The review
Seven seats read the built screen, blind to each other. Refuters attack every finding. The score is the lowest seat, never an average — because an average lets you hide the one thing that would fail you.
2 scored engagements 16 rounds gate 9.5
How a round works
- 01
The surface is frozen
Nothing changes during a round. Every seat reads the same build at the same revision.
- 02
Seven seats sit blind
Each seat reviews on its own axis with no knowledge of what the others found, and returns a score and a list of findings. No seat sees the previous round's scores.
- 03
Refuters attack every finding
A separate pass tries to kill each finding against the actual build. Most die: in the last Tasks round, 31 of 35 findings were refuted. Only survivors get fixed.
- 04
The lowest seat is the score
Not an average. A board that is beautiful and unusable scores as unusable.
- 05
Fix, then run it again
Rounds continue until the lowest seat clears the gate, or until the distance left is small enough to name in days and decide about deliberately.
The seats
UI composition
Layout, rhythm, alignment, density.
Typography
The ramp, measure, tracking, and whether type is doing the hierarchy.
Interaction and states
Hover, focus, press, empty, error, keyboard, and what happens under a screen reader.
UX and information design
Whether the screen answers the question the person actually arrived with.
Brand and copy
Voice, naming, and whether the words are true.
Product taste and emotional resonance
Whether it is memorable, and whether it feels like one hand made it.
Measured evidence
The numbers: contrast, budgets, gate results. This seat reads the results file, not the picture.
Master Suite
Tasks, Notes and Timeline as one suite · August 2026 · 5 rounds, lowest seat 7.2 → 8.6 against a gate of 9.5.
- R1 7.2
- R2 7.5
- R3 7.4
- R4 7.6
- R5 8.6
Each dot is one seat. The bar is the lowest seat — the score. The line is the gate.
Seven seats, 29 findings, 25 standing — 4 blocking, 6 misleading, 12 defect, 3 refinement — and the engagement's first sign-off, from the seat that scored it 9.0.
Every seat, every round
| Seat | R1 | R2 | R3 | R4 | R5 |
|---|---|---|---|---|---|
| Typography | 7.8 | · | 8.7 | 9.0 | 8.7 |
| Brand and copy | 7.2 | · | 8.7 | 8.3 | 8.7 |
| Interaction and states | 7.6 | 8.5 | 7.4 | 8.3 | 8.7 |
| Product taste and emotional resonance | 8.2 | · | 8.6 | 9.0 | 9.0 |
| UI composition | 7.6 | · | 8.4 | 8.9 | 8.6 |
| UX and information design | 7.4 | 7.5 | 8.4 | 7.6 | 8.6 |
| Measured evidence | 7.4 | 8.4 | 8.6 | 8.7 | 8.6 |
| Lowest seat | 7.2 | 7.5 | 7.4 | 7.6 | 8.6 |
| Confirmed | 27 | 7 | 6 | 2 | 3 |
| Refuted | 43 | 7 | 29 | 7 | 4 |
Source · signal-studio-workspace/worktrees/app/design-master-suite-2026-08/docs/design/labs/master-suite-2026-08/panel.json
Tasks
The Tasks board · August 2026 · 11 rounds, lowest seat 6.3 → 8.1 against a gate of 9.5.
- R1 6.3
- R2 7.1
- R3 7.2
- R4 7.2
- R5 6.3
- R6 7.1
- R7 7.4
- R8 8.3
- R9 8.6
- R10 8.5
- R11 8.1
Each dot is one seat. The bar is the lowest seat — the score. The line is the gate.
The final round, graded as a finished artifact rather than as work in progress.
