Althea C.
Back to the demo →

Explorations

Five decisions behind the build.

Forty-eight hours and six versions. These are the five choices that changed what the thing became, including the one that reframed the other four.

This started at a design workshop in London on 1 August 2026.

WED 5 AUG
09:0010:0011:0012:0015:0016:00
Althea Coutinho09:30 · completed
Maya Ellison11:00 · upcoming
Open · 16:00 to 17:00
Maya Ellison
11:00 to 12:00 · Senior Product Designer · London
4 open questions
One hour. Starts in 20 minutes.

Exploration 01 · The user

A recruiter decides who gets an hour. A hiring manager has to spend it.

The first version sorted applications, which is a volume problem and a fairly solved one. The hiring manager has the narrower problem and the worse tools. They get one conversation, and they usually open the CV twenty minutes before it starts. So the surface stopped being about who to see next and became about what to ask in the hour that was already booked.

Professional shape

Design
9 claims · 8 consistent, 1 differs across sources
Product
4 claims · 4 consistent
Engineering
2 claims · 1 consistent, 1 stated once
Consistent Differs across sources Stated once
Composition of what the candidate has claimed, not an assessment of ability.

Exploration 02 · Scoring

A score tells a hiring manager what to conclude. What they need is what to ask.

Every competitor in this category returns a match percentage. I built the shape view on counts and states instead, so a bar shows how much of a candidate's account exists and how much of it agrees, and infers nothing about ability. The radar chart went first. It reads as a rating whatever the caption underneath says.

Role fit

Pass and fail apply only to the factual checks.

Checks
✓
Able to work onsite in London three days a week
Source: interview, 5 Aug
Pass
✓
Full UK working rights, no sponsorship required
Source: application form
Pass
Signals
Has shipped complex products end to end
Evidence found
Cares about the details most people never notice
Not asked yet

Exploration 03 · Pass and fail

Pass and fail are honest about a work authorisation. They are not honest about taste.

Role requirements arrive in one list, but they are not the same kind of thing. Onsite days, working rights and notice are facts. Shipped complex products end to end is a judgement. Splitting them into Checks and Signals let the interface stay definite where it had the right to be, and say only whether evidence was found where it did not.

From this candidate 3
Q:Show me one thing you have built and run yourself.
Not yet asked
No artefact in three sources
Send to my notesDismiss
Q:What was the 34% measured against?
Not yet asked
Stated once
Send to my notesDismiss

Exploration 04 · Agent output

An agent that suggests a question has to say why it is asking, and take no for an answer.

Every suggested question carries its reason underneath: no artefact in three sources, differs across sources, stated once. Dismiss is one click and asks for no justification. Nothing is asked automatically. The reason sits there so the choice is informed rather than obedient.

Exploration 05 · What shipped

Every tool in this category scores the candidate. This one scores the job description.

Four explorations in, the subject was still the candidate. For a role where design, product and engineering blend into one job, the posting is the part nobody has tested. Across forty-one interviews, the requirement the team would least compromise on turned out to be the one with the least evidence behind it. That reframing is the version that shipped.

The bar for this role
Across 41 interviews, you have evidence on two of the four things you said matter.
The requirement you'd least compromise on is the one you have the least evidence for.
24 of 26 Real point of view on product
21 of 26 Ships end to end
9 of 26 Craft at the pixel level
4 of 26 Brings it to life themselves
4 candidates have cleared all →