How this works
Eight positive attributes. Transduce and Data Fusion under SENSE; Algorithm plus three adaptation columns under THINK; Actuation and Closed-Loop under DO. Every cell is strictly 1 or 0. A point is awarded only where a robotic capability is present, which is why the flyball governor scores 1 for all its mechanical ingenuity.
Adaptation is three columns, not one. A single Adaptive bit could not tell a vacuum re-planning around a chair leg apart from a car predicting whether a pedestrian will step off a kerb. Rather than grade one cell 0–2 and break the binary rule, the domain is split by when the adaptation happens. Recover is the present: it meets an unscripted state and departs from the routine. Anticipate is the future: it acts on a projection of how the world will change, including things it does not control — planning a route through known terrain does not count, or every path planner would qualify. Learn is the past: experience changes later behavior and the change persists. This is the reactive / deliberative / learning trichotomy that control architectures have used for decades. 74 entities Recover, 28 Anticipate, 24 Learn, and only 10 do all three.
Verdict bands. 6–8 Robot, 4–5 Borderline, 0–3 Non-Robot — the same proportions as the old six-point scale. Splitting Adaptive moved 30 entities, almost all of them industrial machines dropping from Robot to Borderline because they earn Algorithm and nothing in adaptation. The FANUC arm was always described here as failing adaptation; now the verdict says so too.
A known distortion. THINK now carries four of the eight points while SENSE and DO carry two each, so the total is weighted toward cognition by construction. The score is a lossy summary either way: 11110000 and 00001111 both total 4 and describe opposite machines. Open Profiles to compare by shape rather than by total.
Four exclusions, not one. Grown is biological origin, Unbodied is absence of a physical body, Piloted is intent originating outside the machine, and Unbuilt means it does not exist. These have nothing to do with each other and an entity can trip more than one — six do — which a single disqualifier column could not express. Only Piloted is a property of an operating mode rather than of the object, so it is the flag most worth arguing about.
Fiction is in the table, and excluded from it. Eighteen entries are depictions: Commander Data, HAL 9000, Talos of Crete, and so on. Every cell in those rows records what a story asserts rather than anything anyone measured, so they carry Unbuilt and are marked Depicted rather than given a confidence level they cannot earn. This keeps them out of the robot count — without it, Data's 8 of 8 would sit above Waymo on a screenwriter's say-so. Use the Real only filter for the original real-world dataset, or Fiction to look at the depictions on their own. The pattern that shows up immediately: fiction saturates the matrix. Thirteen of the eighteen score a perfect 8, because writers hand their robots every capability at once and never have to make one work. The matrix cannot tell Data from C-3PO. It has no trouble telling a Roomba from a FANUC arm.
Scope names the box you drew. ABS scores 6 as a Component of a car. A Waymo is a Unit, an autonomous greenhouse is a System, an ant colony is a Collective. Redraw the box and the answer changes, so the box is recorded rather than assumed.
Confidence is not decoration. Settled means nobody who has read the definitions will fight you. Arguable means the call turns on a line the framework leaves soft, and is applied by default to the whole Borderline band. Disputed is set by hand where informed people actively disagree or the machine's own capability is contested. Depicted is not a confidence level at all: it marks a row about something that does not exist.
Everything is editable. Tap any row to open its reasoning, tap any cell to change the call, add your own entities, then export. Compare loads several exported files at once and marks the cells a group disagreed about, which is the part worth doing in a classroom.
Add an entity
New entities start at 0 across the board with no exclusions. Score them yourself, then export. They are stored in this browser alongside your edits.
Profiles
Each distinct six-bit word, read as Transduce, Data Fusion, Algorithm, Adaptive, Actuation, Closed-Loop. Two entities with the same score can have nothing else in common; this is where that shows. Select one to filter the matrix.
Compare exported files
Load several exported JSON files — one per student, per group, or per sitting — and every cell gets a vote count. Cells the group split on are outlined, and the Contested sort key brings the biggest arguments to the top.