audit/tables.md

# Table Provenance

For each paper table: data source plus the audit pass that re-derived
each cell from the underlying run metrics. Audited 2026-04-28 against
commit `a4bf9e6`.

## tab:final_personality_performance — Cross-market 6M summary (Appendix~\ref{appendix:additional_results})

- **Source:** Experiment 1 (US 6M and CN 6M) via comparison bundles
- **Cells verified:** return, MDD, baseline gap (= return − own-market EqW return), rank
- **Status:** ✓ all 10 rows verified

| Market | Mandate | Run ID | return (%) | MDD (%) | Baseline gap | Rank |
|---|---|---|---|---|---|---|
| China | Fundamental Value | `mp_fundamental_value_20260413_163905_878961` | -0.00 | 4.06 | -6.50 | 4 |
| China | Macro Tactical | `mp_macro_tactical_20260413_163905_878979` |  4.02 | 4.55 | -2.48 | 2 |
| China | Behavioral Momentum | `mp_behavioral_momentum_20260413_163905_878974` | -2.52 | 5.28 | -9.02 | 5 |
| China | Low-Volatility | `mp_smart_beta_passive_20260413_163905_878984` |  0.30 | 2.62 | -6.20 | 3 |
| China | EqW | `mp_equal_weight_index_20260413_163905_878987` |  6.50 | 6.25 |  0.00 | 1 |
| U.S. | Fundamental Value | `mp_fundamental_value_20260409_165131_876468` | -4.73 | 6.14 | -14.15 | 5 |
| U.S. | Macro Tactical | `mp_macro_tactical_20260409_165131_876505` |  3.42 | 4.40 | -6.00 | 2 |
| U.S. | Behavioral Momentum | `mp_behavioral_momentum_20260409_165131_876499` |  1.54 | 4.40 | -7.88 | 3 |
| U.S. | Low-Volatility | `mp_smart_beta_passive_20260409_165131_876510` |  0.70 | 2.03 | -8.72 | 4 |
| U.S. | EqW | `mp_equal_weight_index_20260409_165131_876514` |  9.42 | 4.83 |  0.00 | 1 |

## tab:case_study_footprint — Behavior footprint (Appendix~\ref{appendix:additional_results})

- **Source:** Experiment 1 via comparison bundles
- **Cells verified:** trades, avg turnover (%), avg cash (%), avg exposure (%)
- **Status:** ✓ all 10 rows verified (turnover and cash/exposure stored as fractions, displayed as percent)

| Market | Mandate | Trades | Turnover | Cash | Exposure |
|---|---|---|---|---|---|
| China | Fundamental Value | 90 | 1.01% | 38.92% | 61.08% |
| China | Macro Tactical | 1777 | 5.60% | 11.39% | 88.61% |
| China | Behavioral Momentum | 666 | 12.46% | 52.12% | 47.88% |
| China | Low-Volatility | 85 | 0.44% | 69.55% | 30.45% |
| China | EqW | 84 | 0.56% | 1.38% | 98.62% |
| U.S. | Fundamental Value | 123 | 1.72% | 70.56% | 29.44% |
| U.S. | Macro Tactical | 1763 | 3.48% | 36.24% | 63.76% |
| U.S. | Behavioral Momentum | 513 | 10.92% | 69.38% | 30.62% |
| U.S. | Low-Volatility | 27 | 0.28% | 83.76% | 16.24% |
| U.S. | EqW | 63 | 0.45% | 2.40% | 97.60% |

## tab:behavior_validity — Sharpe + holding days (Appendix~\ref{appendix:additional_results})

- **Source:** Experiment 1
- **Cells verified:** trades (CN/US), avg holding days (CN/US), MDD (CN/US), Sharpe (CN/US)
- **Status:** ✓ all 5 rows verified (paper rounds holding days to 1 decimal: 11.82→11.8, 96.79→96.8, etc.)

## tab:reproducibility — Full-run reproducibility (Section~\ref{sec:results}, Q4)

- **Source:** Experiments 1 and 2 (US 6M, R1 vs R2)
- **Cells verified:** R1 return, R2 return, ΔReturn, R1 trades, R2 trades, ΔTrades
- **Status:** ✓ all 5 rows verified

| Mandate | R1 ret | R2 ret | ΔRet | R1 trades | R2 trades | ΔTrades |
|---|---|---|---|---|---|---|
| Equal-Weight | 9.42 | 9.42 | 0.00 | 63 | 63 | 0 |
| Macro Tactical | 3.42 | 2.78 | -0.64 | 1763 | 1817 | +54 |
| Behavioral Momentum | 1.54 | 0.50 | -1.04 | 513 | 512 | -1 |
| Low-Volatility | 0.70 | 0.70 | 0.00 | 27 | 27 | 0 |
| Fundamental Value | -4.73 | -10.29 | -5.56 | 123 | 126 | +3 |

> The 2pp doctrine-effect threshold quoted in the prose is derived from
> the ΔRet column above, not from any external benchmark.

