Custom Analysis - Gradiance Documentation

Custom Analysis (sometimes called “freestyle” in the API) is the assay you pick when your experiment doesn't map cleanly onto any of the named workflows. The agent inspects your data and experiment, then chooses an implemented assay-library method with its exact input and option requirements. The reported numerical result still comes from a bounded, deterministic calculator.

When to reach for Custom Analysis

You do not know which named workflow fits. Describe the measurement and controls, and the agent can choose among the supported library methods after inspecting the table.
Your terminology is ambiguous. A “titer,” “binding,” or “response” request can refer to several implemented methods; Custom Analysis can resolve the intended one.
Your table needs mapping. State which columns carry observations, controls, groups, and units when the headers do not establish those roles.
You need a supported one-off summary. The library includes descriptive summaries, group comparisons, correlation, normality, ROC, Bland-Altman, time-course, and nested comparisons.

If your assay is in the picker, use the named workflow. They're shorter prompts (the agent already knows the model, the QC flags, and the export shape) and the prompts have been hardened against common edge cases.

What the agent will do

Freestyle runs follow the same three-step pattern as every other workflow:

Read & inspect. Load the file, identify columns, surface obvious shapes (long-format vs wide, plate vs tabular, replicate count per condition).
Choose and compute. Search the assay library, load the selected method guide, and supply its required model, controls, units, and other scientific options. If those facts are missing, the agent asks a focused setup question.
Report. Stream a report_fit payload per dataset to the viewer, then a one- to two-sentence narrative interpretation.

That structure means Custom Analysis results land in the same viewer and retain the selected library method. Export availability follows the account entitlement: Free includes20 free runs per calendar month (UTC), plus PNG and basic CSV results, while other formats may require Pro or Team. Cite the result with the same method as named workflows.

Writing a prompt that lands

The agent starts from your prompt plus the file shape; the prompt is what disambiguates when the file alone is insufficient. Three rules of thumb:

Name the metric. For example, request normalized western-band intensity, ΔΔCq fold-change, per-analyte multiplex concentration, or a Welch group comparison.
Identify the columns when names are ambiguous. “Use the OD600_blank_subtracted column, not OD600_raw.” If column names already spell out the role (units in the header, role in the name), you can usually skip this.
State the scientific settings. Name the reference condition, loading control, calibrator, background group, dilution, and units that define the requested method.

Examples that work

Each prompt identifies an implemented method and the metadata needed to run it:

“Normalize imported target-band intensities to GAPDH and report each sample relative to the vehicle sample.”
“Use RPLP0 as the qPCR reference gene and vehicle as the control sample; report efficiency-aware ΔΔCq fold-change.”
“Run a Welch comparison of Yield between the two groups and report the effect size.”
“For each analyte, fit a linear calibration in pg/mL and quantify unknown signals only within its standard range.”
“Compute background-subtracted ELISpot SFU per million cells using media as the matched background; do not make positivity calls.”

Limits worth knowing

Runtime cap. Each run is bounded at 30 minutes of active agent time (waits for user input don't count). Most freestyle runs finish in under a minute; the cap exists to catch loops, not to constrain legitimate analyses.
Step cap. The agent is capped at 20 code steps per run. The standard-workflow prompts target 3 steps end-to-end; freestyle analyses use more, but anything bumping against 20 steps is usually stuck and better off being restarted with a more specific prompt.
Restricted imports. The Python executor allows numpy, pandas, scipy, matplotlib, openpyxl, plus the stdlib json / csv / typing/ functools / operator / traceback. Network modules (requests, urllib, socket) are not on the allow list, so a freestyle run can't fetch a reference dataset from the internet - upload it as a second file instead.
Replay export, when entitled. The analysis record may be captured in the same <runname>_reproducibility.py as the named workflows. Read its input and dependency instructions: it replays saved observations and fit records and does not guarantee a fresh fit from the original raw file.

