Backend implementation checklist
This is an audit of the current Python backend, not a release schedule. The main dashboard modules already have backend services. Most remaining work extends those services or connects them into complete workflows. Optional product expansions are identified separately.
Audited snapshot — October 29, 2026
The dashboard’s Coming soon → Implementation status reads the same audited inventory through GET /api/v1/roadmap. This is a capability review, not a completion percentage or release schedule. Multi-user work is deferred and excluded from the local implementation sequence.
| Area | Current state | Next missing foundation |
|---|---|---|
| Account synchronization | partial | Import pagination and funding support for remaining providers; fiat, staking and subaccounts. |
| Portfolio accounting | partial | Cross-market basis and historical third-currency/transfer-fee valuation. |
| Order management | partial | Broader discovery/recovery adapters, amendment handling and definitive fee/settlement reconciliation. |
| Market-data infrastructure | partial | Broader historical adapters, weighted quota policies, richer operational metrics and WebSockets. |
| Alerts & local operations | partial | Broader risk metrics, silent-worker outage detection, operational diagnostics and external delivery. |
| Automated strategy execution | partial | Broader venue/failure-case acceptance, risk policies, residual exit handling and fee valuation. |
| Strategy research | partial | Additional benchmarks, continuous deployment simulation, nested validation and broader optimization. |
| Sibyl & messaging | partial | Token streaming, evaluated collaboration, resumable/scheduled investigations and messaging clients. |
| ProtoLink workflows | partial | Richer scoped specialist tools, evaluated routing and explicit approval/audit contracts. |
| Forecasting | partial | Rolling evaluation, more model families, automatic recalibration and calibrated drift monitoring. |
| Research & news | partial | Document ingestion, chunking, embeddings, hybrid RAG and cited web search. |
| Explorer, equities & analytics | partial | Historical chain datasets, additional chains, wallet aggregation and disclosure archives. |
| Hosted multi-user support | deferred | User accounts, workspace ownership, permissions and tenant-isolated data/vaults. |
Working foundations
- FastAPI routes, validated inputs, SQLite persistence and an encrypted credential vault.
- Twenty production spot-exchange clients, Binance Testnet, Coinbase Exchange Sandbox, and the persistent offline Demo account.
- Public markets, candles, order-book snapshots, indicators, correlations and account balance reads.
- Portfolio/account views, allocation, observed account-value snapshots and explicit unpriced holdings.
- Manual paper orders, virtual resting limits and guarded manual market/GTC limit orders for Binance, Binance.US, Coinbase, Kraken, OKX, Bybit UTA, KuCoin classic, Bitstamp and the two dedicated test clients; GTC limits and cancellation for Gemini. Settings shows app-level support and gaps per exchange.
- Six rule-based strategies, cost-aware historical backtests, exact-input reruns, chart execution overlays, round-trip accounting, comparable saved runs and explicit Demo research/paper runs.
- Sibyl orchestration and ten ProtoLink module agents with a startup runtime registry, client routing, contextual read-only tools, saved conversations and consultation evidence.
- Live Sibyl consultation progress, request-specific cancellation and retained outcomes/evidence for completed, stopped, failed or interrupted requests. Startup marks unfinished requests interrupted without replaying them.
- Encrypted LLM provider profiles, saved API keys and preset/custom model selection.
- Durable editable source-pinned level/crossing alerts, strategy journal subscriptions and per-operation connection read-failure/recovery watches; a shared persistent unread in-app inbox and bounded sanitized evidence.
- Multi-publisher news ingestion, a persistent archive and filtered text sentiment, fetched crypto profiles/history, a glossary, Bitcoin transaction/address inspection and recent blocks, plus optional equity research.
A paper strategy can consume live prices while still using virtual funds. Its running monitor does not mean it places exchange orders. Initial authenticated balance reads, manual Binance Testnet orders and automated strategy entry/exit/restart with controlled lost-response recovery are recorded in exchange acceptance. Partial-fill/cancellation-race acceptance, broader venues and production execution acceptance remain outstanding.
Core data and trading work
1. Account synchronization — partial
There are on-demand balances, bounded fill/order reads, and a durable normalized fill ledger with background imports for eleven clients/environments, with provider retention shown in the UI. See account synchronization. These are not a complete reconciled account history.
- Persist normalized fills, native order references, fee currencies and rebates in a shared ledger.
- Add explicit observed balance checkpoints and per-asset quantity comparisons across imported markets, native fees/rebates and eligible funding, with gap/timing/scope checks, saved reports, CSV export and read-only Accountant evidence. See balance checks. This does not establish complete account history or settlement.