Every seat, every round
| Seat | R1 | R2 | R3 | R4 | R5 | R6 | R7 | R8 | R9 | R10 | R11 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| UI composition | 6.3 | 7.6 | 7.2 | 7.6 | 7.2 | 7.6 | 7.4 | 9.0 | 8.9 | 8.5 | 8.5 |
| Typography | 7.1 | 7.2 | 7.4 | 7.5 | 7.2 | 7.5 | 7.8 | 8.5 | 8.9 | 8.7 | 8.7 |
| Interaction and states | 6.3 | 7.1 | 7.2 | 7.4 | 6.3 | 7.2 | 8.0 | 8.3 | 8.7 | 8.5 | 8.2 |
| UX and information design | 7.2 | 7.4 | 8.0 | 7.4 | 6.4 | 7.2 | 7.4 | 8.5 | 8.9 | 8.7 | 8.3 |
| Brand and copy | 7.8 | 7.6 | 7.6 | 7.6 | 8.1 | 7.4 | 7.8 | 8.6 | 8.6 | 8.9 | 8.6 |
| Product taste and emotional resonance | 7.1 | 7.1 | 7.2 | 7.4 | 7.4 | 7.1 | 7.4 | 8.5 | 8.7 | 8.7 | 8.4 |
| Measured evidence | 6.4 | 7.2 | 7.3 | 7.2 | 7.4 | 7.3 | 7.9 | 8.3 | 8.6 | 8.7 | 8.1 |
| Lowest seat | 6.3 | 7.1 | 7.2 | 7.2 | 6.3 | 7.1 | 7.4 | 8.3 | 8.6 | 8.5 | 8.1 |
| Findings | 35 | 35 | 35 | 35 | 35 | 35 | 35 | 35 | 35 | 35 | 35 |
| Confirmed | 20 | 17 | 17 | 21 | 29 | 24 | 29 | 28 | 28 | 26 | 4 |
| Refuted | 15 | 18 | 18 | 14 | 6 | 11 | 6 | 7 | 7 | 9 | 31 |
Source · signal-studio-workspace/app/docs/design/labs/tasks-2026-08/panel.json
Rounds without a score
Some engagements produced findings and decisions but no per-seat numbers, or numbers that cannot be treated as independent. They are recorded here rather than turned into a chart.
Weddings
The weddings planning surface · September 2026
Seven seats sat and produced 35 findings against the weddings board. The scores in the record are identical to round one of the Tasks engagement, so they are inherited rather than independently produced — the round is published for its findings and shown without a number.
- Product taste and emotional resonance Give the header a subject. Right now the board is about a workflow; one 15px/600 ink line naming the nearest real event and its countdown — "Menu tasting · 1 Aug · in 14 days" — makes it about a wedding, and demoting the completion ratio…
- Interaction and states Make the completion circle a real control.
- Brand and copy Unify the time chip to one grammar and one vocabulary.
Source · signal-studio-workspace/worktrees/app/design-weddings-exploration/panel-round-1.json
Product films
Drive, Nudge, Review and Blueprint films · September 2026
Three rounds of frame-by-frame critique against the rendered build, with confirmed findings fixed and re-shot. Recorded as a critique log rather than a scored panel: no per-seat numbers were taken.
Source · files-pasted-by-the-user-signal-2/work/film-elevation/
CEO reporting system
The founder-facing report families · September 2026
Six blind seats answered one brief in parallel, then a refuter seat challenged every claim. The outcome was an architecture decision, not a score.
Source · signal-studio-workspace/labs/report-elevation-2026-09/
Process instruments
The fourteen working models · September 2026
A shorter loop on the instrument component itself: stage timing, artefact legibility and the exception path.
Source · files-pasted-by-the-user-signal-2/work/labs/process-stories-2026-09/
What these numbers are and are not
Read this before quoting a score
The seats are models, not people. Each is a separate agent given one axis, one rubric and no sight of the others. That makes them consistent and tireless; it does not make them users. A seat has never been annoyed by a product at 4pm on a Friday.
The scores are internal. They compare a build to the previous build on the same rubric. They are not comparable to anyone else's numbers, and a 9.5 gate is a bar I set.
The refutation rate is the useful signal. In the last Tasks round, 31 of 35 findings did not survive contact with the actual build. A review process that confirms everything it finds is not reviewing, it is agreeing.
Nothing here reached the gate. The best result on this page is 8.6 against 9.5. That gap is the honest state of the work, and it is the reason the loop is still running.