The brief carries its own tension and it is the whole assignment: make someone say wow, for an audience who reads flourish as someone selling me something. So the wow cannot come from the interface. It has to come from what the product knows.
One idea holds the round together. The only material in this product that is not ours is the filing — so the filing is the only thing that gets a material of its own. The app stays charcoal and quiet; source text is warm paper, set as a document, and it is the brightest surface in the room. The eye is drawn to the evidence rather than to us. That is Ive's honest materials applied literally, and it doubles as the positioning: credibility as the aesthetic.
The test I held everything to. Would this make a five-year investor trust the tool more, or notice the tool more? Anything that made them notice the tool got cut — which is why there is exactly one moving element on the centrepiece screen and three motions in the entire vocabulary.
I looked for a better one and did not find it, so I am not going to manufacture a rival to look independent. Scott's candidate is right: the moment a claim you wrote three months ago meets a filing published this morning. It is uncopyable for a structural reason rather than a craft one — it needs a sentence the user wrote before the filing existed, held unedited, and a competitor that returns Supported/Challenged has already trained its users never to write one.
What I added is the part that makes it land, and all three are design decisions:
Asked the way it was posed, and the answer is a real one: 12 → 6, grouped by the question the investor is actually asking. Screener and First Pass are one decision with a screen break in it. Decision, Memo and Allocation are one act split across three screens — today a user can complete two of the three and not know the act is unfinished. Order is unchanged; I have not touched sequencing, because that critique sits in an unapproved section of the brief document. The grouping, with reasoning →
Three motions exist in the whole system: a thing travels to where it now lives, a citation originates from its source line, and the agent recedes to a hairline and never opens itself. Three durations, one curve, no overshoot — overshoot is decoration and decoration reads as salesmanship here. Every motion is sequenced so its first frame is the one that teaches, which is what makes it survive real latency: the destination lights before the card moves, so a stalled network still leaves the relationship learned.
"The agent arrives when the user is stuck." I have not designed detection. For someone deciding where to put real money, a tool that watches for hesitation and acts on it reads as surveillance, and the credibility cost is larger than the help is worth. What I built instead: the agent is one line while you read, and it may change that line to a question after a long pause on an unanswered gate — but it never expands itself. Presence proportional to invitation. If you want true detection, that is a product decision rather than a design one.
A tension I created myself, three tasks ago. I wrote the contract telling Charlene to eliminate every light panel in the dark app, and this round introduces a paper surface. That is not a reversal and I would rather name it than have it read as one: those panels were accidents of an unconverted stylesheet, light because nobody had converted them. This is a material, reserved for verbatim source text and nothing else — never UI, never our prose, never a container we wrote. The rule is testable and it is tested.
No glassmorphism, no particles, no parallax, no gradient fields, no springing easings, no page transitions, no numbers counting up, no hero animation. Every one was available and every one would have made a tool for deciding where to put real money look like something being sold. The constraints that bind held throughout: no scores, ranks, verdicts or recommendations anywhere; honest absence stays visible; citations stay verbatim. The honest-absence state is built as a real screen rather than described — a filing that does not address your invalidator says so, instead of being paired with the nearest thing we found.
The strongest argument for this direction is what it refuses. A competitor can copy a paper texture in an afternoon. What they cannot copy is a product willing to put two documents side by side and say nothing — because their whole value proposition is the sentence we are refusing to write.