- Apply documented Kraken ledger fee effects, retain Binance wallet/completion facts, and add source-version-checked local wallet/amount reviews for ambiguous funding; preserve native observations, audit withdrawal and provenance in Balance checks, exports and read-only Accountant evidence. See funding effects.
- Add durable native pagination and incremental imports for Demo, Binance, Binance.US, Binance Testnet, Coinbase Advanced Trade, Coinbase Exchange Sandbox, Kraken, OKX, Bybit UTA, KuCoin classic and Bitget classic.
- Review durable import pagination for the remaining exchange providers.
- Discover open/recent orders for the reviewed execution clients, retain native observations and link known Sibyl intents and shared fills. See order discovery.
- Expand discovery beyond the reviewed providers/ranges and handle complete amendment/lifecycle history.
- Add a normalized funding ledger and read-only range imports for Binance crypto deposits/withdrawals, Kraken deposit/withdrawal/transfer entries and Bybit master-UID on-chain withdrawals, retaining fee/rebate and status coverage. See funding history.
- Add opt-in, durable pending-record updates for Binance crypto deposits/withdrawals and Bybit on-chain withdrawals, independent of manual import ranges; bounded pagination/retries, missing-source review, control/account revision guards and read-only Accountant status. See pending updates.
- Add opt-in rolling recent funding sync for implemented Binance/Kraken/Bybit streams, completed-record corrections, persistent fetched-window gaps, concurrent-source guards, lazy pass reports and append-only observed funding revisions across all imports. See recent sync. A bounded authenticated Binance deposit-history read is recorded; live correction behavior and other venues still need acceptance.
- Extend funding adapters to remaining exchanges, Bybit deposits/off-chain transfers, fiat, earn/staking and subaccounts; discover older arrivals/corrections outside selected lookbacks and implement native removal/bust semantics.
- Add bounded transfer suggestions, manual pairing, audited user confirmation, stale-source review and reversible link removal/restoration. See transfer review.
- Carry original acquisition lots/cost basis for current confirmed equal-amount, explicitly zero-fee transfers in one unambiguous exact market. See carryover scope.
- Add explicit balanced same-asset fee allocations, missing-fee assertions, distinct-charge acknowledgment and separate fee-unit inventory cost. See transfer fees.
- Add historical fee valuation, other fee currencies/rebates, split/merged transfers, cross-market basis and reviewed external-funding classifications.
- Track per-market progress, successful pages/completed passes, failures and bounded retries; support pause/resume and restart recovery.
- Add available-history rescans, revision guards for stale in-flight responses, repeated-page protection, correction counts and retained execution ranges.
- Add complete cash-event coverage, account-wide balance reconciliation, execution bust/removal handling and versioned fill/order correction audits beyond observed funding revisions.
- Match discovered native IDs to local submissions and resume saved discovery pages after restart; uncertain client IDs stay on the reviewed recovery path.
- Exercise authenticated reads and sandbox order lifecycles against actual test accounts.
2. Portfolio valuation and performance — partial
Valuation now discovers direct, inverse and short bridge routes for held assets from the selected feed, independently of watchlists. It uses observed stablecoin/fiat conversions, exposes paths and quote quality, and reports unavailable coverage explicitly. Historical observations require complete reads. See portfolio valuation.
- Price arbitrary held assets through explicit quote/conversion routes, within the selected feed catalog and bounded read budgets.
- Use observed native stablecoin and fiat conversion rates instead of assuming parity; unavailable routes remain unpriced.
- Compute retained exact-market FIFO with base/quote fees and explicit gaps; offer balance-checked current reference marks. See account performance.
- Reconstruct cross-market cost basis and historical third-currency fee conversions for complete exchange P&L.
- Persist audited opening quantity/cost assertions and disclose unknown basis, ambiguous ordering and incomplete coverage.
- Separate deposits/withdrawals from investment performance; implement cash-flow-aware returns.
- Extend wallet coverage to funding, earn/staking, subaccounts and supported unified-account variants.
3. Order management — partial
Reviewed market/GTC limit orders, exact precision checks, durable cancellation reviews, cumulative fill observations and restart-safe REST tracking are implemented for the existing execution allowlist. Demo reserves funds and matches/cancels virtual resting orders. Ambiguous writes are never automatically retried. See order management.
