Skip to content

Workflow map

Use this page as the shortest route from **what you recorded** to **what to do next**. Start from the measurement, establish the evidence needed for the next step, then follow the domain-specific example or article.
A workflow is complete only when its measurement assumptions, QC, timing, transformations, analysis settings, and reporting evidence remain traceable—not merely when it produces a final feature table.

New: follow one complete workflow

If you want to see the package used from beginning to end before choosing a domain, open the end-to-end EDA research workflow. It connects real public package calls to measurement checks, decision gates, derived tables, figures, exported artifacts and reporting evidence. The companion script is executed in documentation CI.

Choose by recorded data

You have Establish first Typical next steps Go to
I want one complete worked workflow package + data structure QC → process → inspect → export → report End-to-end EDA workflow
EDA / GSR units, activity, missingness, timing tonic/phasic decomposition, SCR detection, event summaries EDA / GSR / SCR
PPG waveform source + observed sampling + pulse quality peak detection, intervals, PRV/HRV-ready diagnostics PPG / HRV
IBI / RR / NN intervals source identity, units, cleaning status time/frequency/nonlinear variability, agreement checks PPG / HRV
Pupil diameter eye/channel, validity, blinks, timing baseline/change summaries, event windows, AOI linkage Pupil / gaze / AOI
Gaze coordinates / fixations coordinate space, validity, event source, timebase AOIs, transitions, saccades, linked physiology Pupil / gaze / AOI
TTL / event markers event identity, edge semantics, clock trial windows, event locking, cross-stream alignment Multimodal alignment
Multiple streams clock identity + overlap + anchors offset/drift correction, alignment, windowed features Timebase guide
External neuro/physiology files backend/version + exchange semantics MNE, LSL/XDF, BioSPPy, HeartPy, pyHRV, NeuroKit bridges Interoperability
Completed analysis QC evidence + software identity + settings visual audit, reproducibility, interpretation guardrails QC + reporting
Repeated/crossed outcome data grouping structure + prediction target grouped prediction or hierarchical location–scale modelling Model selection

Jump straight to the relevant API

The function finder now preserves its selected domain in the page URL. These shortcuts open the matching API lens directly, so a workflow can move from what to do to which functions implement it without repeating the search. The resulting filtered view can also be bookmarked or shared.

Interoperability / exchange formats Pupil / gaze / fixation / AOI EDA / GSR / SCR PPG / HR / IBI / HRV / respiration Events / alignment / multimodal QC / validation / reporting / governance Statistics / design / simulation Core I/O / schema / utilities 
graph LR
  A[1 · Ingest] --> B[2 · Audit]
  B --> C[3 · Process]
  C --> D[4 · Align]
  D --> E[5 · Summarise]
  E --> F[6 · Model]
  F --> G[7 · Report]
  B -. evidence insufficient .-> R[Review / narrow claim]
  D -. timing unsupported .-> R
  F -. design unsupported .-> R
01IngestRead exports, preserve source identity, standardize schema.
02AuditCheck timing, missingness, signal activity, events and design.
03ProcessApply modality-specific transformations with explicit settings.
04AlignRelate clocks, events, trials, AOIs and secondary streams.
05SummariseDerive analysis-ready features and grouped summaries while retaining denominators and QC context.
06ModelMatch validation unit and random structure to the scientific target.
07ReportRetain QC, provenance, plots, software identity and guardrails.

Three workflow checkpoints

Checkpoint 1

Measurement-ready

Source identity, units, timebase, channel validity, event coverage, and grouping variables are sufficiently documented for processing.

Checkpoint 2

Analysis-ready

Transformations are explicit, alignment is defensible, exclusions are traceable, and the model-ready table reflects the intended scientific unit.

Checkpoint 3

Report-ready

Results are accompanied by QC evidence, settings, figures, versions, prediction semantics, sensitivity checks, and restrained interpretation.

What makes a workflow complete?

Retain at least:

  • raw-to-standardized schema decisions;
  • source/provenance declarations where scientific identity matters;
  • observed timebase evidence and any clock mapping;
  • signal validity, missingness, dropout, and artifact evidence;
  • preprocessing/filter/detection settings;
  • event, trial, condition, and AOI coverage;
  • optional backend/software versions;
  • generated figures used for review;
  • exclusion decisions with reasons;
  • model specification and validation unit;
  • final reporting/reproducibility object or evidence bundle.

Visual workflow evidence

Standardised EDA and heart-rate signal overview
Inspect before derivingSignal overview and quality evidence come before feature extraction.
 Multimodal EDA heart-rate pupil and event timeline
Align before windowingDefine event windows only after the relevant timebases are defensible.

Cross-toolbox validation

gpbiometricspy can work with or prepare workflows for several established ecosystems. These bridges are for interoperability and cross-checking, not for silently changing the package's declared analysis contract.

External toolbox bridges MNE / EEG / LSL Version testing Integration matrix

Need a guided path?