Elite remains $50/month, coming soon. This build connects substantial practice, study and account features, but the advertised paid core still requires production, content, coach-quality and usability acceptance. Later expansions remain separate.
This inventory describes the current source build, not a production deployment or proof that the full roadmap is finished. “Preview available” means bounded functionality is implemented and has a route in this build. Target-host migration, independent review and production acceptance remain separate.
Free includes 100 total practice starts, including the first five guest hands. Pro remains $20/month. Elite remains $50/month, coming soon, with checkout disabled. No annual price or fourth public plan is approved; preserve explicit legacy pricing mappings and access to available reviews/exports at the practice cap.
Implemented connected preview
The connected preview now includes 4/6-seat PLO4/PLO5 practice, supported special formats, public decision capture/replay, account-backed study, deterministic coaching, explicit-range equity, supported imports, drills/retests, ICM and an account-backed session journal. These are bounded implementations, not a finished professional curriculum, validated model coach, full solver or hosted multiplayer product.
Notebook and journal D1 APIs use revisions, retry identifiers, local recovery and explicit conflict review. Migrations 0005 and 0006 must be applied and verified on the target host. Local working changes are not automatically confirmed account saves. The practice table cannot resume a live hand after reload; replay balances remain fixed at the saved decision.
Model-quality gate remains open: all eight local baseline evaluation cases timed out, and a separate diagnostic produced a hallucination. Deterministic calculator facts and authored reasoning remain usable previews; the optional model adapter must stay disabled until the actual provider passes independent multi-turn evaluation. No solver or model-quality claim follows from the current tests.
Acceptance-led sequence
Work is ordered by acceptance, not calendar promises. Optional private tables, automatic mixed-game rotation, custom fitted opponents and competitions can ship separately from a complete and honestly advertised core. Arbitrary Omaha solving needs separate feasibility and reference validation. The marketplace and unapproved annual pricing remain parked.
Make the platform dependable
Verify target-host accounts, study/journal migrations, practice limits, recovery and payment operations; reconcile legacy Hold’em QA.
Acceptance: Guest-to-account, multi-device recovery, access isolation and test-mode billing pass.
Connect the full learning loop
Validate the implemented Omaha → saved decision → deterministic coach/calculations → review → drills/retests loop, then close content and operational gaps.
Acceptance: The advertised launch bundle works end to end and passes independent poker and usability review.
Add the other poker products
Harden preview ICM, journal and live-format tools; add private tables, mixed-game rotation, custom profiles and separately gated competitions.
Acceptance: Each module ships only after its own acceptance tests and operational setup.
Earn deeper solver coverage
Benchmark a bounded Omaha reference/solve service and supported node locking.
Acceptance: Reference agreement, licenses, tree matching, costs and response times are measured.
Keep optional businesses separate
Coaching marketplace and any unapproved annual pricing remain outside the launch commitment.
Acceptance: Explicit product decision before work or customer claims.
Paid-release gates
A complete advertised bundle — Every feature sold as included has a working route, authentic data, an owner and a passing end-to-end test. Unfinished expansions stay clearly planned.
Correct poker and honest advice — Independent rule/math fixtures pass. Exact calculation, sampled equity, authored strategy, exploit assumptions and matching solver outputs stay visibly distinct.
A durable account — Study records survive device changes and recovery. The same account owns play, imports, drills and billing. Export and deletion are tested.
A coach that keeps the facts — Validate deterministic coaching with poker reviewers. Keep model explanations disabled until the actual provider passes held-out multi-turn factual, quality, privacy, latency and failure tests; local timeouts and hallucination diagnostics are not acceptable release evidence.
A working subscription lifecycle — Purchase, renewal, failed payment, cancellation and downgrade are verified in test mode before live rollout. Elite checkout remains disabled until release approval.
A usable learning product — Representative players complete play → save → understand → drill → retest on desktop and mobile. Improvement claims require measured unseen transfer, not only quiz repetition.
R01 / Play / Preview available / Foundation
Deep live-table practice
Keep the core Hold’em experience and make its decisions easier to study.
Current implementation
The existing nine-seat Hold’em simulator, selectable stakes, multi-table practice, reviews and analytics remain in the site. Elite changes do not certify older Hold’em findings resolved.
