# ARENA gameplay and match experience research

## Product decision

ARENA should improve the full path from discovering an athlete to entering a round and understanding the result. Expanding a catalogue alone does not make its athletes selectable. The current implementation splits this path among Classic football, Founders cards and five-player Clubs. New accounts receive preselected players, and the leading navigation exposes the restricted modes before the global catalogue. These are observable workflow gaps, independent of visual taste.

The next implementation will make worldwide Fantasy and Ultimate the primary entry points. Each starts with empty positional slots. Fantasy selects from the supported global same-sport pool without collection ownership; Ultimate selects owned cards acquired with earned credits. Existing saved Classic, Founders, Clubs and trading records remain in their original systems, with clearly labelled access to them. Their rules will not silently change or their balances be combined.

## Comparative evidence

Sorare presents a compact loop of collecting players, selecting five-player lineups and earning rewards from real performances. Its officially licensed football, NBA and MLB coverage is a commercial arrangement, not permission for ARENA to copy its card assets or claim the same coverage. The reusable product idea is a visible selection-to-results loop with a comprehensible lineup, not its financial reward model. [Sorare football](https://sorare.com/football).

Sleeper combines scores with distinct fantasy and competition destinations across major North American sports. The reusable navigation principle is to keep the selected sport and game format obvious while the user changes tasks. The existence of social fantasy and score destinations means their combination cannot honestly be advertised as unique. [Sleeper](https://sleeper.com/).

The official FPL site is the reference for the preserved Premier League game. It is a competition-specific product; its restriction must remain explicit wherever ARENA keeps that legacy format. A worldwide entry point must not route users into an unlabeled Premier League-only catalogue. [Fantasy Premier League](https://fantasy.premierleague.com/en/).

The detailed research is provided in [competitor comparison](ARENA-COMPETITOR-RESEARCH-2026-09-10.md), [match-data and rights review](ARENA-MATCH-DATA-RESEARCH-2026-09-10.md), and [portrait/crest audit](ARENA-ASSET-RESEARCH-2026-09-10.md). This review covers a selected market sample and public documentation; it does not establish knowledge of every company, authenticated flow or private app feature.

## Interaction design

Nielsen Norman Group recommends keeping the important choices visible and making secondary options discoverable. Its treatment of staged disclosure also warns against separating controls that users need together. For ARENA, the roster, remaining positions and selection controls belong in the same workspace; advanced rules and receipt details can expand on demand. [Progressive Disclosure](https://www.nngroup.com/articles/progressive-disclosure/).

On phones, a filter tray should retain the context of the results rather than making users reconstruct their selection. ARENA will expose the active role, sport and competition restriction in the picker, with clear reset controls and results counts. [Mobile Faceted Search](https://www.nngroup.com/articles/mobile-faceted-search/).

Tap and keyboard replacement must work without drag-and-drop. WCAG 2.2 specifically requires a non-dragging alternative for applicable drag interactions; its minimum target-size criterion is 24 CSS pixels with exceptions. ARENA's regular selection controls will target at least 44 pixels for comfortable use. [Dragging Movements](https://www.w3.org/WAI/WCAG22/Understanding/dragging-movements.html), [Target Size](https://www.w3.org/WAI/WCAG22/Understanding/target-size-minimum.html).

| Before | After | Why |
| --- | --- | --- |
| A prefilled team implies decisions have already been made | Empty position slots and explicit player choices | The first team belongs to its creator |
| Selection and collection live in disconnected views | Choose a slot, inspect eligible athletes, replace and review in one flow | Users can understand the consequence before saving |
| Competition limits are implicit in the selected mode | Worldwide default; creator restrictions shown before joining and while picking | Membership rules are visible and enforceable |
| Last known data can resemble a completed round | Draft, locked, awaiting evidence and final are separate states | Missing results must not become zero points or invented ranks |
| A large mode banner pushes the useful controls down | Compact mode/sport header and roster-first workspace | The first viewport supports the task |
| Chat draws attention away from match events | Persistent score context and adjacent factual event feed | Conversation retains match context |

Emil Kowalski's design-engineering guidance informs short purposeful transitions, preserved image dimensions, clear pressed states and reduced-motion handling. The existing theme is retained; no claim that a new theme is guaranteed to outperform competitors is supported. [Design-engineering skills](https://github.com/emilkowalski/skills).

