17.5. Pipeline Presets

A pipeline preset is a named, built-in processing recipe. It gives a survey a tested starting order without forcing the user to write every pipeline step code by hand. A preset is not a hidden processing mode: inside pyCSAMT it is an ordered list of (label, Step) tuples. That list can be inspected, expanded, exported to a pipeline configuration file, edited, reviewed, and archived like any other processing pipeline.

Use presets for quick first-pass QC, standard comparisons between processing strategies, teaching safe default step order, and generating starter configs. For serious survey processing, treat the preset as the beginning of the conversation, not the final scientific decision. The reproducible record is the expanded step sequence written to pipeline.yaml or to a committed configuration file.

17.5.1. Preset Mental Model

Each built-in preset is represented by a small object with three fields:

name

Stable identifier used by the CLI and Python API, for example basic_qc.

description

Short explanation shown in preset catalogues.

steps

Ordered list of (label, Step) tuples. The label becomes the step label used in reports and plot directories. The Step object holds the registry code and the parameters passed to the transform.

Mathematically, a preset defines a fixed composition of transforms:

\[P(S_0) = T_n(\cdots T_2(T_1(S_0;\theta_1);\theta_2)\cdots;\theta_n),\]

where \(S_0\) is the input site collection, \(T_j\) is the registered operation at position \(j\), and \(\theta_j\) is the resolved parameter dictionary for that step. The order is part of the meaning. Running FREQ001 before NR001 is not the same workflow as running NR001 before FREQ001 because each step receives a different intermediate survey state.

The preset API is intentionally small:

>>> from pycsamt.pipeline import get_preset, list_presets, preset_catalogue
>>> preset = get_preset("basic_qc")
>>> preset.name
'basic_qc'
>>> [(label, step.spec.code) for label, step in preset.steps]
[('notch', 'NR001'), ('drop_duplicates', 'FREQ002'), ('select_band', 'FREQ001'), ('align_grid', 'FREQ004'), ('qc_snapshot', 'QC001')]
>>> [(p.name, len(p.steps)) for p in list_presets()]
[('basic_qc', 5), ('noise_reduction', 6), ('full_processing', 8), ('tensor_analysis', 5), ('dimensionality_filter', 4), ('publication_ready', 9), ('stratagem_mt', 7), ('mt_qc', 11), ('amt_qc', 11), ('csamt_qc', 14), ('csumt_qc', 15)]

The CLI exposes the same information:

pycsamt pipe presets
pycsamt pipe presets --format json
pycsamt pipe presets --expand basic_qc --format json

Captured expansion excerpt:

{
  "name": "basic_qc",
  "description": "Minimal denoising + frequency cleanup.  Good for quick-look inspection.",
  "n_steps": 5,
  "steps": [
    {
      "label": "notch",
      "code": "NR001",
      "name": "notch_powerline",
      "category": "noise_removal",
      "params": {"mains_hz": 50, "n_harm": 30, "tol_hz": 0.08}
    },
    {
      "label": "drop_duplicates",
      "code": "FREQ002",
      "name": "drop_duplicates",
      "category": "frequency",
      "params": {}
    }
  ]
}

The excerpt is deliberately partial. In practice, use the full expansion when reviewing or exporting a preset.

17.5.2. Run A Preset

From the command line:

pycsamt pipe run data/3edis \
    --preset basic_qc \
    --out results/basic_qc \
    -v

From Python:

>>> from pycsamt.pipeline import Pipeline
>>> pipe = Pipeline.from_preset("basic_qc")
>>> [(label, step.spec.code) for label, step in pipe]
[('notch', 'NR001'), ('drop_duplicates', 'FREQ002'), ('select_band', 'FREQ001'), ('align_grid', 'FREQ004'), ('qc_snapshot', 'QC001')]
>>> result = pipe.run(sites, outdir="results/basic_qc")

When output is enabled, the run writes results/basic_qc/pipeline.yaml. That file is the run-specific canonical pipeline snapshot; it records the step sequence that actually ran.

17.5.3. Built-In Preset Summary

The normal pipeline registry currently provides eleven presets: the seven processing-intent presets below, plus four method-aware presets (mt_qc, amt_qc, csamt_qc, csumt_qc) covered in Method-Aware Presets.