When freestyle is the wrong answer

Two cases where you should pick a named workflow even if freestyle would technically work:

You already know the assay. Pick its named workflow so the agent begins with the exact required columns, options, and assay-specific QC rules.
You need a fixed method before upload. A named workflow records the selected model directly. Custom Analysis must first resolve a library assay and cannot authorize an algorithm that is absent from the catalog. See Statistical methods.

Reference

The run workspace

Refine, stop, share, and the live code panel.

Reference

Statistical methods

The exact algorithms behind every fit, named or freestyle.

When to reach for Custom Analysis

You do not know which named workflow fits. Describe the measurement and controls, and the agent can choose among the supported library methods after inspecting the table.

Your terminology is ambiguous. A “titer,” “binding,” or “response” request can refer to several implemented methods; Custom Analysis can resolve the intended one.

Your table needs mapping. State which columns carry observations, controls, groups, and units when the headers do not establish those roles.

You need a supported one-off summary. The library includes descriptive summaries, group comparisons, correlation, normality, ROC, Bland-Altman, time-course, and nested comparisons.

What the agent will do

Freestyle runs follow the same three-step pattern as every other workflow:

Read & inspect. Load the file, identify columns, surface obvious shapes (long-format vs wide, plate vs tabular, replicate count per condition).

Choose and compute. Search the assay library, load the selected method guide, and supply its required model, controls, units, and other scientific options. If those facts are missing, the agent asks a focused setup question.

Report. Stream a report_fit payload per dataset to the viewer, then a one- to two-sentence narrative interpretation.

Writing a prompt that lands

The agent starts from your prompt plus the file shape; the prompt is what disambiguates when the file alone is insufficient. Three rules of thumb:

Name the metric. For example, request normalized western-band intensity, ΔΔCq fold-change, per-analyte multiplex concentration, or a Welch group comparison.

Identify the columns when names are ambiguous. “Use the OD600_blank_subtracted column, not OD600_raw.” If column names already spell out the role (units in the header, role in the name), you can usually skip this.

State the scientific settings. Name the reference condition, loading control, calibrator, background group, dilution, and units that define the requested method.

Examples that work

Each prompt identifies an implemented method and the metadata needed to run it:

“Normalize imported target-band intensities to GAPDH and report each sample relative to the vehicle sample.”

“Use RPLP0 as the qPCR reference gene and vehicle as the control sample; report efficiency-aware ΔΔCq fold-change.”

“Run a Welch comparison of Yield between the two groups and report the effect size.”

“For each analyte, fit a linear calibration in pg/mL and quantify unknown signals only within its standard range.”

“Compute background-subtracted ELISpot SFU per million cells using media as the matched background; do not make positivity calls.”

Limits worth knowing

Runtime cap. Each run is bounded at 30 minutes of active agent time (waits for user input don't count). Most freestyle runs finish in under a minute; the cap exists to catch loops, not to constrain legitimate analyses.

Step cap. The agent is capped at 20 code steps per run. The standard-workflow prompts target 3 steps end-to-end; freestyle analyses use more, but anything bumping against 20 steps is usually stuck and better off being restarted with a more specific prompt.

Restricted imports. The Python executor allows numpy, pandas, scipy, matplotlib, openpyxl, plus the stdlib json / csv / typing/ functools / operator / traceback. Network modules (requests, urllib, socket) are not on the allow list, so a freestyle run can't fetch a reference dataset from the internet - upload it as a second file instead.

Replay export, when entitled. The analysis record may be captured in the same <runname>_reproducibility.py as the named workflows. Read its input and dependency instructions: it replays saved observations and fit records and does not guarantee a fresh fit from the original raw file.

When freestyle is the wrong answer

Two cases where you should pick a named workflow even if freestyle would technically work:

You already know the assay. Pick its named workflow so the agent begins with the exact required columns, options, and assay-specific QC rules.

You need a fixed method before upload. A named workflow records the selected model directly. Custom Analysis must first resolve a library assay and cannot authorize an algorithm that is absent from the catalog. See Statistical methods.