## Data and identity findings

The current global catalogue contains 29,686 people. The read-only asset audit found 7,234 stored photo URLs and 22,452 missing photos. Football accounts for most missing photos: 1,933 photo URLs among 22,887 footballers. This distinguishes source absence from rendering defects. A wrong face is worse than an explicitly unavailable portrait; exact source identities must govern enrichment.

The audit found crests on 980 of 991 distinct sport/league/team membership records. Therefore shared crest rendering and identity adapters are a useful immediate fix. These counts describe metadata availability, not a guarantee that every remote URL loads or carries unrestricted commercial permission.

The existing partner API supplies future prematch signals and prices. ARENA must not reuse its match IDs as ESPN match IDs or describe a cached price as current in-play odds. Matching must preserve competition, home/away identity and kickoff, reject ambiguity, and display absent or expired markets explicitly. All fixture surfaces should offer the same market-status affordance even when numerical odds are unavailable.

Existing structured match events can support ARENA-authored factual text, using supplied event type, participant, clock and score. That is distinct from copying broadcaster narration or generating unseen match action with AI. Disney's published terms cover ESPN and do not establish an unrestricted commercial scraping or AI-processing licence. [Disney Terms of Use](https://disneytermsofuse.com/english/).

## Differentiation hypotheses

The useful opportunity is a coherent evidence-aware experience: a global same-sport roster, clear creator scope, a pre-entry coverage check, and an inspectable final-score receipt. These are product hypotheses, not claims of invention. Their value should be tested by whether a new fan can build a valid lineup, replace a player, join a restricted league and explain their final points without outside help.

A squad readiness panel should explain incomplete roles, budget, competition scope and data limitations before entry. A match room can offer a short catch-up from supplied events while keeping chat visible. A corrected result should retain its previous value and explain the credit difference. These features distinguish ARENA's execution if they work reliably; they do not establish market superiority by themselves.

## AI connection and launch boundary

The current runtime has the existing prediction-provider secret and the free Signal Engine flag, but no OpenAI secret. The existing backend uses the Responses API, bounds input/output, authenticates users and requires approved source hosts. No key should be embedded in browser code, and generated analysis must remain separate from numerical forecasts. Structured formatting can improve output consistency but does not make unsupported claims true. [OpenAI Structured Outputs](https://developers.openai.com/api/docs/guides/structured-outputs).

This release must test creation, global selection, creator restrictions, substitutions, persistence, deadline races, final evidence, corrections and repeat reward requests. Browser tests should check the actual mobile and desktop flow. Controlled test fixtures can exercise a complete cycle without corrupting real saved users or claiming that an unfinished real-world match has finished.


## Implemented release and measured limits

The new Global Fantasy and Ultimate share a sport-specific full-squad flow: football 11 starters + 5 substitutes (4–3–3), basketball 5 + 5, hockey 6 + 4, baseball 10 + 5 (batting nine and a pitcher), and an explicitly labelled NFL skill-position format of 8 + 4. Native hockey and baseball positions are validated against supplied membership roles. The [lineup research](ARENA-NATIVE-LINEUP-RESEARCH-2026-09-10.md) explains the official sport rules and fantasy-specific bench choices. New teams and Ultimate collections are empty. The same global catalogue is searchable by player, club, country, position and competition. Creators can publish fixed competition restrictions before inviting members. Existing Classic, Founders and Clubs records stay accessible. New legacy Classic/Founders/Clubs teams also start without a selected lineup; existing grants and economics are retained.