Preset	Steps	Best for	Sequence
basic_qc	5	First-pass inspection and quick survey sanity checks.	NR001, FREQ002, FREQ001, FREQ004, QC001
noise_reduction	6	High-EMI data where denoising is the main question.	NR001, NR004, NR005, NR003, NR010, QC001
full_processing	8	Standard end-to-end processing before interpretation.	NR001, FREQ002, FREQ001, FREQ004, SK001, TZ001, SS001, QC001
tensor_analysis	5	Tensor-focused cleanup after frequency and noise handling are already acceptable.	TZ001, TZ002, TZ003, TZ004, QC001
dimensionality_filter	4	Classifying dimensionality and keeping/projecting 2-D-compatible intervals.	DIM001, DIM002, DIM003, QC001
publication_ready	9	Longer reviewed workflow for polished reports and figures.	NR001, FREQ002, FREQ001, FREQ004, SS001, TZ001, TZ002, SK001, QC001
stratagem_mt	7	Stratagem AMT data already loaded as a Sites object.	SS001, FREQ001, FREQ002, NR001, NR004, NR010, QC001

17.5.4. Choosing A Preset

Start with the narrowest preset that answers the current question. A narrow preset is easier to diagnose because fewer transforms stand between \(S_0\) and the output state \(S_n\).

Situation	Start with	Why
You just received a survey and need a quick check.	basic_qc	It does only basic notch, frequency cleanup, alignment, and QC.
Harmonic and spatial noise dominate the data.	noise_reduction	It stacks targeted noise-removal steps before a QC snapshot.
You want a general processing run before inversion preparation.	full_processing	It combines denoising, frequency cleanup, skew gating, rotation, static-shift correction, and QC.
You already trust the frequency grid and want tensor diagnostics.	tensor_analysis	It avoids frequency and noise steps and focuses on tensor operations.
You are deciding which intervals are compatible with 2-D assumptions.	dimensionality_filter	It classifies dimensionality, masks by class, projects to 2-D, and generates QC.
You need a polished, repeatable processing chain.	publication_ready	It is longer and more opinionated, with static-shift correction, tensor cleanup, skew gating, and final QC.
You are working with Stratagem AMT data already represented as Sites.	stratagem_mt	It applies Stratagem-oriented AMT band selection, static shift, denoising, and QC at the emtools pipeline level.

The table above chooses by processing intent only. If you also know the EM survey method (MT, AMT, CSAMT, or CSUMT), start from Method-Aware Presets instead — its QC steps inspect the actual data before deciding what to plot, and CSAMT/CSUMT additionally get near-field correction that the presets above never apply.

17.5.5. Preset Details

17.5.5.1. basic_qc

basic_qc is the safest first preset for most surveys. It does not try to solve every processing problem; it prepares a clean enough view to understand what the data need next.

notch            NR001   notch_powerline
drop_duplicates  FREQ002 drop_duplicates
select_band      FREQ001 select_band
align_grid       FREQ004 align_grid
qc_snapshot      QC001   qc_snapshot

The default transform is

\[S_5 = T_{\mathrm{QC001}}( T_{\mathrm{FREQ004}}( T_{\mathrm{FREQ001}}( T_{\mathrm{FREQ002}}( T_{\mathrm{NR001}}(S_0))))).\]

Use basic_qc when you need quick figures before committing to a processing plan, want to verify that EDI loading and output generation work, or need the same minimal cleanup across several surveys. Move beyond it when static shift, skew gating, dimensionality filtering, or stronger denoising is clearly justified by the data.

17.5.5.2. noise_reduction

noise_reduction concentrates on denoising. It is useful when the first inspection shows power-line harmonics, local spikes, spatially coherent outliers, or incoherent frequency bins.

notch         NR001 notch_powerline
hampel        NR004 hampel_filter
spatial_med   NR005 spatial_median
shrink_trend  NR003 shrink_group_trend
mask_incoher  NR010 mask_incoherent
qc_snapshot   QC001 qc_snapshot

Run it next to basic_qc rather than replacing the first pass silently:

pycsamt pipe run data/3edis --preset basic_qc \
    --out results/compare/basic_qc
pycsamt pipe run data/3edis --preset noise_reduction \
    --out results/compare/noise_reduction

17.5.5.3. full_processing

full_processing is the standard end-to-end workflow. It starts with noise and frequency cleanup, then applies a skew gate, strike rotation, static-shift correction, and QC.

notch          NR001   notch_powerline
drop_dup       FREQ002 drop_duplicates
select_band    FREQ001 select_band
align_grid     FREQ004 align_grid
mask_skew      SK001   mask_by_skew
rotate_strike  TZ001   rotate_strike
correct_ss     SS001   correct_ss_ama
qc_snapshot    QC001   qc_snapshot

Use this preset when the survey needs a broad processing pass, when you want an auditable default before building an inversion-specific config, or when you need one chain that exercises the main processing families.

