When Every Bet Has an Edge, Nothing Has an Edge

Epistemic status: A record of design hypotheses developed during a rapid Dimers product workshop. Some conclusions are embodied in a working prototype; none should be mistaken for validated user research. Confidence is highest in the information-design principles and lower in the particular visual executions.

Summary

We began with what looked like a small interface problem: Dimers displayed sportsbook odds, model probability and Edge as three equal cards, and the relationship between them was difficult to see.

The work eventually produced a broader claim:

A useful prediction product should not merely deliver an answer. It should direct attention, expose the comparison that produced the answer and leave enough work for the user to feel responsible for the decision.

This changed the target mental model from “show me the best bet” to “show me what deserves attention, then help me decide.” The distinction sounds subtle. It affected almost every subsequent decision about hierarchy, interaction, movement, urgency and colour.

The apparent problem: three numbers

The starting interface placed three values beside one another:

1. The sportsbook price.
2. Dimers’ estimated probability.
3. Edge.

This is technically complete but cognitively awkward. Odds and probability are different languages, while Edge is derived from the relationship between them. Giving all three equal visual weight disguises this structure. It asks the user to perform the conversion and infer the comparison before they can understand the conclusion.

Our first hypothesis was therefore simple:

Edge is not a third fact. It is the visible distance between two estimates of the same event.

Once FanDuel’s price is converted to implied probability, the comparison becomes:

FanDuel implied probability → Edge → Dimers model probability

This was the generative idea behind most of the early concepts: equations, aligned bars, a probability gap, a pill between two endpoints and eventually a match-up in which sportsbook and model resembled opposing teams with Edge as the score.

Exploration generated options—and hid the real question

We built a React workshop capable of showing many concepts at once and adjusting spacing, scale, hierarchy and styling. This was useful. It prevented us from prematurely polishing one answer and made recurring patterns easy to spot.

The relationship also seemed capable of becoming a brand device. “The difference” was tested across social, editorial, partnership, out-of-home and motion applications. A short Remotion explainer showed the device forming from market and model values. We created a reverse brief, translated the source brief into Jobs to Be Done and produced job-map and Forces of Progress visuals.

All of this work increased the number of available ideas. It did not necessarily increase confidence that we were solving the right product problem.

The client session made that obvious. The conversation went sideways because the exploration had moved too far from the current product. We had many plausible futures but no stable counterfactual. It was hard to answer the most useful question: is this specific change better than what exists now?

That failure produced the most important methodological update of the project:

In a fast product workshop, fidelity to the current experience is not conservatism. It is experimental control.

The reset: preserve the baseline

We rebuilt the two existing mobile screens side by side on a plain white canvas:

• The opening bets feed.
• The bet detail screen.

They had fixed widths, independent scrolling and functional controls. A Baseline / Version 1 switcher preserved scroll position so the two versions could be compared without the page jumping. Small variations moved into a right-hand control panel rather than becoming additional versions.

This changed the prototype from a concept gallery into an instrument. We could now vary one thing at a time: show or hide metrics, change the comparison module, remove backgrounds, alter Edge shapes, increase bar height or switch signal states.

The reset also forced us to articulate what each screen was actually for.

The first screen should allocate attention, not outcomes

The existing logic risked presenting a feed in which every item had Edge. That seems desirable if Edge is the product’s value proposition. It also destroys the signal.

If every restaurant is “recommended,” recommendation becomes a category label. If every alert is severe, severity stops allocating attention. Likewise, if every bet has Edge, the user cannot experience discovery; the app has converted an exceptional state into background texture.

The opening screen therefore received a different job:

Curate a small set of games and props worth checking now. Do not imply that all of them are good bets.

This led to three principles:

1. Scarcity preserves information. Include items with and without Edge.
2. Progressive disclosure preserves attention. The first screen should be scannable; analysis belongs after the user opens an item.
3. Time creates a reason to return. New discoveries and closing windows should be visible when they are real, not manufactured as generic urgency.

This was not an argument for hiding information. It was an argument for sequencing it.

The second screen should preserve agency

The detail screen created a different problem. If Dimers possesses a strong answer, why not simply provide it?

The practical answer came from a conversation about using the app at the pub. A user who blindly follows a recommendation has an asymmetric attribution pattern:

• If the bet loses, the app chose badly.
• If the bet wins, the user still feels clever.

The product absorbs the downside without reliably receiving credit for the upside.

We therefore hypothesised that a small amount of meaningful effort could improve commitment. Switching a timeframe, comparing market and model, refreshing the estimate or inspecting recent performance asks the user to participate in the decision. The interaction does not need to be difficult. It needs to be causally connected to judgement.

