Aggies Prosperity

Use trader.py for the final Prosperity simulator submission. Earlier algorithmic submissions are archived in trader_round1.py, trader_round2.py, and trader_round3.py. The local datamodel.py mirrors the competition interface closely enough for syntax checks and lightweight local tests.

Final Results

Final leaderboard snapshot:

• Overall placement: #193.
• Total PnL: 487,287 XIRECs.
• Algorithmic placement: #56.
• Manual placement: #2838.
• Country placement: #54.

See logs/final_results.md for the competition retrospective.

Research notebooks:

• round5_final_strategy.ipynb: final Round 5 simulator strategy, public repository lessons, product-by-product alpha isolation, ML rejection tests, passive selected makers, one-tick signal takers, jump-reversion takers, and Ignith manual Ashflow Alpha allocation.
• round4_v14_detailed_findings.ipynb: Round 4 algorithmic findings and the filtered approach to encrypted trader identities / mark signals.
• round3_algorithmic_options_strategy.ipynb: Round 3 Solvenar workflow, Hydrogel/Velvetfruit microstructure diagnostics, foundational options-theory citations, Black-Scholes applicability notes, voucher residual / spread analysis, and replay summary.
• round2_algorithmic_trading_strategy.ipynb: round 2 data regime, contingent-claims framing, ridge/KNN signal screens, risk controls, and strategy replay.
• round2_manual_invest_expand_strategy.ipynb: 50,000 XIREC Research/Scale/Speed allocation optimizer, Speed-rank game theory, and submission recommendation.
• round1_algorithmic_model_trading.ipynb: algorithmic model process, findings, backtests, overfit controls, and profitability thesis.
• round1_manual_auction_strategy.ipynb: manual auction mechanism, exact optimum, and auction backtest.

Round 3 Strategy

• Active submission file: trader.py.
• Trades HYDROGEL_PACK, VELVETFRUIT_EXTRACT, and the 10 VELVETFRUIT_EXTRACT_VOUCHER contracts with their posted position limits.
• Treats Hydrogel as a stationary market around the 10,000 anchor, using a slow EMA, L1 imbalance adjustment, and wide crossing thresholds to avoid getting chopped by noise.
• Treats Velvetfruit Extract as a slower mean-reversion product and only crosses when dislocations are meaningfully larger than the typical spread.
• Uses Black-Scholes-Merton as the first-order no-arbitrage benchmark, then overlays a shrunk live smile on the liquid voucher strikes rather than relying on a single constant volatility.
• Keeps the deep ITM vouchers (VEV_4000, VEV_4500) on an intrinsic-value style leash because their extrinsic value is tiny relative to spread noise.
• Concentrates the active voucher sleeve on VEV_5000 and VEV_5100, where the residual signal is strongest, and only keeps the wider strikes on a tight leash for smile context rather than leaning on them for alpha.
• Uses the VEV_5000 / VEV_5100 adjacent-strike spread first, then selective single-name residual trades with VEV_5000 still given the strongest priority.
• Controls voucher inventory with tighter wing caps, same-side option concentration limits, a portfolio delta budget, and persistent drawdown / reduce-only risk gates.
• Deterministic crossing replay over historical days 0, 1, and 2: +186,273.5 XIRECS, with +131,402.0 from Hydrogel, +36,544.5 from Velvetfruit Extract, and +18,327.0 from the voucher sleeve.
• Option-delta snapshot in the current replay: ending delta 24.9, 255.5, and 261.6 by day, with worst absolute option delta about 262.6.
• Metrics snapshot: logs/round3_diagnostics.json.

Round 3 Theory Notes

• Black, Fischer, and Myron Scholes (1973), The Pricing of Options and Corporate Liabilities: classic closed-form European option benchmark. Citation context: Stanford GSB
• Merton, Robert C. (1973), Theory of Rational Option Pricing: extends the no-arbitrage continuous-time foundation. Citation page: Harvard Business School
• Cox, Ross, and Rubinstein (1979), Option Pricing: A Simplified Approach: discrete-time replication intuition, which is especially useful in spread-constrained simulated markets. Citation / metadata: IDEAS RePEc
• Dumas, Fleming, and Whaley (1996/1998), Implied Volatility Functions: Empirical Tests: practical reminder that constant-vol Black-Scholes is a benchmark, not a literal description of live option markets. Working paper: NBER
• Practical conclusion for Solvenar: Black-Scholes is applicable as the anchor model for voucher fair value and delta, but the observed cross-strike smile and discrete trading frictions make a shrunk live smile plus residual trading more robust than a one-vol implementation.

Round 3 Manual Challenge

• Product: ORNAMENTAL_BIO-POD.
• Fair exit value next round: 920.
• Counterparty reserve prices are uniformly distributed on the 5-XIREC grid from 670 to 920.
• Exact discrete manual optimizer under the symmetric rational-field assumption gives the Pareto-dominant equilibrium: Lowest Bid = 750, Highest Bid = 835.
• Game-theory note: the second-bid game has a continuum of symmetric fixed points once the common second bid is high enough, but 835 is the profit-maximizing equilibrium-selection choice because any higher common bid preserves execution while reducing margin for everyone.
• If the field is irrational and systematically overbids, the best response to the observed average second bid moves upward; the notebook records that distinction explicitly.