17.5.5.4. tensor_analysis

tensor_analysis assumes the data are already in reasonable condition and focuses on tensor operations.

rotate_strike  TZ001 rotate_strike
antisymm       TZ002 antisymmetrize
sigma_clip     TZ003 sigma_clip
balance        TZ004 balance_offdiag
qc_snapshot    QC001 qc_snapshot

Use it when you want to inspect tensor behavior without changing the frequency selection or applying the broader denoising chain.

17.5.5.5. dimensionality_filter

dimensionality_filter is for 1-D / 2-D / 3-D screening and 2-D projection workflows.

classify_dim  DIM001 classify_dim
mask_dim      DIM002 mask_by_dim
project_2d    DIM003 project_2d
qc_snapshot   QC001  qc_snapshot

Use it after basic cleanup when the main question is whether the remaining intervals are compatible with a 2-D interpretation or inversion assumption.

17.5.5.6. publication_ready

publication_ready is the longest built-in general-purpose preset. It is designed for polished processing output rather than quick exploration.

notch          NR001   notch_powerline
drop_dup       FREQ002 drop_duplicates
select_band    FREQ001 select_band
align_grid     FREQ004 align_grid
correct_ss     SS001   correct_ss_ama
rotate_strike  TZ001   rotate_strike
antisymm       TZ002   antisymmetrize
mask_skew      SK001   mask_by_skew
qc_snapshot    QC001   qc_snapshot

Run:

pycsamt pipe run data/3edis \
    --preset publication_ready \
    --out results/publication_ready \
    --dpi 300 \
    --plot-fmt pdf \
    -v

Use this preset after a lighter pass has shown that its assumptions are reasonable for the survey. The name does not make the output publication ready by itself; the review comes from inspecting the report, figures, processed EDIs, and saved pipeline snapshot.

17.5.5.7. stratagem_mt

stratagem_mt is a normal emtools pipeline preset specialized for Stratagem AMT data that are already loaded as a Sites object. It does not perform raw-coordinate injection, raw hardware-file parsing, or station renaming by itself.

correct_ss    SS001   correct_ss_ama
select_band   FREQ001 select_band   band_hz=(10.0, 100000.0)
drop_dup      FREQ002 drop_duplicates
notch         NR001   notch_powerline
hampel        NR004   hampel_filter
mask_incoher  NR010   mask_incoherent
qc_snapshot   QC001   qc_snapshot

Use it with the normal pipeline API when your input is already a site collection:

>>> from pycsamt.pipeline import Pipeline
>>> pipe = Pipeline.from_preset("stratagem_mt")
>>> result = pipe.run(sites, outdir="results/stratagem_mt")

Use pycsamt.pipeline.stratagem.StratagemPipeline or run_stratagem_preset when you also need the full raw EDI plus GPS CSV workflow.

17.5.6. Method-Aware Presets

The seven presets above are chosen by processing intent only — the same step sequence runs regardless of whether the survey is MT, AMT, CSAMT, or CSUMT. In reality these EM methods are not interchangeable: CSAMT/CSUMT use a controlled source and can suffer near-field/transition-zone contamination that MT/AMT never see; a single-component TE- or TM-only CSAMT line cannot produce a meaningful phase-tensor ellipse the way full-tensor MT/AMT data can; and tipper is only sometimes recorded at all. mt_qc, amt_qc, csamt_qc, and csumt_qc are chosen by EM method as well as intent, and their QC steps inspect the actual data before deciding what to plot instead of always firing the same figures.

Selection stays fully explicit — there is no automatic detection of survey method from the data. You (or a config file, or method if you already have one) choose mt_qc/amt_qc/csamt_qc/csumt_qc by name, exactly the same way you choose basic_qc vs. tensor_analysis today.

17.5.6.1. Smart QC And Preview Building Blocks

Three new diagnostic-only steps and two new preview steps back every method-aware preset. They are plain registry steps — nothing about them is special-cased to the four presets above, so any pipeline, existing or new, can add them too.

