YYYY-MM-DD> - <reason>` mark scripts that are16#   skipped because they exceed the per-script timeout cap (300s by17#   default; 1800s for mode=release runs). These are18#   NOT permanent skips — every mega-run surfaces them with a loud warning19#   banner. Fix the performance issue and remove the SLOW marker.20#21# NEEDS_FIX convention:22#   Entries tagged `# NEEDS_FIX <YYYY-MM-DD> - <reason>` mark scripts that23#   are broken and parked as a to-do list. Like SLOW-skips, these are NOT24#   permanent skips — every mega-run surfaces them with a loud warning25#   banner. Investigate the failure, fix the underlying bug, and remove26#   the NEEDS_FIX marker.2728- gui/extra_galaxies_centres # GUI scripts cannot be run29- gui/mask_extra_galaxies # GUI scripts cannot be run30- gui/lens_light_centre # GUI scripts cannot be run31- gui/mask # GUI scripts cannot be run32- gui/positions # GUI scripts cannot be run33- fits_make # Test mode does not output .fits images.34- png_make # Test mode does not output .png images.35- time_delays # Test mode does not support cosmology ift36- guides/plot/searches # Test mode breaks search visualization.37- mass_stellar_dark/modeling # Requires CSE to be JAX enabled.38- mass_stellar_dark/slam # Requires CSE to be JAX enabled.39- detect/database # Unsure but not a feature actively used currently.40- imaging/features/advanced/subhalo/sensitivity/ # All sensitivity scripts need updating when visualization refactored.41- point_source/features/multiple_sources/simulator # Blocked by PyAutoLens #480: solver finds 0 positions for intermediate-plane source42- point_source/features/multiple_sources/modeling # Blocked by PyAutoLens #480: same root cause as simulator above43- interferometer/casa_reduction # Requires CASA MeasurementSet output, not runnable standalone44- cluster/start_here # SLOW 2026-07-22 - hits the full 1800s mode=release cap in workspace-validation (PyAutoHeart run 29912642195). Script-specific, not a shard-wide problem: every other cluster script passes, the next-slowest being lenstool/modeling.py at 137.8s. Consistent with cluster scripts being known un-smoke-able (>500s even in TEST_MODE). Not yet profiled — SLOW-skipped to unblock the release; remove once the cost is found and fixed (autolens_workspace#314).45- multi_galaxy/start_here # SLOW 2026-09-18 - blocked two consecutive release runs after autolens_workspace#554 (2026-09-17) swapped it off the simulated `simple` dataset onto the real SDSS J1011+0143 ACS/WFC F814W frame. On the profile defaults it hit the 1800s mode=release cap (PyAutoHeart run 35319361459, "1 timeout"); with a per-script BUILD_SCRIPT_TIMEOUT of 3600s AND PYAUTO_SMALL_DATASETS lifted (#563, reverted here) the runner died at ~1478s on a shutdown signal (exit 143, PyAutoHeart run 35372831809) before either cap could fire — consistent with the uncapped real frame exhausting runner memory, though that is inferred, not measured. Not yet profiled - SLOW-skipped to unblock the release; remove once the cost is found and fixed (PyAutoMind draft/bug/autolens_workspace/multi_galaxy_start_here_release_cost.md).46- weak/features/strong_lensing/a2744 # SLOW 2026-07-22 - hits the full 1800s mode=release cap in workspace-validation (PyAutoHeart run 29912642195). Script-specific: its sibling weak/real_data/a2744.py passes in 8.9s and the next-slowest weak script is modeling.py at 23.6s, so the cost is in the strong_lensing feature path rather than the A2744 dataset itself. Not yet profiled — SLOW-skipped to unblock the release (autolens_workspace#314).47- guides/modeling/advanced/expectation_propagation.py # NEEDS_FIX 2026-08-03 - EP parked as not release-ready. EP message projection is unstable: a truncated per-factor search projects an ESS=1 posterior with zero weighted variance, which EP feeds back as a delta-function prior. See PyAutoFit #1332 F10 and autofit_workspace_test graphical/ep.py.

