Source: bake-off #1 Form D (Team Landing — Lens Bake-off) — c100 matrix thread, 2026-09-30 · sibling of Tile Map — Variations · plan cycles/c100.landing-pages/matrix-variations-plan.md
The tile map's three asks, applied to the matrix — which has two axes to label, not one. Ink: greyscale turns eight side-by-side charts into one picture. Size: numbers go first, then column names, then row names — and at DataMachine size the move the tile map doesn't have is to label the short axis. Spin: a matrix is the most reorderable form there is — Bertin's point was that reordering is the analysis. Hover any cell: the same team and lens light up in every matrix on the page.
Ink — one picture, or eight charts?
Bake-off #1 gave every column its own colour job — grape for magnitude, status for risk, amber for gaps — so the matrix reads as eight charts side by side. Greyscale makes every column the same heat: one picture, with the columns still shaded against their own max. Critical cells keep a corner notch. The picker sets the hue for 3 and 4 and every other matrix on the page.
Size ladder — table to fingerprint
A matrix sheds labels in a fixed order: numbers, then column names, then row names. Past that it is a texture — and one row of it is a team's fingerprint, small enough to live inside a sentence or a listing column.
Labelling at DataMachine size
Two axes, 88 px of height. Six ways to put the names somewhere — including the one the tile map can't do: turn it sideways and label the eight lenses instead of the forty teams. Each is a real DataMachine.
Other variations — for a spin
Once columns share one ink, the shade's reference becomes a design choice. Per column compares teams on one lens; per row shows each team's own outlier; rank resists one huge team flattening the rest; deviation asks 'unusual for us?' — filled when a team sits past the company median on the lens's attention side, hollow when it sits on the calm side.
Story order sorts teams by the chosen lens (the lever strip in disguise). Size order is stable across lenses. Seriated orders both axes by similarity — teams with the same profile sit together, lenses that move together sit together — so blocks appear that no single sort shows.
Fill is the heatmap. Dot puts value in area and leaves the ground neutral. The in-cell bar trades the image for precision — lengths compare where shades don't. Coloured numerals are the table with a whisper of heat: the most precise and the least 30 ft.
Rows sized by team: big teams are tall, and a cell's area now says how many people the value is about. At 1,500 the small teams become hairlines — the same labelling problem the tile map has, arriving from the other side.
Columns banded by who reads them (the c100 lens ledger: Eng manager · FinOps · Security) — the matrix becomes three small matrices, one per Home lens. The chosen lens stays in full colour; the rest dim to context, so the matrix answers one question while keeping the others in view.
Turn Morph on, then change the Lens: rows glide into the new story order instead of jumping, so your eye keeps hold of a team while the order changes. Worth trying on 4b's story column and the XL matrix.
What to try, and what to look for
Compare §1 1 and 4. Per-job ink says "eight different charts"; grey says "one heat". Which do you read faster — and is the loss of kind (volume vs risk) a loss, when the column header already says it?
Walk §2. At which size does it stop being a table and become a texture? Is the one-row fingerprint a picture you would read in a listing, or decoration?
§3 2 labels the eight lenses instead of the teams. Flip Scale → 1,500: forty thin columns, eight named rows. Does that beat the tile map's best DataMachine strategy?
Walk §4a on Spend and Violations. Deviation-from-median is the only one that says "unusual for us" — does the hollow "calmer than typical" mark read, especially in grey?
Look at §4b's seriated matrix at 200 and 1,500. If blocks appear (the spenders are the violators; the late adopters are the uncovered), that is a finding no single sort shows — and a case for a "similar teams" view.
§4e bands the columns by reader. Is the matrix really three matrices — one per Home lens — and should each Home lens carry only its own band?