Known orders now import native execution pages into the shared Account history ledger, compare quantity/cost and expose fee gaps in an in-place dialog. Three bounded terminal passes can pick up delayed evidence. Client-ID recovery now handles uncertain submissions on the reviewed clients with exact-detail checks and restart-safe bounded lookups; missing/ambiguous evidence remains uncertain. Complete settlement, broader provider support and authenticated acceptance remain pending. See order/fill reconciliation.
- Review and enable execution for the remaining production providers; most added adapters are read-only.
- Add GTC limit intents, typed cancellation, cumulative partial-fill monitoring and persisted automatic status checks.
- Add cancel/replace with venue-specific atomicity and reservation semantics.
- Add stop, take-profit, trailing, OCO, additional time-in-force, post-only and iceberg support where the venue permits it.
- Extend exact-order fill imports to OKX, Bybit UTA and KuCoin classic; retain their provider limits and fee gaps. Bitstamp/Gemini detailed imports remain pending.
- Import native individual fills for confirmed local orders, share history-ledger identities, compare cumulative quantities/costs and report fee gaps; retain lifecycle cancellation/rejection observations. See order reconciliation.
- Add combined/per-account open and recent order discovery for Binance/Testnet, Coinbase Advanced/Sandbox, Kraken and Demo, with read-only imported rows, persistent progress, filters and shared fill evidence. See discovery coverage.
- Extend bounded late-fee checks to definitive settlement, execution bust/removal handling and broader discovery/adapters.
- Recover uncertain submissions using reviewed client-ID lookups, immutable-detail validation, bounded persisted pagination and explicit abstention on missing/ambiguous evidence.
- Exercise actual Binance Testnet client-ID recovery after deliberately discarding a successful strategy-entry response, without a second buy.
- Extend authenticated recovery to delayed indexing, external amendments, permissions, retention and additional providers.
4. Automated strategy execution — partial
Exchange strategy execution is implemented for nine provider/environment variants, alongside the isolated paper worker.
- Bind a run to a fixed account, environment, exact pair and allocated budget with expiring typed approval.
- Keep paper as the default; require both execution flags for exchange test/production runs.
- Submit price-capped GTC intents through the shared at-most-once order pipeline and derive ownership from native fills/fees.
- Persist plans/client IDs and pause on restart; reconcile existing orders without replaying old signals or uncertain writes.
- Define Pause/Stop: prevent future submissions and retain holdings; previously approved expiry continues.
- Persist optional order lifetimes at submission; cancel due remainders through the shared pipeline, retain partial fills and recover uncertain attempts by reading only.
- Add separate typed Stop & cancel approval, durable restart recovery, bounded late native status checks, monitor evidence and journal/CSV events.
- Add separately reviewed position exits for stopped/completed runs: current owned-fill evidence, fresh native book/funds/precision checks, guarded at-most-once submission and realized P&L/residual monitoring. Partial fills and native caps never cause automatic replacement sells.
- Enforce per-order notional, allocated-loss latch, spread/freshness and rolling order-count limits; retain fee/precision dust explicitly.
- Exercise the manual Binance Spot Testnet limit/cancel and market buy/sell lifecycle, native fills/fees, client-ID lookup, duplicate confirmations and retained state after restart; verify read access on one Binance and Coinbase production account.
- Exercise real Binance Testnet strategy signals and capped orders, duplicate approvals, owned fill/fee accounting, paused restart, shared capital rejection, global halt and a separately reviewed owned-position exit; recover a deliberately discarded entry response by exact native client ID without resubmission.
- Harden partial-fill/cancellation races with deterministic scenarios: matching follow-up counters, atomic per-order fill revisions, renewed checks after history-import corrections, invalidated exit readiness, final-submission guards, fill-before-cancel-rejection recovery, restart with writes off and idempotent reimports. Running strategies wait for cancellation evidence before another signal order. These are controlled application tests, not authenticated venue race acceptance.
- Extend authenticated acceptance to observed partial fills, cancellation/fill races, late corrections, real connectivity failures and other venues before deliberate production enablement. Controlled response loss does not establish general outage behavior.
- Add a durable global strategy submission halt: atomically pause exchange runs, invalidate old reviews/unsent signal plans, enforce the shared write boundary, retain audit history and require revision-bound typed release. Release never resumes runs; approved cancellations/manual orders/separately reviewed exits remain available.
- Add opt-in shared allocated-capital, gross-exposure and observed UTC daily-loss policies across accounts, separately for production/test funds; fresh actual conversion paths, retained/pending exposure, native-write transaction admission, persisted daily baselines/locks, deliberate later-day release, bounded audit and read-only Broker evidence. See portfolio limits.
