Fund Optimizations
Today's trillion dollar AI models excel at optimizing millions of parameters at a time while minimizing a loss function. This is how an AI-native hedge fund works. Using proprietary vector embeddings, algorithms and triangulated data the Tuatara hedge fund's AI system is constantly evolving, learning and discovering new edges and alpha in global financial and information markets with selected metrics reported here in real-time.
Every optimization, fine-tuning, pre-training, post-training, calibration and hyperparameter search run across the proprietary set of global algorithms that mine for edge and alpha in global financial markets, with the goal of increasing overall returns. Each record is datestamped and appears here as soon as it lands; algorithm and parameter identities are aliased.
Live running totals
connecting…20,351
Optimizations recorded
0
Pre-training
17
Training run
327
Fine-tuning
19
Post-training
7,456
Calibration
39
Hyperparameter search
199
Optimization
12,294
Evaluation
20,351
Optimizations recorded
5,651
In the last 30 days
19
Algorithm families
136
Registered parameters
Latest datestamp: 2026-10-24T12:00:00Z · first: 2025-12-31T00:05:52Z
Ontology-guided optimizations
Starting from six roles and disciplines, the system walks its own ontology of funds, instruments, risk measures and execution concepts, and every concept it reaches that has a measurable counterpart in the live record fires a rule over real data: Kelly sizing, drawdown limits, implementation shortfall, the out-of-sample gate, the arbitrage band and risk parity across the three fund families combined. A rule proposes a change only when the numbers argue for it; otherwise it says so. Nodes light up as derivations land, in real time.
Starting terms
Concepts reached
Derived optimizations
Walk JSON: https://cymetica.com/api/v1/fund-optimizations/walk
What boosted returns
Which parameters and which optimizations raised returns, measured against the system's own baselines: a search's best candidate against the parameter set it was run against, a basket against its benchmark over the same window, a fine-tuned model against its cold start. Lifts are in percentage points of per-deal return; identities are aliased.
Parameters ranked by attributed lift
| # | Parameter | Family | Changes | Gains | Held-out lift | In-sample lift |
|---|
Optimizations by stage
| Stage | Family | Measure | n | Gains | Lift / detail |
|---|
Method
Growth over time
A race of the stages: each line is one stage of the pipeline drawn forward through time, and the ranking beside it reorders as stages overtake one another. Press play, drag the slider to any date, hover the chart for the figures at that point. The axis grows with the leader; log scale shows the small stages beside the large ones.
Series JSON: https://cymetica.com/api/v1/fund-optimizations/series?bucket=week (bucket = day | week | month)
Ontology
- Stages
- Pre-training — Representation and feature construction before a model sees labels: feature versions, embeddings, universe construction. (0)
- Training run — A full fit of a model on the labelled history from a cold start. (17)
- Fine-tuning — A warm-start fit, a parameter re-tune or an accepted parameter sweep on an existing model or rule set. (327)
- Post-training — Promotion, activation or retirement of a candidate after out-of-sample review. (19)
- Calibration — Mapping raw model scores onto outcome probabilities on held-out samples. (7456)
- Hyperparameter search — A population or grid search over a bounded parameter space behind an out-of-sample overfitting gate. (39)
- Optimization — A proposed or applied change to a live control lever, measured against its baseline. (199)
- Evaluation — A backtest or replay of a rule set over a fixed window against a benchmark. (12294)
- Domains
- Equities — Listed stocks and equity baskets.
- Crypto — Tokens, on-chain venues and perpetuals.
- Derivatives — Perpetual and dated derivatives.
- Events — Headline- and event-driven baskets and cards.
- Cross-asset — Allocation across every strategy family.
- Execution — Order placement, fills, quotes and latency.
- Objectives
- ReturnRisk-adjusted returnNext-day returnDrawdownHit rateCalibration errorFill qualityCapital efficiencyLatency
- Optimizer families
- Genetic searchDifferential evolutionQuality-diversity searchPopulation-based trainingBandit allocationAnalyst-tunedOperator constraint
- Layers
- Signal & dataExecutionLatencyAllocation
- Relations
- Algorithm family has parameter ParameterParameter is searched by Optimizer familyParameter lives in LayerAlgorithm family targets ObjectiveAlgorithm family operates in DomainOptimization is a stage of StageOptimization improves Algorithm familyOptimization is dated by Datestamp
AF-6226
AF-771E
AF-BF5B
AF-2E32
AF-65F1
AF-738D
AF-AF3A
AF-CD08
AF-F276
AF-F80B
AF-0FC1
AF-2401
AF-4023
AF-97DC
AF-BE0C
AF-410F
AF-5C05
AF-C2E5
AF-EF2A
Parameters are listed per family in the ontology feed with their kind, bounds flag, optimizer family and whether they are an operator constraint that is never searched.
