Appendices

Back matter · Companion to the master curriculum · Canonical edition v4

These appendices are a curated desk for the commodity-ETP and exchange-native BTC/ETH intraday curriculum. They are not an exhaustive copy of every citation in legacy module bodies. Until those bodies complete their scope-migration pass, excluded-market material must be labeled wrapper context, legacy/migration context, or transfer evidence as applicable.

Use this order when facts conflict:

1. current product prospectus or supplement;
2. current regulator, listing exchange, broker, or crypto-venue rule/document;
3. timestamped quote, order, fill, statement, or other direct observation;
4. peer-reviewed/reproducible research;
5. practitioner education.

A source about market mechanics does not establish a profitable edge. Product and venue rules must be retrieved, date-stamped, and copied into the project instrument master before each implementation or release.

Appendix A — Scoped Reading Ladder

The core ladder favors market structure, execution, honest research, and model validation. Books using equities, FX, or direct futures provide methods only; their strategies do not enter the canonical universe without new in-scope evidence.

Short reading routes

• Before the first strategy: Harris → Johnson → the product prospectus and venue rules for the chosen instrument.
• Before the first ML experiment: Jansen → M18/M19 sections of López de Prado → primary library documentation.
• Before changing sizing: Carver’s framework sections → Module 23’s intraday risk constitution. Do not import a longer-horizon holding assumption.
• Before RL: M5 and M22 acceptance evidence → Sutton and Barto → Module 21. RL is optional, not a milestone required for deployment.

Appendix B — Product, Venue, Research and Engineering Sources

B1. Primary U.S. commodity-ETP sources

Use the exact product’s documents, not a generic ETF description.

• SEC EDGAR company-filings search — retrieve dated prospectuses, registration statements, product supplements, amendments, and material event filings.
• SEC — Updated Investor Bulletin: Exchange-Traded Funds — explains intraday share trading, NAV, premiums/discounts, and creation/redemption. Coverage caveat: its 1940 Act ETF discussion does not cover every exchange-traded commodity fund or ETN.
• NYSE market hours and calendars — primary calendar input when the selected product lists or trades under applicable NYSE rules.
• Nasdaq Trader calendar — primary holiday and early-close reference for applicable Nasdaq trading.
• The selected listing exchange’s current rulebook and trading notices — verify regular hours, auctions, order eligibility, halts, clearly erroneous trades, tick increments, and effective dates.
• The selected broker’s current account and order documentation — verify routing, fees, short/borrow availability, margin, fractional-share support, auction access, API behavior, and forced-liquidation rules.
• The product issuer’s current prospectus, summary or supplement, methodology, holdings, roll schedule where applicable, tax disclosures, and lifecycle notices. Store the exact URLs and document hashes in the project card.

B2. Commodity-wrapper education

These sources help interpret product mechanics; they are not trading-edge evidence.

• Fidelity — Commodity ETFs: contango and backwardation — explains why a futures-based product can diverge from spot through contract rolling.
• SEC investor publications — regulator-level background. Always check whether a publication applies to the selected legal wrapper.
• Product issuer methodology and holdings pages — preferred over generic articles for collateral, contract selection, roll timing, fees, and concentration.

Direct commodity-futures specifications may be consulted to understand a futures-based ETP’s underlying process. They remain wrapper-mechanics context only; the canonical strategy trades the listed ETP share.

B3. Primary native-crypto sources

A venue is approved only after legal/account review and technical due diligence. A listing here is a documentation starting point, not an endorsement or guarantee of access.

• Coinbase International Exchange — Instruments — contract and instrument concepts, including metadata needed by an adapter.
• Coinbase — Funding rate — funding explanation. Verify the current interval, formula, timestamps, and eligible instrument in primary venue metadata.
• Coinbase Developer Platform documentation — API, authentication, products, market data, orders, errors, rate limits, and changelog entry point.
• Kraken API Center — alternative CEX API and market documentation entry point if that venue passes approval.
• The approved venue’s current rulebook, fee schedule, status page, maintenance notices, contract specification, mark/index composition, liquidation rules, and API changelog. Archive effective versions with the instrument master.

Never teach that perpetual funding is “always every eight hours.” Interval, formula, cap/floor, settlement currency, and exceptional handling are venue- and instrument-specific.

B4. Data, sessions and direct observations

• Databento documentation — normalized exchange data, schemas, symbology, and timestamps for eligible U.S.-listed products.
• Polygon.io documentation — U.S. market data API option; confirm plan entitlements, corrections, conditions, and historical coverage.
• Official broker/CEX REST, WebSocket, FIX, drop-copy, and statement documentation — the canonical serving contracts for the chosen venue.
• Apache Parquet and DuckDB documentation — durable columnar storage and reproducible local analytics.
• Polars documentation and Apache Arrow — typed, columnar transformation contracts used by the reference pipeline.

Vendor bars are not sufficient evidence for short-horizon fills. Preserve native trades/quotes, condition codes, sequence or update IDs, receive timestamps, order intents, acknowledgements, fills, cancels, rejects, balances, and statements where available.

