Capstone: From Engineering Closure to Research Design
Package note. Run instructions in the engineering document apply to the final, complete course package. This Markdown-only candidate does not include runnable course entrypoints. Links into outputs/ identify read-only I3 sources; they are not command working directories. Final package navigation will be mapped during integration.
This reader edition was added on 2026-10-14. The research question concerns the paired net monthly return difference between industry momentum strategy M0 and EW10 under the fixed revised French ten-industry history, hypothetical monthly allocations, and an all-in assumed cost. The A10-A design received limited acceptance, and the A10-A.1 v2 execution protocol received technical acceptance; G02 (dates) and G04 (statistical and cost conventions) are closed. G01 (the actual subject, use, and applicable rights evidence) remains unknown; G05 has not authorized A10-B, so execution is blocked. Historical available / ingested / revision times remain unknown. full_preregistration, market_verified, and execution_authorized are all false. Raw redistribution is not_requested, which does not mean granted. The historical data have been exposed and are not a blind holdout.
Reading order and prerequisites
Read Research Question, Sources, and Prior Exposure, then the Monthly Return Data Dictionary and Time and Label Contract, then the Full Mathematical Protocol. Finally, follow the 18-item report guide while completing the report template. The read-only I3 source map records source hashes, versions, and reader destinations; readers do not need production QA to understand this protocol.
Prerequisites by topic: Chapters 3 and 11 cover data sources and compounding; Chapter 15 covers RF / FF3 attribution; Chapters 22–25 cover hypotheses and signals; Chapters 29 and 31–33 cover costs, time, and inference; Chapters 35–36 cover cost units and execution boundaries; Chapter 38 covers label maturity; Chapter 45 covers identity, status, and recovery. Projects P1–P3 show historical statistics and benchmarks, P4 shows constructed costs, and P6 shows prediction, labels, and unresolved real-research boundaries. Direct links to the current English chapter sources: Chapter 3, Chapter 11, Chapter 15, Chapter 22, Chapter 23, Chapter 24, Chapter 25, Chapter 29, Chapter 31, Chapter 32, Chapter 33, Chapter 35, Chapter 36, Chapter 38, and Chapter 45. Return through the read-only body index, Loom index, and P1, P2, P3, P4, and P6.
The two configs and the two dictionaries
| Material | Units and purpose | Use |
|---|---|---|
offline_constructed.json and the engineering dictionary |
AssetA / AssetB; USD/share, shares, second-level constructed timestamps; two fixed trades | Run the constructed example in an isolated environment using the engineering entrypoint; preserve the config bytes |
french_protocol_v2.json and the monthly-return dictionary |
Monthly decimal total returns for ten industries, portfolio weights, and normalized amounts | Read alongside the parameter teaching material and preserve the config bytes; there is no run command or implemented monthly-return adapter |
The French parameter file has SHA-256 d49ac62cff7d6496e478211921d507fe7216c18b0e4f13b57e48fc893eb12505. Its candidate gate fields retain their source identity. The original main conversation closed G02 and G04 on 2026-10-13. Do not treat old pending fields as current open items, and do not use those closures to change execution_authorized to true. Full source hashes are recorded in the read-only source map.
The bridge from engineering to research
The engineering pipeline can carry over run_id, input/config/code identity, quality rejections, idempotency, and recovery semantics. Its two-asset, single-period price/share interface cannot represent multi-period monthly total returns for ten industries. A monthly-return ledger adapter would be needed; it has not been implemented and is not authorized. The following is a data-flow sketch only:
fixed snapshot / units / month key -> monthly quality gate -> cutoff and 11-term compounded score
-> fixed-order top-k and EW targets -> carry forward prior-month drifted amounts
-> solve post-cost NAV and target trades -> monthly total return -> month-end holdings / final liquidation
-> continuous 440-row ledger -> primary 132-row paired difference -> fixed statistics / risk / post-hoc attribution
A return or wealth index cannot stand in for a USD/share price. A 3,600-second staleness gate cannot be applied as a monthly-return cleaning rule. Without minute prices, volume, order-book data, or execution slots, TWAP / VWAP / POV cannot model real execution. The engineering example's USD 9,995.998 NAV and USD 4.002 cost explain only the constructed engineering example; they do not fill French research results.
This teaching package explains how to design and interpret the study. Target-market results do not exist, so it cannot answer the primary hypothesis. It does not register the full Capstone as verified or formally published.