QC005 (tensor_qc_smart)

Plots the phase-tensor-ellipse pseudosection only when at least one station has both off-diagonal impedance components (Zxy and Zyx) populated. A single-component TE/TM-only CSAMT line skips this plot entirely rather than drawing a degenerate ellipse.

QC006 (tipper_qc_smart)

Plots the induction-vector map, induction section, response-tipper panels, tipper components, and tipper hodograms only when at least one station carries a real tipper channel. Most AMT surveys have none, so these five plots are silently skipped rather than drawn empty.

QC007 (strike_qc)

Always plots the strike rose and the combined strike/phase-tensor/ tipper-strike analysis figure. plot_strike_analysis is already internally tipper-aware — it draws a two-panel figure when no tipper is present and three panels when it is — so it needs no gating here.

PRE001/PRE002 (raw_preview/processed_preview)

PRE001 opens a method-aware preset and plots up to three randomly (but deterministically) chosen stations’ raw 1-D response; PRE002 closes the preset and plots the same station-selection logic against the processed result, so a before/after comparison is always available without hunting through the full station list. Both steps independently re-derive the station subset with the same default seed rather than sharing state, so if an intermediate step drops one of the previewed stations, the after-plot’s reselection (drawn from the smaller surviving set) can end up choosing a different station than the before-plot — a deliberate trade-off documented in pycsamt.pipeline._preview rather than solved with a cross-step state channel.

A QC step’s plot functions are always called as fn(sites) with no way to forward extra parameters (see Pipeline Steps), which is why every gating decision above is made inside the plot function from the data itself (pycsamt.pipeline._smart_qc), not from a step parameter.

17.5.6.2. Method-Aware Preset Summary

Preset	Steps	Best for	What differs from mt_qc
mt_qc	11	Broadband MT quick-look.	Baseline: denoise, frequency cleanup, strike rotation, smart QC, raw/processed preview.
amt_qc	11	AMT quick-look.	Trims select_band to the standard AMT band (10 Hz - 100 kHz).
csamt_qc	14	CSAMT quick-look.	Adds near-field/transition-zone correction (SRC001), its zone pseudosection (QC002), and source-response normalization (SRC002).
csumt_qc	15	CSUMT quick-look.	Everything csamt_qc adds, trimmed to the CSUMT acquisition band, plus a Bostick depth-section snapshot (QC004).

17.5.6.3. mt_qc

raw_preview        PRE001  raw_preview
notch              NR001   notch_powerline
drop_dup           FREQ002 drop_duplicates
select_band        FREQ001 select_band
align_grid         FREQ004 align_grid
rotate_strike      TZ001   rotate_strike
qc_snapshot        QC001   qc_snapshot
strike_qc          QC007   strike_qc
tensor_qc_smart    QC005   tensor_qc_smart
tipper_qc_smart    QC006   tipper_qc_smart
processed_preview  PRE002  processed_preview

select_band is left at its registry default (0.001 Hz - 10 kHz), which is already a broadband-MT-appropriate range — no method-specific override is applied.

17.5.6.4. amt_qc

Identical to mt_qc except select_band(FREQ001, band_hz=(10.0, 1e5)) — the exact AMT band already used by stratagem_mt, not a new number.

17.5.6.5. csamt_qc

raw_preview           PRE001  raw_preview
notch                 NR001   notch_powerline
drop_dup              FREQ002 drop_duplicates
select_band           FREQ001 select_band
align_grid            FREQ004 align_grid
correct_near_field    SRC001  correct_near_field  source_offset=None
field_zone_snapshot   QC002   field_zone_snapshot
normalize_response    SRC002  normalize_response
rotate_strike         TZ001   rotate_strike
qc_snapshot           QC001   qc_snapshot
strike_qc             QC007   strike_qc
tensor_qc_smart       QC005   tensor_qc_smart
tipper_qc_smart       QC006   tipper_qc_smart
processed_preview     PRE002  processed_preview

correct_near_field runs with source_offset=None — this is not “no correction,” it is “resolve per station from the site’s own source_offset/offset/dist metadata, and warn instead of failing when nothing resolves.” Against the real Tongkeng CSAMT example dataset (data/CSAMT, whose EDI headers carry no offset metadata), a real run warns once per station rather than raising:

UserWarning: correct_near_field: no source offset for 'csa000'; station skipped.
UserWarning: correct_near_field: no source offset for 'csa050'; station skipped.

and the impedance tensor for those stations passes through unchanged. When a station’s offset is resolvable, the correction is real — dividing \(Z_{\mathrm{obs}}\) by the complex near-field factor \(F(p) = 1 - 3/p^2 + 3/p^3\) — not a no-op; see pycsamt.emtools.source_effects.correct_near_field() for the full reference. No CSAMT-specific frequency band is applied — there is no authoritative CSAMT band constant in the codebase to draw from, so select_band stays at its registry default and the real method-differentiating behavior here is the near-field correction itself.

