← ProjectsDesign or Disaster

Criticism · spatial evidence

Case 05 · Interface critique tool2026 · Working archive

Design or Disaster

A verdict needs evidence first: pick a lens, mark the screen, and write your reasoning. Only then do the five juror readings appear beside yours.

Open interactive project New tab Inspect the source

What I did: I framed the critique method, wrote ten cases and five jurors, designed the evidence-map interaction, and built the archive.
Context: Self-directed research-through-design · Independent · concept to production
Main design decision: Evidence before verdict: Require a lens, a located mark and an explanation before any verdict can be stated.
What exists: A complete five-minute path plus a persistent ten-case archive.
Still unproven: The jurors, cases, annotations, and several interface images are authored or reconstructed critique objects. This is not expert research, a usability study, or a population-level result.

Try the interactionPoint at the evidence first

The image, regions, and five readings are authored or reconstructed critique material, not participant data.

What this showsPlace your own mark on the evidence before any other reading appears. The five perspectives are authored examples, not expert truth.

A marked screen, liveCase 001 on the working archive: the Hierarchy lens chosen, four markers placed, each one carrying an evidence sentence. Only after the evidence exists can a verdict be submitted.

Scale · 10 cases · 5 jurors · 2 experience pathsBuilt with · React · JavaScript · Vite · CSS Modules · Browser-local state

The situation

Design criticism often begins with a verdict and works backward.

A review says that an interface is clear, confusing, delightful, or broken. But the precise detail that produced the judgment is usually left implicit. Once that evidence disappears, disagreement becomes a contest of taste, confidence, or seniority.

I wanted critique to leave a visible trail. Before anyone could rule, the interface would require them to choose a lens, locate evidence, explain what it showed, and only then state a verdict.

What it is

Every case ends with five other readings on the same screen. They can agree with your mark, contradict it, or miss what you saw. I wrote them as fallible critics, not experts.

Every mark lives on one coordinate system, so disagreement stays visible instead of collapsing into a score.

Two people can disagree more usefully when both marks are still on the screen.

What changed during the build

Using one coordinate system changed the critique from a sequence of opinions into something people could inspect together.

The archive works, but whether it improves a real critique session still needs to be tested with people.

The reasoning

The project stopped asking who was right and started showing how each reading was made.

The turn

Where it started
The obvious versions were a design quiz, an expert score, or a crowd percentage. Each would end by telling the visitor how their judgment compared with authority.
What changed
The useful object was the evidence map created before any score: a record of what entered one person's attention, and what never did.
Where it landed
I built five coherent but fallible jurors on one shared coordinate system. Their annotations can agree, collide, or miss one another without any perspective becoming the official answer.

How the interaction works

Every case uses the same sequence. The five-minute path teaches it; the full archive shows how your attention repeats across ten cases.

Choose what kind of evidence you are looking for: hierarchy, access, task, trust, or feeling.
Place a mark on the exact coordinate, or choose a named region when precision input is unavailable.
Describe what that place makes visible and write the sentence the evidence supports.
File a ruling and state your own confidence. The interface does not infer confidence from behavior.
Open the juror maps. Five authored interpretations, written for the archive rather than research participants, appear beside your mark on the same screen.
Reach the sealed case, where every other interpretation is withheld and your evidence is the only record.

The system underneath

Input
A spatial claim
Lens, coordinate or named region, evidence sentence, verdict, and self-reported confidence form one review record.
Rule
Evidence precedes authority
Pending cases hide curator notes and panel status. Comparison unlocks only after the visitor has committed a reading.
Output
A trace, not a grade
The Design Eye record shows which lenses you used and preserves ties. It never claims to diagnose you or to represent anyone else.

Considered, then dropped

Expert verdict
It would hide the expert's own selection of evidence behind institutional authority.
Crowd consensus
Agreement would look like accuracy, even though the archive contains no live population data.
Opinion-first prompt
It would let a verdict form before the visitor had made their evidence inspectable.

The record

Decisions and open questions.

One coordinate system

Store every mark as screenshot-relative percentages.

Visitor and juror evidence stays aligned through responsive resizing and perspective changes.

Fallible jurors

Write five authored positions instead of simulating experts.

The panel demonstrates structured disagreement without manufacturing authority or telemetry.

Ties stay ties

Preserve shared strongest and weakest lenses.

The report refuses to invent a personalized winner from ambiguous evidence.

Equivalent input paths

Support pointer, keyboard cursor, touch, and named-region placement.

The research grammar survives when precise pointing is unavailable.

Built and working

A complete five-minute path plus a persistent ten-case archive.
Five spatially annotated juror perspectives for every open case.
Pointer, keyboard, touch, and named-region evidence placement.
Validation for annotation bounds, evidence alternatives, content invariants, assets, encoding, and production build.

Not yet proven

The jurors, cases, annotations, and several interface images are authored or reconstructed critique objects. This is not expert research, a usability study, or a population-level result.
The browser-local Design Eye record is a trace of selection, not a diagnosis of the visitor.

Next test

Use the grammar inside real critique sessions.

Give design teams the same interface and compare an ordinary verbal critique with an evidence-map critique. Study whether claims become more specific, whether overlooked lenses enter the conversation, and whether disagreement becomes easier to inspect.

What would countThe success measure is better-supported critique, not higher agreement between reviewers.

AI assisted ideation, critique, writing iteration, and code iteration. Final concept selection, case construction, design decisions, editing, implementation, and authorship are Tanishk's.

A critique needs evidence before a verdict. Pentimento gives the person being described the final word.

Interactive essayPentimento

An interactive essay about who gets to tell someone's story. Software writes a sentence about a person; they can keep it, rewrite it, or cross it out. Their words take priority.

Also: Atlas →

← Back to projects