## tab:us_3m_backend_robustness — Default vs GPT-5.4 (Section~\ref{sec:results}, Q4)

- **Source:** Experiments 3 (default backend) and 4 (GPT-5.4) on US 3M
- **Aggregator:** `derived/gpt54_robustness/backend_comparison.csv` (also feeds `fig:backend_robustness`)
- **Cells verified:** return, MDD, trades, avg cash for each of 5 mandates × 2 backends
- **Status:** ✓ all 5 rows × 8 columns verified

| Mandate | Default ret | GPT-5.4 ret | Default MDD | GPT-5.4 MDD | Default trades | GPT-5.4 trades | Default cash | GPT-5.4 cash |
|---|---|---|---|---|---|---|---|---|
| Fundamental Value | 0.00 | 1.23 | 0.00 | 1.10 | 0 | 31 | 100.00% | 95.21% |
| Behavioral Momentum | -0.90 | 4.77 | 1.08 | 1.40 | 513 | 621 | 91.06% | 46.33% |
| Macro Tactical | -0.45 | 4.29 | 4.77 | 3.13 | 893 | 901 | 39.71% | 17.31% |
| Low-Volatility | -2.44 | -2.44 | 3.33 | 3.33 | 15 | 15 | 83.38% | 83.38% |
| Equal-Weight | 5.96 | 5.96 | 3.52 | 3.52 | 52 | 52 | 2.14% | 2.14% |

## tab:us_3m_ablation_main / tab:us_3m_ablation — Mechanism ablation (Section~\ref{sec:results} + Appendix)

- **Source:** Experiment 3 (US 3M default backend; Full and Ablated variants per mandate)
- **Cells verified:** return, MDD, trades, avg cash for 8 variants
- **Baseline gap:** computed as return − 5.96 (EqW return on this window)
- **Status:** ✓ all 8 rows verified

| Mandate / Variant | Run ID | Return | Baseline gap | MDD | Trades | Cash |
|---|---|---|---|---|---|---|
| Fundamental Value / Full | `20260417_160942` | 0.00 | -5.96 | 0.00 | 0 | 100.00% |
| Fundamental Value / No filter | `20260417_191443` | 5.80 | -0.16 | 2.10 | 611 | 20.72% |
| Behavioral Momentum / Full | `20260417_220721` | -0.90 | -6.86 | 1.08 | 513 | 91.06% |
| Behavioral Momentum / No guardrails | `20260418_004927` | -0.75 | -6.71 | 1.60 | 392 | 89.05% |
| Macro Tactical / Full | `20260418_033940` | -0.45 | -6.41 | 4.77 | 893 | 39.71% |
| Macro Tactical / No tilt | `20260418_072633` | -0.33 | -6.29 | 5.35 | 918 | 35.76% |
| Low-Volatility / Reference | `20260418_110947` | -2.44 | -8.40 | 3.33 | 15 | 83.38% |
| Equal-Weight / Reference | `20260418_111210` | 5.96 |  0.00 | 3.52 | 52 | 2.14% |

## tab:efficiency_ablation — Execution paths (Appendix)

- **Source:** Experiment 5 source doc `docs/2026-03-23-cn-10tickers-6m-efficiency-ablation.md`
- **Cells verified:** total runtime, speedup, LLM call count, total tokens, estimated cost (CNY), cache bundle counts, shared phase-one timing, phase-two mode
- **Status:** ✓ verified against source doc

## tab:released_work_comparison — Related-work comparison (Section~\ref{sec:related})

- **Source:** Hand-coded from cited papers' reported scope (DeepFund, AMA, STOCKBENCH, InvestorBench, TradingAgents, FinMem, HedgeAgents)
- **No backtest runs to verify against;** the row for QuantArena (#assets=40, #days=226) is the sum of the two case-study universes (20 US + 20 CN tickers, 124 + 102 trading days).

## tab:sector_style_universe — Universe matrix (Appendix)

- **Source:** `universe/sector_style_universe.csv` (40 rows: 20 US + 20 CN tickers, each tagged with sector and size-style cell)
- **Status:** the figure-generation scripts and the paper text both consume this same CSV, so the 5×4 grid in the table is by construction what the case studies trade.

## tab:implementation_matrix — Implementation matrix (Appendix)

- **Source:** Hand-coded mapping of mandate → engine class / LLM role / signal sources / guardrails / diagnostics (no per-run artifact to verify).