1# Per-script environment variable configuration for the RELEASE-FIDELITY2# validation run (Heart's workspace-validation.yml, mode=release — the M33# wheel-based release-fidelity path). Distinct from profile_smoke.yaml, which is the4# `smoke` profile used by the per-PR CI gate.5#6# The `release` profile trades speed for fidelity: it is run once per release7# rehearsal against the TestPyPI wheels, not on every PR, so it can afford a8# reduced (not bypassed) sampler and real fit output/visualization/checks.9# Spec + acceptance table: PyAutoHeart/docs/release_validation.md.10#11# "defaults" are applied to every script on top of the inherited environment.12# Every var this profile cares about is given an EXPLICIT value in "defaults"13# (not left absent) — the runner only ever *sets* keys it's given, it never14# clears unrelated inherited env vars, so an absent key silently falls through15# to whatever the calling process already had (a leftover smoke-mode "1" from16# an earlier step, a developer's local shell, ...). Pinning everything here17# makes the profile self-contained regardless of the caller's environment.18#19# "overrides" should normally `set:` a var away from this profile's own default.20# Use `unset:` only when the script genuinely needs the variable absent, such as21# guides that must not run under PyAuto test mode at all.22#23# Pattern convention (same as no_run.yaml):24#   - Patterns containing '/' do a substring match against the file path25#   - Patterns without '/' match the file stem exactly2627defaults:28  PYAUTO_TEST_MODE: "1"                     # reduced iterations (real sampler), not bypassed29  PYAUTO_SKIP_FIT_OUTPUT: "0"               # real fit output (release fidelity, not smoke)30  PYAUTO_SKIP_VISUALIZATION: "0"            # real visualization (release fidelity, not smoke)31  PYAUTO_SKIP_CHECKS: "0"                   # real mesh/position/weight checks (release fidelity)32  PYAUTO_SMALL_DATASETS: "1"                # cap grids/masks to 15x15, reduce MGE gaussians33  PYAUTO_DISABLE_JAX: "0"                   # JAX enabled (release fidelity, not smoke)34  PYAUTO_FAST_PLOTS: "1"                    # skip tight_layout() + critical curve/caustic overlays35  PYAUTO_SKIP_WORKSPACE_VERSION_CHECK: "1"  # TestPyPI dev version won't match the workspace pin36  JAX_ENABLE_X64: "True"                    # enable 64-bit precision in JAX37  NUMBA_CACHE_DIR: "/tmp/numba_cache"       # writable cache dir for numba38  MPLCONFIGDIR: "/tmp/matplotlib"           # writable config dir for matplotlib3940overrides:41  # start_here scripts load real FITS data that breaks with small datasets42  #43  # imaging/start_here.py additionally gets its own per-script BUILD_SCRIPT_TIMEOUT44  # (autolens_workspace#547). It was killed at the release leg's run-wide 1800s cap45  # in 3 of the last 8 Release Integrate runs (2026-09-09, 09-12, 09-14 run46  # 34898325503) at 1805s, while passing runs of the same script span 586-1695s and47  # every sibling in the leg is stable. Each kill lands inside one ~26 min XLA CPU48  # compile of MultiStartProdigy (n_starts=48, batch_size=None) under this script's49  # non-uniform over-sample map, BEFORE the first gradient step - a compile-time50  # flake, not a regression. 3600s buys the compile room; worst case it adds +30 min51  # to the single slowest leg of a ~72 min job, and only when the flake actually52  # fires. The run-wide 1800s cap is deliberately left alone: it is the only guard on53  # the other 84 entries in this profile.54  #55  # Spelled with `set:` because validate_env_profiles.ALLOWED_OVERRIDE_KEYS is56  # {pattern, set, unset} - a `timeout:` key would be rejected. build_util.timeout_for57  # reads it parent-side and it WINS over the workflow global (PyAutoHands#227);58  # precedent: autolens_workspace_test/config/build/profile_smoke.yaml "jax_grad/".59  #60  # The real fix - an explicit batch_size or a uniform over-sample map so the compile61  # is cheap - is tracked separately; retire this entry once that lands.62  - pattern: "imaging/start_here"63    set:64      PYAUTO_SMALL_DATASETS: "0"65      BUILD_SCRIPT_TIMEOUT: "3600"66  - pattern: "interferometer/start_here"67    set: { PYAUTO_SMALL_DATASETS: "0" }68  - pattern: "group/start_here"69    set: { PYAUTO_SMALL_DATASETS: "0" }70  - pattern: "multi_dataset/start_here"71    set: { PYAUTO_SMALL_DATASETS: "0" }72  # Potential correction builds a dpsi mesh by subsampling the image grid, so the73  # 15x15 small-datasets cap starves it: al.pc.PairRegularDpsiMesh(dpsi_factor=2)74  # raises "The dpsi grid is too sparse" from mesh.py's get_itp_box_ctr. Both the75  # interferometer and imaging siblings under features/advanced/potential_correction/ need76  # the cap lifted; the "*/start_here" patterns above do NOT cover them — their77  # paths are {interferometer,imaging}/features/advanced/potential_correction/..., which78  # that substring misses. Uncapped they are cheap: start_here.py runs in ~58s79  # against the 1800s cap.80  - pattern: "interferometer/features/advanced/potential_correction/"81    set: { PYAUTO_SMALL_DATASETS: "0" }82  - pattern: "imaging/features/advanced/potential_correction/"83    set: { PYAUTO_SMALL_DATASETS: "0" }84  # Non-simulator scripts that load committed FITS data need full-size85  # datasets to avoid shape mismatch with pre-existing 100x100 data.86  # Simulators run first and regenerate data at the capped size, but87  # scripts using the old `if not dataset_path.exists()` pattern skip88  # re-simulation when data already exists.89  - pattern: "guides/"90    set: { PYAUTO_SMALL_DATASETS: "0" }91  # (guides/results/ formerly unset PYAUTO_TEST_MODE here; that intent now92  # lives in-file as '# ENV: real_search' declarations on the scripts, which93  # apply in every profile — the override became redundant.)94  #95  # fits_make / png_make produce .fits / .png outputs from real fits, so they96  # need PYAUTO_FAST_PLOTS forced off (it would otherwise close every figure97  # without saving via the subplot_save / save_figure short-circuit in98  # autoarray/plot/utils.py). PYAUTO_SKIP_VISUALIZATION is already "0" above.99  - pattern: fits_make100    set: { PYAUTO_FAST_PLOTS: "0" }101  - pattern: png_make102    set: { PYAUTO_FAST_PLOTS: "0" }103  # The datacube Delaunay fit runs a reduced-but-real sampler over a 4-channel104  # Delaunay + per-channel NUFFT FactorGraph. Its per-likelihood cost is far105  # higher than the rectangular sibling (interferometer/features/datacube/106  # modeling.py, which passes under the reduced sampler), so the reduced-sampler107  # run overruns the 1800s per-script cap on CI CPU. Bypass the sampler here so108  # it still validates the full Delaunay pipeline (compose + one likelihood109  # eval) within budget; the reduced-sampler datacube path stays covered by the110  # rectangular sibling.111  - pattern: "interferometer/features/datacube/delaunay"112    set: { PYAUTO_TEST_MODE: "2" }