B5. Execution and transaction-cost analysis

• Harris — Trading and Exchanges — who supplies/takes liquidity and why costs exist.
• Johnson — Algorithmic Trading & DMA — arrival-price accounting, order tactics, and implementation shortfall.
• Almgren and Chriss — Optimal Execution of Portfolio Transactions — impact-versus-timing-risk framework. Institutional sizing assumptions require adaptation; the accounting concepts remain useful.
• The selected venue and broker’s primary order-type, routing, auction, fee, and execution-quality documentation.
• Your own timestamped decision/order/fill ledger — the deciding evidence for cost distributions, not a generic spread statistic from a blog.

B6. Research and validation

• Machine Learning for Trading repository and ml4trading.io — end-to-end workflow and code references.
• Bailey and López de Prado — The Deflated Sharpe Ratio — adjusts an observed Sharpe for selection and non-normality.
• Bailey et al. — The Probability of Backtest Overfitting — selection-risk framing used with combinatorial validation.
• scikit-learn probability calibration — reliability diagrams and calibrated classifiers.
• scikit-learn permutation importance — model-agnostic feature audit with documented caveats.
• statsmodels time-series API — primary implementation reference for stationarity, serial dependence, and related tests.
• PyTorch documentation — primary M20 implementation reference.
• Gymnasium documentation and Stable-Baselines3 documentation — M21 extension tooling only after simulator acceptance.

Published results on individual equities, equity-index products, FX, options, or other excluded instruments can motivate a preregistered mechanism. Label them transfer evidence only and retest from zero on the approved instrument, venue, clock, costs, and fills.

B7. Reference-system and MT5 migration sources

These sources support migration and comparison with m1-trading-model. They are not the canonical deployment lane.

• MetaTrader5 Python integration — legacy adapter semantics.
• MQL5 ONNX reference — optional model-serving portability for the legacy adapter.
• MQL5 market information constants — contract fields needed to reproduce legacy behavior.
• ONNX Runtime documentation and Netron — venue-neutral model export/runtime inspection.

XAUUSD MT5 remains an OTC-CFD migration benchmark. Do not use its symbol name, spread, lot, session, or swap assumptions for a commodity ETP.

Appendix C — Tooling Landscape

Choose tools after defining contracts. A connector’s symbol list does not approve a venue, and a backtest package does not make its default fill model honest.

Selection rules

• Pin versions and archive API schemas/changelogs used by a release.
• Prefer native venue APIs for behavior that a generic connector cannot represent.
• Test decimal precision, rounding, minimum notional, rate limits, sequence gaps, reconnect, cancel races, partial fills, and idempotency before paper trading.
• Keep research, replay, shadow, and live feature/order contracts versioned together.
• Treat any black-box data correction or synthetic bar as a model input with lineage, not as ground truth.

Appendix D — Glossary

Definitions are scoped to this curriculum. “Current” always means current for the specific product, venue, account, and effective date recorded in the project card.