17.5.6.6. csumt_qc

Same chain as csamt_qc, plus:

• select_band(FREQ001, band_hz=(F_MIN_CSUMT, F_MAX_CSUMT)) — the real 9.6 kHz - 614.4 kHz CSUMT band constants from pycsamt.emtools.csumt, not invented numbers.

• depth_section_snapshot (QC004) — the Bostick depth-section snapshot, CSUMT-specific and already registered.

17.5.6.7. Selecting By Method String

get_preset_for_method() maps an explicit method string to the matching preset, so a caller that already knows (or has a SurveyMeta) doesn’t have to hand-build the preset name:

>>> from pycsamt.pipeline import get_preset_for_method
>>> get_preset_for_method("CSAMT").name
'csamt_qc'
>>> get_preset_for_method("amt").name
'amt_qc'

This performs no data inspection — method must be supplied explicitly, the same way get_preset() requires an explicit preset name. "MT", "BBMT", "LAMT", and "LMT" all map to mt_qc; "CSUMT" is recognised here even though it predates SurveyMeta’s own method vocabulary. "CSEM"/"TEM" are recognised as valid EM methods but have no method-aware preset yet (see below) and raise ValueError.

17.5.6.8. Not Yet Covered

Two things are deliberately out of scope for the current method-aware presets, rather than silently missing:

• A ``csem_qc`` preset. No CSEM-specific processing code exists anywhere in pycsamt.emtools today — fabricating CSEM corrections without real domain code behind them would be guessing at physics.

• Method-aware ``*_publication_ready`` tiers. The QC-level presets above are the first pass; fuller, publication-oriented method-aware chains are a natural follow-up once these are validated against more real surveys.

17.5.7. Export A Preset To A Config

Preset expansion is the safest transition from exploration to reproducible work. Generate the preset once, review the explicit list, then commit the config with the survey project.

pycsamt pipe init \
    --preset publication_ready \
    --name line22_publication_ready \
    --outdir results/line22_publication_ready \
    --output config/line22_publication_ready.yaml

Preview before running:

pycsamt pipe show config/line22_publication_ready.yaml
pycsamt pipe run data/3edis \
    --config config/line22_publication_ready.yaml \
    --dry-run

Generate Python or JSON instead:

pycsamt pipe init --preset basic_qc --format py \
    --output config/basic_qc.py
pycsamt pipe init --preset basic_qc --format json \
    --output config/basic_qc.json

17.5.8. Customizing Presets Safely

There are three supported customization patterns.

Edit a pipeline object in Python:

>>> from pycsamt.pipeline import Pipeline, Step
>>> pipe = Pipeline.from_preset("basic_qc")
>>> pipe.replace("notch", Step("NR001", mains_hz=60, n_harm=25)) is pipe
True
>>> pipe.append("static_shift", Step("SS001")) is pipe
True

Expand a preset into an explicit config and edit the generated parameters:

name: basic_qc_60hz
output_dir: results/basic_qc_60hz

steps:
  - name: notch
    code: NR001
    params:
      mains_hz: 60
      n_harm: 25
      tol_hz: 0.08
  - name: drop_duplicates
    code: FREQ002
  - name: select_band
    code: FREQ001
  - name: align_grid
    code: FREQ004
  - name: qc_snapshot
    code: QC001

Append extra steps after a preset in a config:

name: basic_qc_plus_static_shift
output_dir: results/basic_qc_plus_static_shift
preset: basic_qc

steps:
  - name: static_shift
    code: SS001
  - name: final_qc
    code: QC001

The third pattern appends. It does not edit the preset. In symbols, the loaded step list is

\[S_{\mathrm{final}} = S_{\mathrm{preset}} \Vert S_{\mathrm{explicit}},\]

where \Vert means append. If you need to change NR001 from 50 Hz to 60 Hz, use an explicit expanded step list instead of preset: basic_qc plus a second NR001 step.

17.5.9. CLI Priority

pycsamt pipe run resolves the pipeline definition in this order:

1. --config FILE;

2. --preset NAME;

3. --steps CODE,CODE,....