- Add broader policy options (asset/account concentration, configurable loss windows and risk sizing), archived-run accounting beyond current bounds, and distributed coordination. Any future automatic multi-order liquidation needs its own explicit policy; current residual exits need fresh reviews.
- Add reviewed residual/dust ownership handoff to new runs; value third-currency fees; reconcile/adopt manual position changes; handle definitive settlement/busts and more providers.
5. Market-data infrastructure — partial
The backend uses cached REST snapshots, source-specific catalogs/capabilities and provider-sized candle pages. Settings selects public exchanges, saved account environments, or CryptoCompare/CoinGecko APIs with encrypted keys. Markets supports temporary overrides, exact pair analysis, timestamp/stale labels and quotes-only feeds. Descriptive statistics, a versioned explainable technical-bias score and lightweight drag-region chart zoom are implemented; see statistics scope and formulas. See Market data.
- Persist exact-source OHLCV, reviewed resumable backfills, coverage gaps, manual retention and immutable research snapshots. See candle history.
- Review historical pagination for the remaining providers and exercise authenticated provider acceptance.
- Add WebSocket market and account/order subscriptions with reconnection handling.
- Add shared local read admission, bounded HTTP concurrency, conservative cross-client spacing and read-side cooldowns. See provider control.
- Review exact endpoint-weight/IP/account quota policies; coordinate multiple processes if hosted.
- Expose observed connectivity, separate quote freshness, sanitized errors and queue state in Settings and feed status lines. Live session counters reset on API restart.
- Persist bounded hourly HTTP read outcomes, success rate, mean/peak response time and sanitized errors; inspect 24-hour/7-day windows across restart. History never establishes current credential readiness or uptime. See retained read history.
- Add richer latency distributions and application-wide operational metrics. Continuous availability monitoring remains a separate future capability.
- Support held-asset valuation and editable watchlists beyond the six defaults with bounded provider reads; broader aggregator discovery remains future work.
6. Strategy research — partial
Six configurable strategies, fee/slippage-aware backtests, performance metrics and a paper monitor are implemented.
- Add bounded grid search, chronological holdout/expanding walk-forward evaluation, fixed-settings comparisons, recorded training rankings/test executions, exact-study reruns and Broker evidence tools. See validation.
- Add reproducible seeded finite random search with an explicit candidate budget, full-space validation, coverage/fingerprints, saved plans, exact-study replay checks and native Broker evidence.
- Add nested validation, continuous rebalanced deployment simulations and Bayesian/distribution-based optimization; repeated user experiments can still overfit.
- Retain immutable input datasets and enable offline backtest replay with recorded fingerprints and engine version; current-feed runs also retain their inputs.
- Add additional explicit benchmarks and strategy-version comparisons.
- Add multi-asset strategy execution and allocation rules if required.
- Define a plugin contract for custom strategies and optional Freqtrade integration.
- Implement the planned ML/RL strategies with their own training/evaluation pipeline.
AI and research work
7. Sibyl and ProtoLink workflows — partial
Sibyl and the ten specialists start through ProtoLink 0.8.3 AgentGroup with registry discovery. Each has its own file and prompt. ProtoLink automatically supplies discovered agents and handles native delegation for Sibyl; module requests use native group handles for their registered specialist; owned local consultations inherit explicit private context. The agents have page context and read-only domain tools. They are request-driven assistants. Models choose evidence tools on demand. Every specialist has topic-based page help; Broker can also inspect retained order/fill reconciliation evidence. Tool descriptions explain purpose, arguments and returned information. Agents inherit a workspace model default or use an explicit per-agent provider/model override. Each request freezes the whole team’s configuration; profiles retain shared encrypted provider keys/endpoints.
- Adopt ProtoLink 0.8.3 native thread execution, lazy model factories, private request context, group lifecycle, SQLite conversations and correlated report/evidence APIs. Remove application tool wrappers, LLM adapter and telemetry reconstruction; keep UI projections only.
- Split each agent into its own native declaration with prompt and typed tools; remove consult gateways and per-request Agent construction.
- Add on-demand evidence tools, cached Reporter reads and practical Technician setup guides.
- Review model-facing tool docstrings; add typed, focused help/FAQ tools for every specialist, including Trade history, backtesting, chart controls and source setup. Keep instructions separate from observations and update guides with module changes.
- Let Broker inspect bounded, filtered retained order/fill evidence, with native ProtoLink correction for invalid record selections and no provider refresh or mutation.
