Understanding Your Results
What every number on the result panel actually means, and what to look at first
A successful supported fit shows the fields produced by that method. Promoted fits include a curve, parameters, R², and a factual headline; parameter intervals appear only where the fixed calculator estimated them. This page explains how to review those fields.
Read the curve first, the table second
The plot tells you whether the model is plausible before any number can. Look for the four things that go wrong most often:
- No upper plateau. The top asymptote is unconstrained because the highest dose did not yet flatten. The fitted Top is a guess; the IC₅₀ may shift if you add a higher dose.
- No lower plateau. Same problem, mirrored: the assay never fully “turned off.” Common with cytotoxicity at low end.
- One outlier dragging the inflection. A single hot replicate near the IC₅₀ moves it more than the same outlier at the ends.
- Inflection outside the dose range. The IC₅₀ is being extrapolated. It is a model estimate beyond the observed range, not a directly observed result.
The parameter table
Each row is a fitted parameter with a value and, where available, units and an interval. The supported 4PL reports Top, Bottom, IC₅₀, and Hill; only IC₅₀ and Hill receive parameter intervals.
| Parameter | What it is | Review note |
|---|---|---|
| IC₅₀ | Concentration at half-maximal response | Check whether the tested range brackets it and whether its interval is available |
| Top | Upper asymptote of the response | No interval is reported; inspect plateau coverage and the safety bounds |
| Bottom | Lower asymptote of the response | No interval is reported; inspect plateau coverage and the safety bounds |
| Hill | Dimensionless steepness and direction in this empirical 4PL | Its value alone does not establish stoichiometry or cooperativity |
What parameter intervals do and do not describe
When estimable, supported 4PL fits report local covariance-based 95% Student-t intervals for IC₅₀ and Hill; Michaelis-Menten reports them for Vmax and Km. Calibration coefficients and back-calculated unknowns do not receive intervals. These intervals describe uncertainty in the fitted model under its assumptions, not biological or across-experiment uncertainty.
- Narrower interval. The local fit estimates that parameter more precisely under the model. There is no universal acceptable relative width.
- Wider interval. Review range coverage, residuals, replicate handling, and whether the model is appropriate before interpreting the point estimate.
- Unavailable interval. Too few residual degrees of freedom or near-singular covariance can prevent estimation. The missing interval does not by itself diagnose the cause.
R² and the quality tier
R² is the fraction of variance the model explains. Gradiance auto-derives a quality tier from it:
R² is a descriptive fit statistic, not an acceptance test. A high value does not establish that the model, data mapping, or experiment is valid; inspect the fitted curve, residuals, and executed method together.
Residuals: where the curve and the data disagree
Switch to the residual view to see (observed minus predicted) plotted against concentration. Healthy residuals are scattered evenly above and below zero with no structure. Three patterns worth recognizing:
- S-shape or U-shape. The supported model may not describe the data. Review the mapping and experimental range. Public calibration supports linear or quadratic fits; public dose response supports only the free, unweighted 4PL.
- Funnel (variance grows with concentration). This may indicate nonconstant variance. Weighting is not available in the supported launch methods, so record the limitation rather than assuming a weighted refit was performed.
- One spike with the rest tight. A single outlier. Check the well in the source plate. Exclude it only with an explicit, scientifically justified reason retained in the run.
The plain-English interpretation
Supported promoted fits include a deterministic factual headline assembled from the verified calculator result. The separate executed-method record names the model, weighting, preprocessing, units, and available source counts. Review both before adapting the wording for a scientific report.
QC warnings you may see
- IC₅₀ extrapolated. The fitted IC₅₀ falls outside your tested concentration range. The model is reaching past your data.
- Possible missing plateaus. The fitted 4PL span is less than 70% of the observed response span. Inspect range coverage; this warning does not identify the experimental cause.
- Shallow slope. |Hill| below ~0.5. Either the data are unusually shallow or the model is fitting noise.
- Steep slope. The supported 4PL warns when |Hill| is above 4. Treat it as a prompt to inspect the data and model, not as a mechanistic conclusion.
What to do next
For most runs, the answer is “export.” When something looks off, use Refine to ask the agent to review the mapping or a specific observation. A source-row exclusion must be explicit and retain its reason. Supported launch methods remain unweighted linear or quadratic calibration, free 4PL dose response, and free Michaelis-Menten; unsupported model, weighting, and fixed-parameter requests are refused. Each refinement counts as another run against your shared monthly allowance.
Provenance and the audit bundle
Entitled accounts may have an Audit Bundle (JSON) export that captures available provenance such as input file hashes, recorded agent steps, the chosen model, fitted parameters and CIs, and the Gradiance version. The Python export, when available, is a saved-fit replay; it does not replace the original data and settings needed to reproduce the full analysis.