• Approved venue — a broker or centralized crypto exchange that has passed legal, account, custody/counterparty, product, data, API, operational, and risk review for a named instrument. Library support alone is not approval.
• As-of join — joining a slower or irregular record to an intraday row using the latest value that was actually available at that row’s decision time.
• Authorized participant (AP) — an institution able to create or redeem eligible exchange-traded product shares in large units under the product’s process. AP activity does not guarantee that market price always equals NAV.
• Auction — an exchange mechanism that matches accumulated interest at a defined opening or closing event. Eligibility, imbalance data, cutoffs, and fills are exchange- and order-type-specific.
• Basis — the price difference between related instruments, such as spot and a linear perpetual. It can move, incur different costs on each leg, and fail to converge inside the allowed session.
• Borrow availability — whether an account can short a specific ETP in the requested size at that time and cost. A historical backtest cannot assume it.
• Centralized exchange (CEX) — a venue operated by an intermediary that maintains accounts and matches or intermediates crypto orders under its own rules.
• Commodity ETP — the curriculum’s umbrella for a U.S.-listed exchange-traded commodity exposure. The exact legal wrapper may be a trust, commodity pool/product, note, or another disclosed structure; never infer it from the ticker.
• Commodity pool/product — a structure that can obtain commodity exposure through futures or related instruments. Its prospectus and methodology determine rolls, collateral, fees, and tax treatment.
• Completed bar — a bar whose interval has ended and whose value has passed the declared source/processing release lag. H1/H4/D1 bars are context-only here.
• Contango / backwardation — an upward- or downward-sloping futures curve. For a futures-based commodity ETP, contract selection and rolling can make share returns differ from spot; this is wrapper context, not a direct-futures strategy.
• Creation/redemption — the primary-market process that changes ETP shares outstanding, often in large units through eligible participants. Intraday traders normally transact shares on the secondary market.
• Decision price — the benchmark price observable when the strategy decided to trade. It begins the implementation-shortfall ledger.
• Deflated Sharpe ratio — a Sharpe assessment adjusted for selection across many trials and for non-normal returns. It is one diagnostic, not deployment approval.
• Effective spread — a fill-cost measure relative to a contemporaneous midpoint. It must use synchronized and valid quotes.
• Embargo — a time buffer that excludes observations near a test fold to reduce leakage from overlapping labels or dependence.
• ETN — an unsecured exchange-traded debt obligation whose payoff follows stated terms. In addition to market and tracking risk, the holder has issuer credit risk.
• ETF — an exchange-traded fund. Do not use “ETF” as shorthand for every commodity ETP; legal and regulatory structures differ.
• Executable price — a price at which the strategy could buy or sell the allowed size under the declared order/fill model, not merely a last, mark, index, midpoint, or candle close.
• Forced flatten — an unconditional order/risk process that reduces all strategy exposure to zero by the declared session cutoff, with retries, escalation, and an incident path if the first attempt fails.
• Funding payment — a venue-defined transfer between eligible perpetual longs and shorts. Sign, interval, formula, cap/floor, settlement, and exceptional behavior must come from current instrument documentation.
• Funding window — a declared interval around a funding timestamp in which the strategy either permits and books funding exposure or prohibits it.
• Implementation shortfall — the difference between the outcome at the decision benchmark and the actual executed outcome, including delay, slippage, fees, impact, and unfilled quantity.
• Instrument master — the versioned point-in-time table describing identity, wrapper/contract, venue, clocks, tick/lot/minimums, price fields, fees, funding, borrow, margin, lifecycle, and source evidence.
• Latency budget — the maximum time allowed for each stage from market-data receipt through decision, risk, order submission, acknowledgement, and state update.
• Linear perpetual — a perpetual crypto derivative whose P&L and margin are expressed in a quote currency or stable-value settlement asset according to the venue contract. Only explicitly approved linear contracts are in the initial lane.
• Liquidity impact — adverse price movement associated with executing size. It is conditional on instrument, venue, clock, volatility, order tactic, and available liquidity.
• Mark / index / last price — distinct fields: the index is a venue-defined reference, the mark is a venue-defined risk/valuation price, and last is the most recent trade. None is automatically the strategy’s executable bid or ask.
• Meta-labeling — a secondary model estimating whether a primary route’s candidate should be accepted or sized. It must be validated route by route.
• NAV, premium and discount — net asset value is a product valuation measure; a premium or discount is the share market price above or below the applicable NAV measure. NAV is not necessarily a real-time executable price.
• OTC CFD — an over-the-counter contract for difference against a broker rather than a U.S.-listed commodity ETP share. XAUUSD MT5 is legacy/migration context.
• Partial fill — execution of only part of an order. The remaining quantity and hedge/risk state require explicit cancel, retry, or flatten behavior.
• Physical commodity trust — an exchange-traded trust intended to hold a physical commodity under its governing documents. Share price can still differ from NAV and incurs spread, fees, and market-structure risk.
• Point-in-time data — data represented as it was knowable at a historical decision time, including effective/revision timestamps and without later metadata silently overwriting history.
• Probability of backtest overfitting (PBO) — an estimate of how often a selected configuration performs well in training combinations but poorly out of sample.
• Purging — removing training observations whose label intervals overlap a test interval.
• Quote-aware replay — a simulation using valid bid/ask or book updates and explicit order rules rather than filling every trade at a candle close.
• Regular trading hours (RTH) — the declared core exchange session applicable to the selected U.S.-listed product. Calendar, early-close, and auction rules still require primary verification.
• Roll return — the component of a futures-based product’s outcome associated with replacing one underlying contract with another under its methodology.
• Route — a deterministic candidate-generating sleeve before a meta-model. In the verified legacy system, enabled routes are XAU R0/R2 and BTC R2.
• Same-session exit — closure before the project’s declared ETP or synthetic crypto session ends. Historical depth does not change this maximum hold.
• Shortfall ledger — decision, order, acknowledgement, cancel/reject, fill, fee, funding/borrow, and benchmark records joined under durable IDs for TCA.
• Synthetic crypto session — a UTC-bounded research and risk day imposed on a continuously trading venue so starts, limits, attribution, and forced flattening are deterministic.
• Transaction-cost analysis (TCA) — measurement of the path from decision to executed outcome, including spread, fees, delay, slippage, impact, missed fills, funding, and borrow where applicable.
• Triple-barrier label — a path-dependent label determined by profit, stop, or time barriers. In this curriculum barriers and exits must use executable-price logic and respect forced flattening.
• Venue fragmentation — differences among venues in liquidity, participants, fees, risk rules, mark/index design, funding, and outages. Same-symbol contracts are not assumed fungible.
• Walk-forward validation — repeatedly train on past data and evaluate the next unseen interval in chronological order, with purging/embargo where labels overlap.

Maintain these appendices with docs/CURRICULUM.md. Add a source only when its scope, authority, effective date, and intended use are clear.