If --config is supplied, --preset and --steps are ignored because the config file is the source of truth.

pycsamt pipe run data/3edis \
    --config config/basic_qc.yaml \
    --preset publication_ready

pycsamt pipe run data/3edis \
    --preset publication_ready

pycsamt pipe run data/3edis \
    --steps FREQ002,FREQ001,FREQ004,NR001,QC001

17.5.10. Compare Presets

A preset comparison run keeps the input fixed and changes only the recipe. Let \(P_a\) and \(P_b\) be two presets. The comparison is meaningful only when both are applied to the same initial state:

\[S_n^{(a)} = P_a(S_0), \qquad S_m^{(b)} = P_b(S_0).\]

Run the branches into separate output roots:

pycsamt pipe run data/3edis \
    --preset basic_qc \
    --out results/compare/basic_qc

pycsamt pipe run data/3edis \
    --preset noise_reduction \
    --out results/compare/noise_reduction

pycsamt pipe run data/3edis \
    --preset full_processing \
    --out results/compare/full_processing

Compare summary.txt for step failures and site counts, report.html for per-step status and embedded YAML, plots/ for visual differences, processed/ for exported EDI differences, and pipeline.yaml for the exact recipe behind each branch.

17.5.11. Stratagem Presets

There are two related but different Stratagem preset systems.

stratagem_mt

A normal pycsamt.pipeline.Pipeline preset. It expects data that can already be processed as Sites and runs normal registered pipeline steps.

StratagemPreset

A convenience workflow in pycsamt.pipeline.stratagem for raw Stratagem EDI directories plus coordinate CSV files. It calls StratagemSurvey methods such as remove_static_shift, drop_frequencies, remove_noises, export, and rename.

The Stratagem convenience presets are:

Preset	Main workflow	Best for
basic	Coordinate injection, AMA static shift, frequency trim, noise removal, export, rename.	Direct replacement for the legacy Stratagem processing script.
full_processing	QC, AMA static shift, hardware-aware frequency filtering, smoothed noise removal, export, rename.	Raw Stratagem workflows with hardware files and a full QC pass.
publication_ready	Stricter QC, hardware SNR masking, AMT band trimming, stronger smoothing, export, rename.	Polished Stratagem outputs after the basic workflow has been reviewed.

Run a raw Stratagem convenience preset:

>>> from pycsamt.pipeline.stratagem import run_stratagem_preset
>>> survey = run_stratagem_preset(
...     "full_processing",
...     edi_dir="2/2EDI",
...     coord_file="2.csv",
...     raw_dir="raw/2HX",
...     outdir="results/stratagem",
...     epsg=32649,
...     utm_zone="49N",
...     rename_basename="T2.",
...     overwrite=True,
...     verbose=1,
... )

Build a Stratagem pipeline object from a normal emtools preset:

>>> from pycsamt.pipeline.stratagem import StratagemPipeline
>>> pipe = StratagemPipeline.from_preset(
...     "stratagem_mt",
...     coord_file="2.csv",
...     raw_dir="raw/2HX",
...     epsg=32649,
...     utm_zone="49N",
...     rename_basename="T2.",
... )
>>> result = pipe.run("2/2EDI", outdir="results/stratagem_mt")

17.5.12. Troubleshooting

Unknown preset

Run pycsamt pipe presets. Preset names are exact and lowercase, for example basic_qc or publication_ready.

I changed preset: basic_qc but the notch is still 50 Hz

A config preset expands the preset first. Explicit steps are appended; they do not edit existing preset steps. Generate an expanded config and edit the NR001 parameters directly.

The preset is too aggressive

Move to a narrower preset such as basic_qc or export the preset to a config and remove the steps that are not justified by the data.

The preset does not include a step I need

Append the step in Python, or add it to an explicit config after generating the preset scaffold.

I need raw Stratagem coordinate injection and renaming

Use pycsamt.pipeline.stratagem rather than the normal stratagem_mt preset alone.

17.5.13. Related Pages

• Pipeline Steps explains the registered step codes used by each preset.

• Pipeline Configuration Files explains how presets expand inside YAML, JSON, and Python configs.

• Pipeline CLI explains pycsamt pipe presets, pycsamt pipe show, and pycsamt pipe run.

• Pipeline Concepts explains pipeline execution and result objects.

• Pipeline Outputs explains generated reports, plots, processed EDIs, and saved pipeline snapshots.