Full scope
Cash display by default with BB toggle; position labels, visible bets, legal sizing, pre-fold, speed controls, top-ups and session restarts.
Opponent notes, public hand memory, similar-spot search, against-you versus whole-table views, rare tilt and occasional player turnover.
After-hand recap, available showdown cards, hand names and replay; concealed cards stay unknown to the live decision model.
Next work
Reconcile legacy Hold’em findings against the release candidate and verify mobile play, session continuity and deployed account limits.
Ready when
Independent action-order, all-in, side-pot, chip-conservation and hidden-information tests pass; mobile play and session continuity pass.
Plan allocation: Free includes 100 total hands; Pro continues Hold’em play. Elite would inherit Pro.
Dependencies: R02
Evidence and limits
app/play, public/table and existing Hold’em tests; the latest Elite tests are not a substitute for a deployed Hold’em regression.
One account for play, review and study across devices.
Current implementation
Practice starts are account-bound. Study notebook and session journal now have authenticated D1 APIs with revision checks, idempotent retry, local recovery and explicit conflict choices; guest work is imported explicitly.
Full scope
First 5 guest hands count toward 100 total Free practice hands; login after 5; new sessions do not reset the allowance.
Merge guest work once; sync decisions, drill attempts, notes and preferences; export, deletion, backups and recovery.
Account isolation, concurrent-table limits, audit logs and resilient save/retry flows.
Next work
Apply migrations 0005 and 0006 in the chosen target environment; verify two-device recovery, backup/restore, export/deletion and production identity flows.
Ready when
Two devices recover the same work; 101st Free start is denied; concurrent requests cannot overspend; cross-account reads and writes fail.
Plan allocation: Account ownership and export apply to every plan. Elite study sync is part of the planned Elite core.
Dependencies: None
Evidence and limits
app/api/elite/study and journal; lib/elite/study-sync.ts and journal-sync.ts; drizzle/0005_glossy_umar.sql and 0006_wise_blindfold.sql. Actual-handler/client tests establish implementation, not target-host rollout.
Learn Omaha decisions at a table that remembers public play.
Current implementation
PLO4/PLO5 practice supports 4 or 6 seats, $1/$2 through $10/$25, 200–1000 BB hero buy-ins, cash/BB display and heterogeneous opponent stacks. Saves preserve a public decision while the live hand stays paused.
Full scope
Configurable supported seats, unequal deep stacks, selectable stakes and clear pot-limit min/max sizing.
Persisted opponents, safe replacement seats, shared practice quota, save a frozen decision and return to the same hand.
Study multiple active opponents and eligible pots without treating heads-up advice as a multiway solution.
Next work
Complete deployed browser/mobile acceptance and durable full-hand recovery. Keep 4/6-seat and 200–1000 BB limits explicit; reload still does not resume a live hand.
Ready when
PLO4/PLO5 cases at 2, 3 and 6 players and 50/100/250/500 BB conserve chips and obey pot limits, short-raise reopening and deck bounds.
Plan allocation: Omaha is planned for Elite. Current preview access is not a promise of permanent Free access.
Dependencies: R02
Evidence and limits
components/elite-table.tsx, lib/elite/table-session.ts and tests/elite-table-session.test.cjs; 144 complete-hand configuration checks plus six-seat snapshot/coach compatibility. No real-money or multiplayer claim.
Practice the live formats where split pots and nut potential matter.
Current implementation
Table setup now exposes single/double boards, a 2-BB UTG straddle, 5-BB-per-seat bomb pots and eligible all-in run-it-twice. Run-it-twice is disabled for double-board hands; fractional play-chip awards are disclosed.
Full scope
Explicit ante and straddle rules; button and action order; run-it-twice only when eligible.
Board-by-board equity, eligible side pots, odd-chip rules, quartering and scoop explanations.
Scenario drills for strong-on-one-board and weak-on-the-other decisions.
Next work
Add format-specific drills and validated board-by-board range equity/side-pot analysis; verify every exposed setup and result in browser acceptance.
Ready when
Every supported format passes independent multiway all-in and split-pot fixtures; no duplicated runout cards or lost chips.