Football starts with 360 earned-game credits: the entire 16-player squad costs 320, leaving 40. Each sport receives at least its full-squad signing cost plus one pack; there is no paid entry requirement. Starters score, substitutes replace only confirmed non-participants in compatible positions and frozen bench order. Unknown participation stays pending, unused bench points never count, and a substitute cannot be reused. Before lock, compatible starters and substitutes can be swapped.

Daily entries are independent by owner, league context and UTC day. A private league lasts 1, 7 or 28 daily rounds, supports up to 20 members, freezes picks at midnight UTC, and checks each submitted team's final evidence incrementally. Every manager's tournament total counts the same fully verified days. A fully elapsed empty day counts zero only after its final-check window; missing evidence on a submitted team never becomes zero. Only the public daily Ultimate result earns the bounded performance-credit award, preventing duplicate private-league reward farming. Corrections reconcile the net award and preserve actual committed evidence history.

The interface provides a four-step illustrated guide, entry receipts, creator-rule preview, role-aware replacement, owned-collection search, direct signing and a short reduced-motion-aware pack reveal. Guest drafts are retained in the current tab's session through sign-in; successful authenticated saves consume the guest record. Account export/deletion includes all new personal game records, and owner deletion transfers shared league organizer duties. Database-clock leases and atomic parent-entry guards prevent in-flight result checks from restoring deleted data.

Seventeen additional source portraits passed exact identity crosswalks and fresh availability checks. The catalogue now has 7,251 photo references among 29,686 athletes; 22,435 still lack source photos. Independently, 617 roster rows representing 495 athlete identities can now reuse photos already in the catalogue. These are distinct measures, not a claim that 634 new portraits were acquired. Team and match records now use that exact-ID fallback, and public squad lists actually render their available portraits. URL accessibility does not grant commercial image rights; no blocked Commons thumbnail was bypassed or shipped.

Fixture markets join only approved league identities, unique scoped team IDs, equal kickoff instants and scheduled future states. The existing partner supplies selected-outcome prematch quotes, not universal in-play prices. A row displays its recorded quote, unknown quote-update time or an explicit unavailable/unverified state. Core schedule and match pages share a batched source lookup and the existing partner’s D1 shared response cache is enabled in the publication configuration; FPL surfaces require a second exact schedule crosswalk. Non-covered international/gateway records remain labelled. Only the first 200 supplied league signals are inspected per batch; a missing row is not proof no odds exist.

Match chat now retains the score and supplied action context, with ARENA-authored factual event sentences generated from native structured records. There is no copied broadcaster narration or invented AI action. Browser polling inherits the current source delays and evidence labels. Mobile delivery is a responsive installable web-app shell with shortcuts and an honest offline screen; private records and live scores are never cached for offline replay. This is not an App Store or Play Store binary.

## Activation list

1. **OpenAI:** provide the secure location of the existing `OPENAI_API_KEY`, or set that secret on this Site. The three existing evidence-based brief endpoints (match, planner and prediction) are implemented and bounded; a real OpenAI request still requires this missing credential. The current model is `gpt-5-mini`. No new recurring subscription or trial was started.
2. **Global scoring:** a listed athlete is selectable where their existing membership is eligible, but complete real-time scoring for every world league requires a licensed/statistically complete source. Current gaps remain pending. Further connections need provider permission and credentials; they cannot be replaced with invented ratings.
3. **Commercial portraits and narration:** confirm the required display/reuse rights for provider-controlled images and any desired full editorial commentary. Openly reachable resources are not an all-sports commercial licence.
4. **Universal numerical odds:** the existing partner is football/prematch and often supplies a single outcome signal. Complete current three-way sets across every fixture and other sports require supplier coverage or an approved additional source. No paid feed was started.

## Competitive test after release

Measure a first-time user's successful empty-team completion, time to replace a player, proportion who correctly identify the league's scope, percentage who understand why a round is pending, and return visits after a match. Benchmark those tasks against the publicly accessible flows in the research; do not substitute a visual opinion for evidence of superiority. The strongest near-term differentiator is the inspectable path from global selection to source evidence and fair corrections, not unsupported claims to contain every competitor's feature.