Uploading Data - Gradiance Documentation

Gradiance reads CSV and Excel directly. There is no required template — the agent inspects whatever you upload and asks if it can't identify the shape. This page covers the formats that work without prompting and the few cases where the file needs adjusting.

What we accept

CSV (.csv) — comma- or semicolon-separated. UTF-8 or Latin-1 encoding. Headers in the first non-empty row.
Excel (.xlsx, .xls) — multi-sheet workbooks supported. The inspector picks the first sheet with numeric data; specify a different one either through the picker or in your prompt (“use the Results sheet”).
Tab-separated (.tsv, .txt) — parsed when extension or delimiter indicates tabs.
Multi-file uploads — drop several files at once and Gradiance treats each one as a related dataset inside the same run. Datasets are labelled with the source filename (e.g. plate_A.csv · Sheet1) so the painter and result panels stay legible. Common case: one plate of standards and one of unknowns, fitted together.

Max upload size is 50 MB per file. Plate-reader exports are rarely close to that.

What happens after the drop

Gradiance inspects every file the moment it lands. From shape alone — the number of rows, the dilution-factor pattern, the presence of a 96-well grid — it suggests likely assays at the top of the picker. You can accept the suggestion, override it, or pick Custom Analysis if nothing fits.

For plate-style data the plate painter opens automatically when you select an assay that uses one (standard curve, dose-response, viability, BCA, binding, kinetic ELISA). Standards, unknowns, blanks, and controls get labelled in a few clicks per plate; the agent reads the layout directly without needing role columns in the file.

The two layouts that work

Long format (one row per measurement)

Each row is a single observation. Replicates repeat the concentration; conditions get their own column. This is the format the agent prefers — every supported workflow accepts it.

Concentration_uM,Response,Replicate
1,45.2,1
1,47.1,2
1,46.8,3
10,12.3,1
10,11.9,2
10,12.8,3

Wide format (one column per replicate)

Each replicate gets its own column. The agent reshapes this to long internally before fitting.

Concentration,Rep1,Rep2,Rep3
1,45.2,47.1,46.8
10,12.3,11.9,12.8

Plate-reader exports (96/384-well grids)

Plate-reader software typically exports a 2D grid: row labels (A–H), column labels (1–12), and a value in each cell. Gradiance auto-detects this shape and opens the plate painter so you can label each well's role (standard, unknown, blank, control) with their concentration or sample ID.

You don't need to reshape these by hand. Drop the export as-is. The painter is the fastest path for any data that came off a plate reader.

Column conventions that help the agent

Include units in headers

Concentration_uM, Dose_nM, Viability_Percent, Absorbance_450nm. The agent uses these to label axes correctly without asking, and to convert units when comparing across runs.

Use plain ASCII

Underscores and hyphens are fine. Avoid:

Brackets and parentheses ([Conc], Response (µM))
Spaces if you can avoid them (Sample ID → Sample_ID)
Greek letters in column headers (use them in values; not headers)
Generic names (Column1, Var1)

Numbers should be numbers

Decimal separator. Use a period (3.14), not a comma. Most non-US locales export Excel with comma decimals — change the locale before exporting, or run a search-and-replace.
No units in cells. 2.4 uM doesn't parse. Put the unit in the header.
Special values. NA, N/A, nd (not detected), OOR (out of range) are recognized as missing and excluded from the fit but kept in the audit trail.TNTC is recognized for CFU counts.
Concentrations should be raw, not log-transformed. The agent applies log scaling for fitting; if your data is already log10([conc]), tell the agent (“these are already log-transformed”) so it doesn't double-transform.

Common parse errors and what fixes them

“Could not detect a concentration column.”

Either your concentration values are formatted as text (Excel does this with custom number formats), or the column header looks like a date or sample ID. Open the file in Excel, set the column to General or Number, and re-export.

“Header row not found.”

Plate-reader software often dumps metadata into the first 5–15 rows before the actual table. The agent skips them automatically if they're sparse (a few cells per row); if they look table-shaped (full rows of text), it will ask. Either delete the metadata before upload or tell the agent which row the headers are on.

“Multiple sheets — which one?”

Pick the one with the data. Or in your prompt: “use the ‘Results’ sheet,” “use sheet 2.”

Merged cells

Excel merged cells are unmerged on read. The merged value lands in the top-left cell only; the rest are blank. If your file relies on merged cells for grouping, unmerge and fill down before uploading.

Formulas

The agent reads computed values, not formulas. If you exported with formulas evaluated to errors (#REF!, #N/A), they're treated as missing. Re-save with Paste Special → Values if you want every cell explicit.

Practical checklist before uploading

Headers in row 1, with units
Numbers as numbers, periods for decimals
One sheet of data; remove decorative summary tables
No merged cells in the data region
Empty columns and trailing rows trimmed
If it's a plate export, leave the grid intact — the painter handles it