Round 2 Manual Challenge

• Recommended submission: Research 16%, Scale 49%, Speed 35%.
• Uses the full 50,000 XIREC budget.
• Continuous optimum conditional on 35% Speed: Research 16.17%, Scale 48.83%, Speed 35.00%; the integer submission is easier and nearly identical.
• The 35% Speed allocation is chosen to sit above common integer-balanced 33/33/34 submissions while preserving enough budget for the deterministic Research/Scale engine.
• The allocation clears 200,000 XIRECs if it beats at least 61.7% of Speed investments; at the 70th Speed percentile it scores about 227,948 XIRECs, and at top rank it scores about 329,021 XIRECs.
• Metrics snapshot: logs/round2_manual_invest_expand.json.

Round 2 Strategy

• Active submission file: trader.py.
• Uses data/round2 for the reproducible local research set.
• Bids 825 XIRECs for Market Access Fee coverage. The bid is above the common placeholder/example range while staying close to the measured one-day marginal value of the extra 25% book access in local volume-sensitivity checks.
• Trades ASH_COATED_OSMIUM and INTARIAN_PEPPER_ROOT with the 80-unit position limits.
• Models pepper root as a deterministic drift claim with slope 0.001 * timestamp, calibrated from the live day-open intercept, then builds to the long 80-unit cap while the intercept remains stable.
• Models osmium as stationary around the 10,000 anchor with short-horizon mean-reversion, EMA fair-value smoothing, capped best-level order-book imbalance, active crossing inventory skew, and a crowded-inventory gate that suppresses imbalance when it would add to an already large position.
• Uses risk gates rather than unconditional averaging down: pepper flattens if the live intercept falls more than 35 XIRECs below day-open intercept, and osmium flattens if observed fair value moves more than 35 XIRECs from the stationary anchor.
• Deterministic local replay over days -1, 0, and 1: +249,317 XIRECs, with +238,195 from pepper root and +11,122 from osmium. A trade-print-aware replay reaches +249,352 XIRECs.
• Replay improvement versus archived round 1 parameters on the same round 2 data: +2,636 XIRECs, and +33 XIRECs versus the restored second Round 2 iteration in the quote-only replay.
• Execution-stress Monte Carlo, with 97% crossing-fill probability, up to 1 XIREC adverse slippage, and 3 XIRECs closing mark noise: 5th percentile +245,406.2 XIRECs, 100% of draws above +245,000 XIRECs.
• Metrics snapshot: logs/round2_diagnostics.json.

Round 1 Results

• Overall position: 1359.
• Previous total: 0 XIRECs.
• Algorithmic trading: +94,022.5625 XIRECs, displayed as +94,023, with round ranking 2046th.
• Manual trading: +87,995.10 XIRECs, displayed as +87,995, with round ranking 1st.
• Round 1 total: +182,017.6625 XIRECs, displayed as 182,018.
• New total PnL: 182,018 XIRECs.
• Mission progress: 91%.
• Badges unlocked: 5.
• Manual orders: buy 9,999 DRYLAND_FLAX at +30 for +9,999 PnL; buy 19,999 EMBER_MUSHROOM at +17 for +77,996.10 PnL.
• Metrics snapshot: logs/round1_metrics.md.

Round 1 Strategy

• Trades ASH_COATED_OSMIUM and INTARIAN_PEPPER_ROOT with the 80-unit position limits.
• Models pepper root as a slowly growing product with live-book intercept calibration and a 0.001 * timestamp slope.
• Builds a long pepper inventory early and keeps it through the close to maximize marked-to-market drift capture.
• Treats osmium as a stationary mean-reversion market around 10,000, with an EMA/imbalance microstructure signal and a looser crossing edge.
• Includes model-based stop-loss gates: pepper de-risks if the live drift intercept breaks below the day-open intercept, and osmium de-risks if it leaves the stationary anchor regime.
• Keeps the research notes in logs/round1_strategy_findings.ipynb.

Diagnostics

Run the dependency-free diagnostics harness with:

python scripts\round1_diagnostics.py
python scripts\round2_diagnostics.py --mc-chains 1 --mc-draws 40
python scripts\round3_diagnostics.py
python scripts\manual_auction_backtest.py
python scripts\build_round3_notebook.py

The latest run is saved to logs/round1_diagnostics.json and includes:

• Profit-first crossing-fill replay: 249,384.5 XIRECs across the three historical days, 49,384.5 above the 200,000 XIREC floor.
• Parameter sensitivity grid: 144/144 tested parameter combinations remained above the 200,000 XIREC target.
• Selection protocol: candidates are gated by a train-only robust score on days -2 and -1, then the highest-PnL row inside that train-validated plateau is selected; day 0 is reported as the local holdout stress check.
• Selected parameter set: rank 2/144 by train-only robust score, rank 1/144 by all-days combined PnL, with 83,273 XIRECs on holdout day 0.
• Lightweight linear signal screen: osmium day-0 holdout MSE improves from 8.772 to 7.301, with 70.54% next-tick directional accuracy.
• Risk-control checks: the pepper trend-break and osmium anchor-break guards had zero historical triggers, so they protect live regime breaks without reducing the normal-path backtest.
• Monte Carlo execution stress: 4 chains x 40 draws with 97% crossing-fill probability, up to 1 XIREC adverse slippage, and 3 XIREC mark noise.
• Gelman-Rubin R-hat: 1.0000.
• Geweke checks: all chains passed abs(z) < 2.
• Monte Carlo stressed 5th percentile: 245,764.4 XIRECs, with 100% of draws above 245,000.
• Anderson-Darling and K-S normality checks: neither rejects the Monte Carlo PnL sample at the 5% level.

The manual auction backtest is saved to logs/manual_auction_backtest.json. It exhaustively scans integer bid prices and quantities, confirms the recommended orders are exact optima under the screenshot order books, and records threshold checks around the one-unit quantities.