Plan allocation: Advanced Omaha formats are demonstrated in the preview. Their paid inclusion remains part of the approved Elite launch-scope decision.
Dependencies: R03, R06
Evidence and limits
Table settings and unchanged Omaha engine; four/six-seat special-format fixtures cover all supported stakes and both variants. The separate equity calculator models one equal-eligibility pot.
Make reads from what players actually showed and did.
Current implementation
Stable opponent identities retain per-observer public reads, including bot-versus-bot actions. New seats have random styles/depths; busted and occasional departing opponents are replaced with fresh identities.
Full scope
Separate skill from looseness and aggression; eight Omaha archetypes, noisy decisions and occasional bounded emotional shifts.
Explain which public examples changed a read; replacements get new identities; uncertainty persists when cards were not shown.
Next work
Calibrate on held-out scenarios and compare adaptation on/off under matched seeds. Review changes by street, size and opponent rather than equating aggregate statistics with realism.
Ready when
Unseen cards cannot change another player’s policy; bot A changes appropriately after B’s public pattern; replacement IDs never inherit the prior occupant’s reads.
Plan allocation: Hold’em table reads remain in the existing plans; Omaha adaptation is planned for Elite.
Dependencies: R01, R03
Evidence and limits
public/elite/omaha-bots.js and table-session tests; bounded turnover is authored behavior, not measured human psychology. Continuing players retain public memories; replacements do not inherit the departed observer state.
See exactly which cards make your hand—and what could beat it.
Current implementation
Studio highlights the legal two-hole/three-board hand. The workbench reader enumerates compatible better two-card combinations, respects entered dead cards and distinguishes combination counts from range probabilities.
Full scope
Made-hand names and selected cards for every board; nut versus non-nut distinctions and redraw explanations.
What-beats-you drawer, blockers, backdoor paths and clean versus dirty outs where supported by validated enumeration.
Distinguish combinatorial possibilities from opponent-range probabilities.
Next work
Connect richer reader views directly to saved table decisions and review clean/dirty outs, backdoors and redraw explanations before expanding their claims.
Ready when
Independent PLO4/PLO5 scorer fixtures agree; card-removal, double-board and four-to-a-flush traps pass; no pair-count statistic is mislabeled as range equity.
Plan allocation: Keep existing basic hand labels. The richer Omaha reader is planned for Elite.
Dependencies: R03
Evidence and limits
components/elite-tools.tsx, public/elite/research-worker.js and omaha-coach.js; independent dead-card regression. Reader output describes the current board, not future win probability.
Follow a range through position, actions and board texture.
Current implementation
The workbench accepts explicit weighted Omaha hands or uniform random ranges for up to three opponents, removes blocked combinations and exports reproducible inputs/results. A chart library and graphical feature-filtered range explorer are not implemented.
Full scope
Filter Omaha hands by rank groups, suits, connectivity, blockers and made-hand/draw categories.
Compare position and action-conditioned ranges; show available combinations, weights, source, depth, rake and tree.
Import/export supported range formats and turn a selected subset into a drill; basic range tools stay free.
Next work
Add versioned authored or licensed reference ranges, combination filters and drill links. Label authored assumptions separately from matched solved references.
Ready when
Weights and card removal reconcile; actions narrow ranges without inventing frequencies; unmatched depth or multiway trees are clearly unsupported.
Plan allocation: Basic Hold’em range tools remain free. Advanced Omaha range exploration is planned for Elite.
Dependencies: R06
Evidence and limits
public/elite/range-equity.js and components/elite-tools.tsx; explicit user-entered ranges are assumptions. No licensed Omaha chart collection or action-conditioned range inference is supplied.
Test whether a read changes a decision, not just the equity number.
Current implementation
Range equity now uses bounded exact enumeration or collision-conditioned joint Monte Carlo with tie shares and sampling intervals. Studio retains terminal heads-up EV sensitivity, separating fold probability from equity when called.
Full scope
Direct price, eligible pot, effective stacks, SPR, blockers and who is still to act.
Fold/call/raise size comparisons with conditional cashflows, uncertainty bands and break-even assumptions.
Future betting, implied/reverse-implied odds and multiway branches only under an explicit validated model; no arbitrary future-value bonus.
