Capstone Research Report Template

Status: The bounded research design and technical protocol have been accepted; target-market execution is blocked. See the current Capstone README. The engineering appendix below retains its historical identity and is not a French-market result. The original generic prompts are provided for comparison. The selected French protocol does not include model training, neutralization, a Cartesian grid, or mandatory DSR/PBO calculations; use the current design and report guide for each section.

1. Research Hypothesis

Fixed now: S01/S02: fixed French ten industries, L11/skip1/top3, EW10; $H_0:\mu_d\le0$ / $H_1:\mu_d>0$ ; one month is the experimental unit; slow reaction is not an identified cause. See Guide item 1. Target-market output: not run.

Prior economic mechanism, target assets/frequency, falsifiable expectation, baseline, and registration date.

2. Data Sources

Fixed now: S08/S09, R/P: 2026-10-05 revised snapshot, industry/FF3 identity, and download UTC; G01 subject/use is unknown; raw redistribution is not_requested. See Guide item 2. Target-market output: not run.

Source text and URL, license, download UTC, coverage, hashes, and supplier revisions; do not invent missing details.

3. Data Dictionary

Fixed now: S04/S03: long table (snapshot_id, month, industry_id), decimal total returns for ten industries, FF3/RF, score/weight/amount/net-return/label-maturity fields. See Guide item 3. Target-market output: not run.

Fields, keys, types, units, frequency, time fields, missingness, and revision rules.

4. Point-in-Time Logic

Fixed now: S03: half-open $[B_t,B_{t+1})$ ; $S=B_t-1\text{ second}$ and $\mathrm{allocation}=B_t$ are hypothetical ordering markers only; available / ingested / revision remain unknown; the normal/future rows are CONSTRUCTED. See Guide item 4. Target-market output: not run.

Announcement/availability/ingestion/signal/trade/return/label timeline; as-of join and counterexample tests.

5. Data Cleaning

Fixed now: S02/S04/S06: reject duplicate month/industry keys, missing months, non-finite values, and missing values; divide source percentages by 100 once only; do not add corporate-action cash already reflected in total return. See Guide item 5. Target-market output: not run.

Deduplication, missing values, bad prices, splits/dividends/delisting; before/after counts and traceable logs.

6. Feature Engineering

Fixed now: S02/S03/S05: $\operatorname{score}_{i,t}=\prod_{m=t-12}^{t-2}(1+R_{i,m})-1$ , 11 terms; general window $[t-\mathrm{skip}-L,\ t-\mathrm{skip}-1]$ ; first-evaluation-month endpoints are listed. See Guide item 6. Target-market output: not run.

Mathematical definition, units, window endpoints, train-only fitting, economic meaning, and ablation.

7. Signal Construction

Fixed now: S02/S05: fixed order of ten industries; descending score / ascending manifest index for ties; primary top3 at $1/3$ each, sensitivities at $1/k$ ; no fitting. See Guide item 7. Target-market output: not run.

Ranking, winsorization, neutralization, missing-value treatment, and combined weights; verify direction and time.

8. Portfolio Construction

Fixed now: S02: $V_{post}=V_{pre}-c\sum_i|w_iV_{post}-x_i|$ ; $x_i$ is the drifted holding amount; fully invested long-only, gross=net=1; first month $1/(1+c)$ ; no segment reset. See Guide item 8. Target-market output: not run.

Budget, gross/net, sector/beta, leverage, cash, drift, feasibility, and rebalancing.

9. Transaction Cost Model

Fixed now: S02/S05/S06: all-in cost is $c\times|\text{traded notional}|$ , $c=0.001$ ; do not divide by 2 or charge spread again. $200.10\times0.001=0.20010$ ; $(1-0.001)(1+0.02)-1=0.01898$ is a separate constructed amount example. See Guide item 9. Target-market output: not run.

Commissions, spread, impact, financing, borrow, capacity, calibration limits, and gross-to-net reconciliation.