- Expand investigations with deeper source-specific tools as domain features become available.
- Sibyl uses native ProtoLink discovery/delegation, with delegation disabled for specialists; saved chats include collaboration evidence.
- Show live consultation lifecycle events, allow stopping the exact queued/running request, and retain request outcomes and completed consultation evidence separately from successful chat turns.
- Mark unfinished requests interrupted on startup; block late consultations/answers after Stop and preserve already committed answers.
- Add scheduled investigations and resumable long-running work. Retained request history is an audit, not workflow resumption.
- Add Ollama installation/server/model detection and explicit local server Start/Stop ownership controls, with setup links.
- Add per-agent provider/model overrides with shared encrypted profiles, preset/custom IDs, revision-checked edits, frozen delegated request snapshots and accurate model attribution. See contracts.
- Stream native tool/consultation lifecycle activity into browser chats; retain polling for reopened Sibyl requests and cancel native work on stream disconnect. Show safe on-demand run summaries with tool timing/outcomes and available per-call token/cost metrics, labeled estimated, partial or unavailable.
- Run an initial real Ollama/Gemma acceptance pass for specialist evidence, native coordinator tool/inference consultations, memory/isolation, Stop and persistence; align provider/context-policy windows and preserve setup instructions in native previews. See findings and remaining limits.
- Add user-facing token streaming, broader repeated routing/grounding evaluation and fix the upstream owned-subagent metrics gap. Activity streaming does not expose raw model chunks or internal action JSON.
- Implement direct-HTTP Telegram polling, private-user binding, encrypted credentials, conversation mapping and durable delivery outcomes. Explicit rate limits retry; ambiguous sends never auto-retry.
- Validate Telegram with a real bot; add media and richer delivery review. WhatsApp, Discord, Slack, Signal and Email remain unavailable previews.
- Design approval/action contracts and audit records before introducing mutation tools.
8. Forecasting / Chronos — partial
v2.1.0 adds direct autoregressive ridge models over immutable daily input, alongside the separate untrained baseline. See the user workflow and engine contracts.
- Implement reproducible daily dataset preparation and real numerical model fitting, with explicit synthetic Demo input through the same engine.
- Add JSON coefficients/scalers, source and engine fingerprints, numbered model versions, exact-input reruns, current/saved-input inference and retained forecast observations.
- Add chronological 60/20/20 training/calibration/test partitions, purged boundary-crossing targets, naive/drift comparisons and independent test metrics/empirical band coverage.
- Connect the UI to trained results, explicit baseline/model tabs, test/forecast charts, exports and native read-only Chronos registry/report tools.
- Expose descriptive lag-feature departures from training ranges without claiming a calibrated drift detector.
- Add rolling/repeated forward evaluation, calibrated drift monitoring, automated recalibration and managed archive/import beyond the initial bounded registry.
- Add neural/ARIMA and other model families, richer features and larger evaluation studies. Historical held-out results do not establish future profitability.
9. Research knowledge and news — partial
Crypto profiles/history are fetched by the backend. Knowledge search is a curated glossary. News has five publisher RSS/Atom adapters, retained excerpts, deduplication, source health, backend filters, extractive summaries, VADER distributions/trends and contextual Reporter tools. See the news backend.
- Add document ingestion, chunking, embeddings and retrieval storage.
- Implement hybrid RAG with attributable citations and optional web-search tools.
- Add multiple news providers, deduplication, retained history, source health and filtered sentiment evidence.
- Add licensed full-article processing and evaluated finance-trained/entity-targeted sentiment models.
- Add configurable external research API credentials/endpoints where integrations need them.
10. Explorer, equities and advanced analytics — partial
Recent Bitcoin blocks, mainnet transaction/address inspection, optional equity quotes/history/statistics and Pearson/Spearman correlation exist.
- Add read-only mainnet transaction/address inspection with checksum validation, exact fee/confirmation evidence, paged inputs/outputs and confirmed activity, separate pending observations, source-labeled UI/exports, independent Demo and native Explorer tools. See Bitcoin explorer contracts.
- Add historical chain datasets, additional chains, wallet aggregation, output-spend tracking and local chain verification.
- Implement the planned public-official portfolio/disclosure archive.
- Add distance correlation, MIC, Granger causality, regression and SHAP only with clear statistical assumptions.
Cross-cutting and optional expansion
Distribution — Python package ready; publication pending
The release path is a complete Python wheel and source archive published from the existing repository. Source checkout and local Docker Compose remain available. No Homebrew repository or registry is required; see distribution.