Equity against specified ranges uses joint legal deals, card removal, split-pot shares and board-by-board accounting. Show exact versus sampled mode, seed/sample count and uncertainty; equity is separate from EV.
Next work
Review boundary-case uncertainty displays and add explicit validated multi-street, multiway and side-pot models before extending the terminal EV tool.
Ready when
Independent cashflow fixtures match; sunk costs and uncalled returns are treated consistently; no double-counting fold equity as showdown equity. Equity fixtures include ties, blocked combinations and correlated multiway deals; repeated-seed coverage is measured.
Plan allocation: Existing Pro live-spot tools stay in Pro. Deeper Omaha equity and exploit modeling is planned for Elite.
Dependencies: R06, R07
Evidence and limits
lib/elite/study.ts, public/elite/range-equity.js and independent combinatorial/cashflow tests. Equity supports one board/equal eligibility, no rake or future betting; sampling intervals exclude uncertainty in ranges.
Ask follow-up questions about the exact decision and the assumptions behind it.
Current implementation
A connected deterministic coach accepts saved/manual/imported decisions, bounded follow-ups and public calculation facts. An optional provider adapter exists but model coaching is not release-quality: all eight local baseline cases timed out, and a separate diagnostic produced a hallucination.
Full scope
Freeze known cards, legal actions, pot, stacks, reads and action prefix; ask a useful clarification when an essential fact is missing.
Separate computed facts, theory-informed guidance, matched solver references and conditional exploits; compare alternative sizes.
Retain five-turn context, label hypothetical changes, support a deterministic fallback, limits and clear service errors.
Next work
Keep model-backed coaching disabled until an actual provider passes held-out multi-turn poker review, latency, privacy and failure tests. Validate the deterministic core with educators and representative users.
Ready when
50 held-out spots × 5 turns have zero critical card, pot, hidden-card or invented-solve errors; report fallback rate and latency separately from answer quality.
Plan allocation: Current authored Hold’em chat remains in its existing plans. Grounded conversational Elite coaching and its message allowance must be verified before sale.
Dependencies: R02, R06, R08
Evidence and limits
components/elite-coach.tsx, lib/elite/coach.ts and coach-provider.ts, coach tests and scripts/qa/coach-local-eval.cjs. Authored reasoning and calculator facts are not solver output; the local evaluation does not support a model-quality claim.
Manual live-hand builder for players, positions, stacks, blinds, bets and street-by-street actions.
Tag game source and session; separate actual winnings, imported facts and reconstructed assumptions.
Next work
Build the manual legal-action reconstruction flow and broaden supported history fixtures. Preserve unknown cards, parser warnings and source units; do not imply every site/format is supported.
Ready when
12 expert-specified hands reconstruct with reconciled pots and stacks; malformed or unsupported formats fail visibly; duplicate imports do not duplicate results.
Plan allocation: Existing Hold’em imports remain available as configured. Deeper Omaha import and reconstruction is planned for Elite.
Dependencies: R02, R03
Evidence and limits
public/elite/hh-import.js and review.js, research worker and import UI; independent cashflow/future-information/dedup/currency tests. Multi-board, bomb-pot and non-USD imports are rejected.
Move from a saved hand to the exact choice you want to understand.
Current implementation
Table and supported imported decisions connect to the notebook and coach. Captured table replay steps through public actions/known boards; balances are explicitly fixed at the saved decision. Authenticated D1 notebook sync has conflict review and local recovery.
Full scope
Replay only information available at the chosen street, with player identity, style, pot, stacks and bets.
Save questions, reasoning, tags and confidence; link decisions to drills, retests and a coach conversation.
Review an opponent’s public line and similar spots against you or the whole table; explicit post-hand reveal mode stays separate.
Next work
Finish historical stack-by-action replay, full-hand resume and cross-device acceptance on the target host. Unify tags, reasoning and imported/simulated review without fabricating missing facts.
Ready when
20 decision round-trips preserve exact facts and IDs; later boards or hidden cards never contaminate earlier advice; replay and export agree.
Plan allocation: Keep Free latest-100 review and Pro extended history. Elite adds the connected Omaha study notebook and sync.