10. Walk-Forward Validation

Fixed now: S02/S03/S05/S07: 18 warm-up months; $440=176+132+132$ ; sole primary period 2015-09..2026-08; 13 paths continuous and carried across phase boundaries; do not invent training; history is exposed and is not a blind holdout. See Guide item 10. Target-market output: not run.

Training/selection/final test, rolling calendar, overlap/purge/gap, and fit/prediction availability times.

11. Performance Analysis

Fixed now: S02/S05: $d_t=r_{M0,net,t}-r_{EW10,net,t}$ ; primary $n=132$ , $k=1$ , common- $n$ gamma denominator, Bartlett lag 6, $q=132/131$ applied to covariance, $\mathrm{SE}=\sqrt{\widehat\Omega/131}$ , approximate $t(131)$ ; bootstrap block 12 / 5,000 / fixed seed 20261012. See Guide item 11. Target-market output: not run.

CAGR/Sharpe/drawdown, intervals under dependence, simple benchmarks, and phase-specific descriptions; do not report only the best period.

12. Risk Analysis

Fixed now: S01/S02/S06: top3 concentration, industry weights, maximum drawdown, historical tails; a uniform constructed $-10\%$ shock checks only the risk formula; without volume/order-book data, do not estimate ADV capacity. See Guide item 12. Target-market output: not run.

Factors, industries, concentration, tails, liquidity, leverage, margin, and stress scenarios.

13. Factor Attribution

Fixed now: S02/S05: subtract same-month FF3 RF once from net total return; do not subtract RF again from published Mkt-RF / SMB / HML; use an intercept and $k=4$ correction $n/(n-4)$ . The primary paired $d_t$ does not subtract RF and uses $k=1$ . See Guide item 13. Target-market output: not run.

Same-frequency, same-currency RF; CAPM/FF; HAC choice; residuals and scope of alpha interpretation.

14. Robustness Tests

Fixed now: S02/S05/S06: report all cost/skip sensitivities, the prespecified phases, and statistical block-length 6/24 sensitivities; retain negative/zero results and do not choose conventions based on intervals. See Guide item 14. Target-market output: not run.

Prespecified subsamples, alternate conventions, costs, delays, extreme periods, ablations, and alternate baselines.

15. Parameter Sensitivity

Fixed now: S02/S05: M0/C0/C25/C50/L5/L17/K2/K4/S2 (nine strategies) plus four EW cost paths; zero optimization; one factor at a time, no Cartesian product. See Guide item 15. Target-market output: not run.

Parameter grid, all trials, stable regions, and figures; do not re-register based on the best test result.

16. Overfitting Discussion

Fixed now: S07/S02: P3 saw all of 1990–2026; P1/P2 saw portions; other access is unknown; not blind and full_preregistration=false; PSR/DSR/PBO may be N/A if applicability is insufficient. See Guide item 16. Target-market output: not run.

Research degrees of freedom, multiple testing, PSR/DSR/PBO applicability, omitted searches, and limits on the final holdout.

17. Limitations

Fixed now: R/P, S01/S02/S06: rights are unresolved; historical PIT is unknown; hypothetical allocations are not actual fills or ETF replication; costs are not calibrated to capacity; retain old P6/39 gaps; the monthly-return adapter is not implemented. See Guide item 17. Target-market output: not run.

Data, trading, statistical, model, market-regime, capacity, and deployment gaps; state unsupported inferences explicitly.

18. Final Research Conclusion

Fixed now: R/P, S01/S02/S06: current reportable status is design/protocol technically accepted and execution blocked; future positive, zero, negative, or interval-crossing-zero outcomes each have their own interpretation. Engineering PASS does not answer $\mu_d>0$ . See Guide item 18. Target-market output: not run.

Answer the initial hypothesis; distinguish evidence, conjecture, and unresolved questions. Report honestly if the net result is not economically meaningful.