The detail-screen principle became:

If the system has an answer, show the evidence that makes the answer available without removing the user’s responsibility for reaching it.

This is the difference between an assistive tool and an autopilot. It may also be the difference between a user blaming the product and a user learning to reason with it. That claim remains unvalidated, but it is specific enough to test.

Three visual modules, three explanatory models

Version 1 tested three ways to explain the same market–model relationship.

1. Refined cards

This retained the familiar three-card layout while clarifying authorship and movement. It had the lowest learning cost but continued to treat the values as somewhat independent objects.

2. The Edge bar

FanDuel and Dimers occupied one probability scale. Both boring extremes—towards zero and one hundred—were compressed so the middle difference had enough room to read. The bar made the mathematical relationship explicit, although scale compression introduces a trust risk if it is not legible.

3. The match-up

FanDuel appeared on the left, Dimers on the right and Edge sat between them as a smaller score-like device. This was less chart-like and more immediately narrative: one estimate versus another. It also connected most naturally to the earlier brand exploration.

These should not be treated as decorative variants. Each one teaches a different model of the system. Cards say “three facts.” The bar says “one measured gap.” The match-up says “two competing estimates.” Choosing among them should ultimately depend on comprehension, not preference.

Movement is not value

One of the prototype’s smallest changes carried a large semantic correction. The original cards used small green arrows for movement. Green implied improvement, but a change in odds or probability is not inherently good. Its meaning depends on which value moved, in which direction, over what period and relative to what other estimate.

We replaced generic positive colour with neutral direction, magnitude and timeframe. Edge movement remained a percentage. The movement chart was rebuilt around the same language as the comparison module:

• FanDuel is blue.
• Dimers is yellow.
• The space between them is Edge.
• History reaches a marked “Now.”
• Projection continues beyond it as a trend with a high–low range.

This made uncertainty visible instead of allowing the projected line to impersonate observed fact.

Urgency should describe danger, not excitement

Our first urgency experiments added separate pills for “new,” “ending soon” and Edge. They increased the number of signals without making the decision clearer.

The better metaphor came from retail: a solid red label works because the danger is losing access to an opportunity. Red does not mean the bet is good. It means the current relationship may not remain available.

The resulting Edge window combines two claims:

1. A meaningful Edge exists now.
2. There is a genuine reason to believe the window is closing.

If either claim is absent, the urgency treatment should be absent. This constraint matters because false scarcity may increase short-term clicks while degrading the trust required for a prediction product.

Colour needed an ontology

During exploration, colours accumulated meanings: positive, model, movement, Edge, severity and brand. The collisions made interpretation unstable. We eventually assigned each colour one job:

• Blue: FanDuel and market information.
• Yellow: Dimers model information.
• Green: the historically preferable Sweet Spot.
• Orange: unusually high Edge or caution.
• Red: a time-sensitive Edge window.
• Grey: neutral movement and supporting information.
• Cyan: recent trend evidence.

The strongest rule was reserving yellow for Dimers. Anything the model says can be recognised as model output. “Dimers,” “good” and “moving up” are no longer encoded as the same thing.

What changed in our model of the product

The project produced six linked updates:

The most consequential shift is from answer to comparison. Once the product is understood as a comparison tool, Edge becomes an explanation rather than a badge, movement becomes evidence rather than decoration and interaction becomes part of judgement rather than friction.

What we believe—and what we do not yet know

We now have relatively high confidence in four propositions:

1. Market and model need explicit authorship.
2. Edge should be explained as their difference.
3. Movement needs direction, magnitude and timeframe without automatic positive colour.
4. The opening and detail experiences should disclose different amounts of information.

Confidence is lower in the particular visual solution. We do not yet know whether users understand a compressed probability scale, whether the match-up metaphor creates false competitiveness, whether the working Sweet Spot applies across bet types or how much interaction produces agency before it becomes effort.

Those are useful uncertainties. They can be tested independently because the prototype preserves a baseline and exposes the variables as controls.

Conclusion

The project started by trying to improve three cards. It ended by reconsidering what kind of promise the product should make.

A weak promise is: we will tell you which bet to make. It creates dependence, compresses uncertainty and gives the product ownership of failure.

A stronger promise may be: we will show you where the market and our model disagree, why the disagreement matters now and what evidence should inform your decision.

The durable design rule is therefore:

Show the market and model honestly. Make the difference understandable. Leave the decision with the user.

That rule is more valuable than any individual card, bar or ribbon because it can survive the next interface, the next feature and the next iteration of the model.