Dependencies: R02, R10
Evidence and limits
components/elite-decision-context.tsx, elite-studio.tsx, study API/sync and table/coach tests. Snapshot text is bounded; oversized captures reject without truncation. Working local changes are not silently treated as confirmed cloud saves.
Turn a repeated pattern into a specific study task.
Current implementation
Imported Omaha review now shows opportunity-based patterns, example-hand links and a dated study plan. Existing Hold’em progress views remain separate; results are review prompts, not verified strategic errors.
Full scope
Filter by variant, source, stake, position, street, pot type, size, SPR and player count; click any metric to its contributing hands.
VPIP, PFR, 3-bet, squeeze, c-bet, aggression, check-raise, folds by opportunity, showdown, sizing and result trends.
Keep last-100 style labels separate from last-1,000 decision scores; unknown grades stay unknown; variance is not automatically bad play.
Next work
Unify variant/source filters, simulation/import evidence and drill outcomes. Add full metric-to-hand drill-down and independent denominator checks before claiming a complete improvement dashboard.
Ready when
Every reported percentage has the right opportunity denominator; imported/simulated outcomes stay separate; no winnings-only or repeated-quiz metric is presented as skill.
Plan allocation: Existing Hold’em Pro analytics stay in Pro. Unified Omaha reporting and its drill recommendations are planned Elite additions.
Dependencies: R10, R11
Evidence and limits
public/elite/review.js and workbench import UI; existing progress dashboard is separate. No winnings-only skill score, validated EV-loss grade or full Omaha statistics dashboard is claimed.
Know what to work on next, then test whether it stuck.
Current implementation
42 authored exercises, confidence records, first/repeat tracking, goal/time sessions and due reviews remain connected to saved decisions. Notebook decisions and attempts can be saved to the authenticated account with explicit conflict handling.
Full scope
Reviewed curricula for PLO4/PLO5, preflop pressure, nuts/redraws, blockers, sizing, deep/multiway play and uncertain reads.
Prerequisites, saved drills, pause/resume, seven-day retests and unseen transfer questions.
A weekly plan tied to actual observed opportunities; optional narrated explanations use the same reviewed facts.
Next work
Expand the independently reviewed curriculum and unseen transfer sets; verify due queues across devices. Local due dates are not notifications or an automatic cross-device scheduler.
Ready when
Every answer has author/reviewer/version and basis; repeats never count as unseen; 12 scenario families include 3 held-out transfer variants each.
Plan allocation: Existing Pro study tools stay in Pro. The reviewed Omaha curriculum and connected retests are planned for Elite.
Dependencies: R02, R11, R12
Evidence and limits
lib/elite/study.ts, elite-studio.tsx and study sync tests; exercise arithmetic checks do not establish expert review, retention or a full professional curriculum.
Keep real-session results, decisions and context together.
Current implementation
A session journal now records/edit/deletes USD buy-ins, cash-outs, hours, stakes and notes, with stake filters, cumulative results and backup import/export. D1 account save/load includes revisions, conflicts and local recovery.
Full scope
Buy-ins, cash-outs, top-ups, currency, stakes, duration, venue and personal notes with correction history.
Cash and BB results by session, source and stake, plus hourly trends and sample-size limits.
Link a difficult live hand to reconstruction and the next practice session.
Next work
Verify journal migration and two-device conflicts/recovery on staging; add richer session-to-hand links and any approved currency conversion/correction history without mixing training chips into cash results.
Ready when
Rebuys and stake changes reconcile; hourly rates use valid durations; edits recompute reports; small samples do not establish skill.
Plan allocation: The journal is allocated to Pro and inherited by Elite; account-backed implementation is present, with production rollout still gated.
Dependencies: R02, R10
Evidence and limits
components/elite-tools.tsx, lib/elite/research.ts, journal API/sync and migration 0006; journal-sync tests cover actual handlers/client behavior. Edits save locally until Save to account confirms them.
Compare chip value with prize-equity consequences in supported spots.
Current implementation
The tournament workbench exposes bounded 2–12-player Malmuth–Harville ICM and one-caller full-effective-stack shove branches, with explicit dead-pot treatment, no-tie assumptions and equal-payout indifference.
Full scope
Input payouts, stacks and dead pot; compare chip EV, ICM equity, bubble factor and required equity.