1# Per-script environment variable configuration for automated runs2# (smoke tests, pre-release checks, CI).3#4# "defaults" are applied to every script on top of the inherited environment.5# "overrides" selectively unset or replace vars for matching path patterns.6#7# Pattern convention (same as no_run.yaml):8#   - Patterns containing '/' do a substring match against the file path9#   - Patterns without '/' match the file stem exactly1011defaults:12  PYAUTO_TEST_MODE: "2"                     # 0=normal, 1=reduced iterations, 2=skip sampler (fastest)13  PYAUTO_SKIP_FIT_OUTPUT: "1"               # Skip pre/post-fit I/O, VRAM profiling, result text14  PYAUTO_SKIP_VISUALIZATION: "1"            # Skip fit visualization and plotting15  PYAUTO_SKIP_CHECKS: "1"                   # Skip mesh validation, position checks, weight thresholds16  PYAUTO_SMALL_DATASETS: "1"                # Cap grids/masks to 15x15, reduce MGE gaussians17  PYAUTO_DISABLE_JAX: "1"                   # Force use_jax=False, avoid JIT compilation overhead18  PYAUTO_FAST_PLOTS: "1"                    # Skip tight_layout() + critical curve/caustic overlays19  JAX_ENABLE_X64: "True"                    # Enable 64-bit precision in JAX20  NUMBA_CACHE_DIR: "/tmp/numba_cache"       # Writable cache dir for numba21  MPLCONFIGDIR: "/tmp/matplotlib"           # Writable config dir for matplotlib2223overrides:24  # The */start_here and guides/ PYAUTO_SMALL_DATASETS unsets migrated to25  # in-file `ENV: full_datasets` declarations on each matched script, and26  # the PYAUTO_TEST_MODE / PYAUTO_FAST_PLOTS unsets below to `ENV:27  # real_search` / `ENV: real_plots` — #187 Stage 2. A declaration is an28  # `__Env__` docstring SECTION holding a bare `ENV: <tokens>` line; the older29  # `# ENV:` comment form was removed and now RAISES (PyAutoHands#189/#190).30  # The non-declarable PYAUTO_SKIP_* vars remain here.31  # guides/results/start_here.py must produce real samples so the example32  # scripts that read from `output/results_folder` afterwards33  # (data_fitting, queries, models, samples_via_aggregator) find a34  # non-empty aggregator. Covers all scripts under guides/results/.35  # (PYAUTO_TEST_MODE moved to `ENV: real_search`.)36  - pattern: "guides/results/"37    unset: [PYAUTO_SKIP_FIT_OUTPUT]38  # fits_make / png_make produce .fits / .png outputs from real fits, so they need39  # visualization turned on. (PYAUTO_FAST_PLOTS moved to `ENV: real_plots`.)40  - pattern: fits_make41    unset: [PYAUTO_SKIP_VISUALIZATION]42  - pattern: png_make43    unset: [PYAUTO_SKIP_VISUALIZATION]

1# Notebook stems that should run when PyAutoHands's generate / run pipeline2# is invoked with --visualise. Used to refresh notebook output cells in main.3#4# Format: flat list of notebook stems (no extension, no path).5# An empty list means no notebooks need re-visualisation in this workspace.6#7# This file overrides PyAutoHands/autohands/config/visualise_notebooks.yaml8# for this workspace. Add or remove entries here, not there.910- start_here

1# The settings of a parallelized grid search of non-linear searches.23parallel:4  number_of_cores: 3 # The number of cores the search is parallelized over by default, using Python multiprocessing.5  step_size: 0.1     # The default step size of each grid search parameter, in terms of unit values of the priors.

1Basis: {}