- Bundle React in wheels/source archives, resolve installed assets, add a per-user installed workspace and launcher, and preserve source workspaces.
- Synchronize release metadata and lock entries in one version command; derive current docs/install examples from metadata while preserving history.
- Exercise a clean installed wheel and add Linux, Windows and both Mac architectures to package CI.
- Add a Python-package release workflow with dry-run artifacts, checksums and GitHub Release uploads.
- Complete remote CI acceptance and publish the wheel/source assets.
- Add Homebrew optionally after the package release; a recipe can live in this repository or a separate tap.
- Validate local Docker runtime and optionally publish prebuilt amd64/arm64 registry images in a separate workflow.
- Decide whether to claim a PyPI distribution name and configure trusted publishing.
- Consider signed desktop installers and optional background-service setup.
11. Alerts and operational jobs — partial
Basic price, strategy and connection alerts, account-risk watches and the local inbox are implemented. Their worker is a bounded domain loop, not a general-purpose job scheduler.
- Persist exact-pair, source-pinned one-shot price rules, Demo attribution, pause/rearm/archive/restore and a 50-rule non-archived budget.
- Evaluate fresh timestamped quotes with persisted scheduling/backoff; atomically save a trigger and its notification, including across restart and lifecycle races.
- Add retained in-app notification history, pagination, unread/read state and a top-bar inbox alongside existing run activity.
- Add revision-checked edits that save paused, inclusive level/observed crossing conditions, reset-based recurrence and persisted cooldowns; preserve original notification evidence.
- Add per-run journal subscriptions for pauses/errors/recovery and lifecycle changes, atomic cursors/notifications, restart catch-up, revision-checked controls and monitor links. See strategy alerts.
- Add connection read-failure/recovery rules per operation, bounded credential-free evidence, atomic cursor/inbox commits, restart catch-up and revision-checked controls. No polling/uptime claim; see connection alerts.
- Add fixed-account-set value floors, asset concentration and observed-value drawdown watches with complete fresh valuation, recovery/cooldown, atomic evidence/inbox commits and read-only Accountant tools. See account risk.
- Add broader risk metrics, silent-worker outage detection and optional fill/order notification policies. Shared strategy exposure/daily-loss enforcement is implemented separately; alerts do not change execution policy.
- Add optional external delivery integrations with explicit recipients, delivery attempts and retries.
- Add a general job queue/scheduler for tasks beyond the current domain workers.
- Add structured application/provider metrics, tracing and operational diagnostics beyond basic health/log output.
- Add managed backup/restore, notification retention/export and credential/key lifecycle tools.
12. Hosted/multi-user product — deferred
Deferred by user request. These items remain visible in Coming soon and are excluded from current implementation priorities. The present deployment is explicitly a single-user workspace with a bearer token and one API worker. These additions are necessary for a hosted multi-user product, not prerequisites for local use.
- User authentication, workspace ownership and per-account authorization.
- Role/permission enforcement and tenant-isolated data/vault handling.
- Database-backed job claims or a separate worker system before running multiple API workers.
- Deployment-specific rate limiting, retention, auditing and recovery operations.
Other optional expansions include derivatives, on-chain wallets/DeFi, exchange transfers, additional regional endpoints and managed local-model downloads/quantization. They should be scoped explicitly rather than counted as broken existing features.
Local implementation sequence
- Market-data foundations: durable candle archive, bounded backfills, gap/health diagnostics and shared read budgets.
- Expand synchronized accounting: remaining provider imports/funding, fee conversion and complete reconciliation; validate authenticated test-account behavior.
- Close the order loop: authenticated acceptance of status/fill/recovery/discovery, broader adapters and final-fee reconciliation, then provider-specific conditional and cancel/replace orders.
- Build on local alerts: broader risk metrics, strategy-event policies and operational diagnostics, then explicit external delivery and backup/restore.
- Sibyl and richer ProtoLink tools: evaluated collaboration, scoped specialist tools and resumable investigations; per-agent models are implemented. Messaging clients remain a separately scoped future integration.
- Test-environment strategies: reconcile actual orders/fills and implement risk controls before any production automation.
- Trained forecasts and advanced research: benchmarked models, RAG, explorer/equities expansions and custom strategies.
Multi-user hosting is deferred. No user accounts, permissions, tenant isolation or distributed-worker work is included in this sequence.
The feature coverage table tracks current user-facing scope. The exchange reference contains the provider matrix and exchange-specific backlog.