Save final-table scenarios and state the treatment of eliminated players and ties.
No bounty, arbitrary postflop tournament solve or future-game-simulation claim until separately implemented.
Next work
Add saved tournament scenarios and reviewed edge-case coverage; keep bounties, multiple-elimination ordering, future-game simulation and tournament solving out of the offer until implemented.
Ready when
Published hand-verifiable examples agree; dead-pot conventions are explicit; unsupported multiple-elimination order and excessive jobs are rejected.
Plan allocation: The bounded ICM tool is an Elite preview. Final paid inclusion and limits must be approved before release.
Dependencies: R02, R20
Evidence and limits
lib/elite/research.ts and tournament UI; independent finish-order permutation and winner-take-all cashflow oracles. Model prize equity is not a solved push/fold strategy.
The standalone server-authoritative private-table API and signed-ledger prototype remain separate from the browser practice table; their in-memory stores are not production durability.
Full scope
Account-bound seats and invite permissions; reconnect, turn ownership, timeout policy and private/public card views.
Practice-only chips, replayable shared sessions, durable table state and clear spectator permissions.
Server-owned randomness and verifiable deal records without exposing concealed cards during play.
Next work
Build the authenticated lobby, durable table/ledger adapter, reconnect and spectator rules; adversarially test concurrent actions and restart recovery.
Ready when
Forged seats, actions and records fail; no private cards leak; server restarts preserve a recoverable hand; parallel clients cannot double-act.
Plan allocation: Private tables are planned for Elite. Table/seat limits will be confirmed before release; no real-money games.
Dependencies: R02, R03, R20
Evidence and limits
Enhanced engine src/table-server.js; no hosted private-table lobby or durable multiplayer deployment is delivered by the browser previews.
R17 / Formats / Integration needed / Expansion
Mixed-game rotation
Switch between supported Hold’em and Omaha formats in one practice session.
Current implementation
An orbit-based mixed-game scaffold exists in the engine package. The website supports manual between-hand PLO4/PLO5 selection, not automatic Hold’em/Omaha rotation.
Full scope
Choose an explicit rotation, seats, stakes and bomb-pot schedule; show the next format before the deal.
Preserve identity, stack accounting and variant-specific public memories through transitions.
Keep each game’s legal actions, deck constraints and statistics distinct.
Next work
Connect tested per-variant adapters and an explicit rotation lobby, preserving separate game rules, state and public-model contexts.
Ready when
Orbit boundaries and replacements rotate correctly; no stack or card state leaks between formats; per-format reports reconcile.
Plan allocation: Mixed-game rotation is an Elite expansion after its adapters and formats pass QA.
Dependencies: R01, R03, R04, R16
Evidence and limits
Enhanced engine src/mixed.js; no automatic orbit rotation UI or integrated mixed-game history exists.
R18 / Study / Planned / Research
Solver and reference library
Use an identified solution when the spot actually matches.
Current implementation
The existing bounded Hold’em lab remains separate. New Omaha calculators use explicit inputs; no PLO solution/reference service, arbitrary deep multiway solver or licensed chart library is integrated.
Full scope
Provenance for every reference: engine/version, range source, rake, depths, tree, sizes, run settings and accuracy evidence.
A matching check before showing frequencies or EV loss; node locking and trainable profiles only inside supported solved trees.
A bounded server-job pilot with licensed or compatible open-source logic; label approximation and unsupported spots clearly.
Next work
Benchmark sources, licenses, matching rules and bounded compute before committing to solver coverage. Core can use reviewed authored assumptions and exact arithmetic with honest labels.
Ready when
Independent reference fixtures match; unsupported trees fail clearly; resource/cancellation limits hold; expert review confirms labels and recommendations.
Plan allocation: Keep the current bounded Hold’em lab free. Future Omaha solve coverage and compute allowances are a separate Elite release decision.
Dependencies: R07, R08, R20
Evidence and limits
Existing /solver and /solver-lab routes; range-equity and ICM calculations are not solved strategy references. No model output is evidence of a solve.
Practice against a tendency profile without claiming to clone a person.
Current implementation
The standalone aggregate-profile fitter remains an uncertain tendency prototype. The live preview offers randomized authored archetypes but no user-built or fitted opponent profile editor.