Engineering Foundation Evidence Appendix (A9-B Engineering Base Interface)

The following preserves the historical status and figures from the original A9-B engineering appendix. These historical pending / candidate labels do not override the current A10-A / A10-A.1 acceptance status and do not mean that market research has been run. See the current research entry and the 18 guide items above for current conventions.

Historical status: The A9-B engineering-foundation phase was completed, providing an offline production pipeline, quality gates, persistent ledger, and reconciliation loop. All real research and strategy-development work was explicitly marked pending_A10 at that time.

Engineering evidence and research status by report section

Report section	Historical status	Engineering evidence / artifact index	Note
1. Research Hypothesis	`pending_A10`	No strategy hypothesis; framework interface only	Formal research hypothesis awaited A10
2. Data Sources	`pending_A10`	`CONSTRUCTED_TEACHING_DATA` v1.0.0	No external real-data download
3. Data Dictionary	`pending_A10`	`capstone/data_dictionary.md` appendix	Constructed-field engineering dictionary only
4. Point-in-Time Logic	`author_candidate_pending_independent_acceptance`	$\max(\texttt{avail},\texttt{ingest},\texttt{as\_of})\le\texttt{signal}<\texttt{trade}$	Constructed time contract and rejection tests; pending independent acceptance at the time
5. Data Cleaning	`author_candidate_pending_independent_acceptance`	Staleness threshold ( $\le 3600\text{ s}$ ), strictly positive prices, non-null/non-duplicate checks	Rejects the anomalies listed in the pre-fill contract
6. Feature Engineering	`pending_A10`	One-period teaching momentum $(P_t-P_{t-1})/P_{t-1}>0$	Formal A10 feature engineering and ablation were pending
7. Signal Construction	`pending_A10`	Fixed teaching allocation quantities (A: 10, B: 20)	Cross-sectional ranking and neutralization awaited A10
8. Portfolio Construction	`pending_A10`	Fixed teaching positions; ex-ante leverage check $0.20\le0.50$	Formal portfolio method awaited A10 preregistration
9. Transaction Cost Model	`author_candidate_pending_independent_acceptance`	10 bps spread embedded in fill price; 10 bps commission charged explicitly	No duplicate charge; total cost USD 4.002
10. Walk-Forward Validation	`pending_A10`	No rolling-window split run	Real rolling out-of-sample backtest awaited A10
11. Performance Analysis	`pending_A10`	Baseline ending NAV $9{,}995.998\,\mathrm{USD}$ ; net loss $-4.002\,\mathrm{USD}$	Teaching example, not real strategy alpha
12. Risk Analysis	`author_candidate_pending_independent_acceptance`	$-10\%$ market-shock scenario loss USD 200.0 $\le$ USD 1000.0 limit	Stress test and ex-ante leverage gate passed
13. Factor Attribution	`pending_A10`	No multi-factor decomposition	Attribution awaited A10 under Project 2 units/RF/HAC rules
14. Robustness Tests	`pending_A10`	No parameter-robustness perturbations	Subsample and sensitivity analyses awaited A10
15. Parameter Sensitivity	`pending_A10`	No parameter-grid scan	Parameter response surface awaited A10
16. Overfitting Discussion	`pending_A10`	No multiple-testing check	PSR/DSR/PBO calculations awaited A10
17. Limitations	`author_candidate_pending_independent_acceptance`	Offline, single-process environment; constructed teaching data; no live-market impact	Engineering boundaries stated; no extrapolation
18. Final Research Conclusion	`pending_A10`	Constructed engineering-interface candidate; strategy-validity test awaited A10	No unsupported empirical investment conclusion

R2 engineering repair

R01–R04 counterexamples / STOPPED / overwrite and actual-order evidence are in the historical source qa/a9/ch45_batch/revisions/r2_20261012. The amount tolerance is $10^{-6}\,\mathrm{USD}$ ; real-valued share quantities and three recovery branches are described in ENGINEERING_BASE.md. All real-research sections were still pending_A10 in this historical appendix. It does not fill real results or register acceptance.