Full scope
Fit supported public opportunity statistics, with small-sample shrinkage and explicit unknown traits.
Separate observation-based estimates from learner-entered assumptions; compare the same scenario under different profiles.
Link an exploit adjustment to evidence and show what would invalidate it.
Next work
Build a profile editor, held-out conditional fit checks and evidence disclosures; avoid sharing real-player identity or private history in presets.
Ready when
Fit error and sample counts are visible; aggregate agreement is never claimed to reproduce a person’s conditional strategy.
Plan allocation: Custom tendency profiles are planned for Elite. They are not guaranteed replicas of real people.
Dependencies: R05, R07, R12
Evidence and limits
Enhanced engine src/clone.js; randomized table styles do not constitute opponent cloning or calibrated conditional strategy.
R20 / Platform / Integration needed / Foundation
Reliable compute and subscription service
Make paid access, saved work and expensive jobs dependable.
Current implementation
Account-bound practice, study and journal persistence, entitlement guards, bounded research workers and provider adapter limits now exist. Production payment lifecycle, host recovery and model quality are still release gates.
Full scope
Free $0, Pro $20/month, Elite $50/month coming soon; no fourth public tier or silent legacy repricing.
Signed webhooks, durable event inbox, renew/fail/grace/cancel/downgrade, account recovery and verified entitlements.
Bounded coach/solver jobs, worker isolation, actual cancellation, rate limits, retries, observability, backups and rollback.
Next work
Verify target-host migrations/bindings, backup/rollback and production-equivalent test-mode billing. Keep Elite checkout and unvalidated model service disabled; isolate any future CPU jobs.
Ready when
Payment lifecycle and stale/repeated events pass; process restart preserves access; per-plan limits hold; CPU jobs can be stopped without stalling the service.
Plan allocation: The three public plans remain Free, $20 Pro and $50 Elite coming soon. No annual offer or separate league plan is approved here.
Dependencies: R02
Evidence and limits
Site APIs, D1 migrations 0005/0006, browser worker termination, enhanced billing/job modules and access tests. Local tests are not live Stripe, provider reliability or disaster-recovery evidence.
R21 / Community / Integration needed / Expansion
Leagues and competition tools
Make rankings auditable before adding rewards.
Current implementation
Provisional standings code and a leaderboard route remain; prize entry, awards and automated fulfillment are not active. New study tools do not certify competition integrity.
Full scope
Server-verified records, season rules, minimum eligible sample, excluded bots, suspicious-pattern review and appeals.
Reproducible standings with ties, reversals and corrections; distinguish result rank from measured decision quality.
Competition is under consideration and not open. No entries, awards, dates, funding formula or eligibility have been announced. Practice subscriptions do not include competition entry.
Next work
Resolve eligibility, ranking, ties, appeals, funding and reward rules; complete authoritative-ledger abuse tests and separate provider/jurisdiction review before prize activation.
Ready when
Standings reproduce from authoritative records; forged/replayed records and collusion scenarios are handled; published rules, funding and fulfillment are approved.
Plan allocation: Competition eligibility and its subscription allocation are undecided under the final three-plan lineup. Do not advertise prize access as included.
Dependencies: R02, R16, R20
Evidence and limits
Enhanced engine src/league.js and /leaderboard; no claim of production collusion resistance or approved reward operation. Core Elite need not wait for competitions unless sold as included.
Let players report problems and follow improvements.
Current implementation
Account-bound community submissions/voting and roadmap links exist alongside the expanded study previews. Production moderation and issue-to-release operations still need acceptance.
Full scope
Bug and suggestion categories, duplicate handling, one vote per account and status updates.
Attach a redacted decision reference so a problem can be reproduced without sharing private history.
Link release notes and roadmap updates to the reported issue.
Next work
Verify moderation, permissions and pagination on the target host; connect redacted repro cases and release notes to resolved issues.
Ready when
Ownership, voting, moderation and pagination tests pass on the deployed host; public posts contain no private account data.
Plan allocation: Community and roadmap feedback remain available to all plans; contributions require the applicable account permissions.
Dependencies: R02
Evidence and limits
components/community-board.tsx, community tests and /roadmap. Public feedback must not expose private imported hands or account notebooks.