Optimization ledger
OPT-F2D548CACF2026-10-24T12:00:00Z
Ontology-guided optimization: fund manager → AI Portfolio Rebalancing reached the capital weight of algorithm family AF-1088; risk parity across the three funds rule: held — the measured evidence does not justify a change.
days=28 · path=['fund manager', 'AI Portfolio Rebalancing'] · rule=Risk parity across the three funds · lever=capital weight · nodes=['instr:ai-portfolio-rebalancing', 'instr:fund-of-funds', 'instr:global-macro', 'instr:hedge-fund', 'instr:index-fund', 'instr:long-short-equity', 'instr:quant-systematic', 'master:ai-portfolio-rebalancing', 'master:fund-of-funds', 'master:global-macro', 'master:hedge-fund', 'master:index-fund', 'master:long-short-equity', 'master:quant-systematic'] · paths=7 · outcome=held · excluded=True · direction=held · daily_volatility=0.0004
OPT-D849F82A2C2026-10-24T12:00:00Z
Ontology-guided optimization: fund manager → AI Portfolio Rebalancing reached the capital weight of algorithm family AF-1B63; risk parity across the three funds rule: held — the measured evidence does not justify a change.
days=28 · path=['fund manager', 'AI Portfolio Rebalancing'] · rule=Risk parity across the three funds · lever=capital weight · nodes=['instr:ai-portfolio-rebalancing', 'instr:fund-of-funds', 'instr:global-macro', 'instr:hedge-fund', 'instr:index-fund', 'instr:long-short-equity', 'instr:quant-systematic', 'master:ai-portfolio-rebalancing', 'master:fund-of-funds', 'master:global-macro', 'master:hedge-fund', 'master:index-fund', 'master:long-short-equity', 'master:quant-systematic'] · paths=7 · weight=0.404 · outcome=held · direction=held · current_share=0.47 · funds_weighted=2 · daily_volatility=0.0127
OPT-B57120B4F52026-10-24T12:00:00Z
Ontology-guided optimization: fund manager → AI Portfolio Rebalancing reached the capital weight of algorithm family AF-BDFF; risk parity across the three funds rule: held — the measured evidence does not justify a change.
days=28 · path=['fund manager', 'AI Portfolio Rebalancing'] · rule=Risk parity across the three funds · lever=capital weight · nodes=['instr:ai-portfolio-rebalancing', 'instr:fund-of-funds', 'instr:global-macro', 'instr:hedge-fund', 'instr:index-fund', 'instr:long-short-equity', 'instr:quant-systematic', 'master:ai-portfolio-rebalancing', 'master:fund-of-funds', 'master:global-macro', 'master:hedge-fund', 'master:index-fund', 'master:long-short-equity', 'master:quant-systematic'] · paths=7 · weight=0.596 · outcome=held · direction=held · current_share=0.53 · funds_weighted=2 · daily_volatility=0.0086
OPT-C4E58F404D2026-10-24T12:00:00Z
Ontology-guided optimization: quantitative financial analyst → Sharpe Ratio reached the volatility band of algorithm family AF-BDFF; out-of-sample gate rule: held — the measured evidence does not justify a change.
path=['quantitative financial analyst', 'Sharpe Ratio'] · rule=Out-of-sample gate · lever=volatility band · nodes=['instr:quant-systematic', 'instr:sharpe-ratio', 'instr:sortino-ratio', 'master:metric-sharpe', 'master:metric-sortino', 'master:quant-systematic', 'master:sharpe-ratio', 'master:sortino-ratio'] · paths=3 · outcome=held · direction=held · band_in_use=0.35 · candidate_band=0.622 · held_out_deals=14 · held_out_lift_pp=-0.382 · overfitting_probability=0.6
OPT-FF9059A7002026-10-24T12:00:00Z
Ontology-guided optimization: High Frequency Trading algorithm → Best Execution reached the position size of algorithm family AF-BDFF; implementation shortfall rule: held — the measured evidence does not justify a change.
path=['High Frequency Trading algorithm', 'Best Execution'] · rule=Implementation shortfall · fills=30 · lever=position size · nodes=['clob:best-execution', 'clob:latency-routing', 'clob:multi-venue-execution', 'clob:slippage', 'clob:tick-to-trade-latency', 'master:clob:best-execution', 'master:clob:latency-routing', 'master:clob:multi-venue-execution', 'master:clob:slippage', 'master:clob:tick-to-trade-latency', 'master:exec-best-execution'] · paths=5 · outcome=held · direction=held · buy_adverse_bps=-21.0 · mean_adverse_bps=-11.0 · sell_adverse_bps=2.2
OPT-15A2DC9B492026-10-24T12:00:00Z
Ontology-guided optimization: quantitative financial analyst → Maximum Drawdown reached the rebalance band of algorithm family AF-1088; drawdown limit rule: held — the measured evidence does not justify a change.
days=28 · path=['quantitative financial analyst', 'Maximum Drawdown'] · rule=Drawdown limit · lever=rebalance band · nodes=['instr:max-drawdown', 'instr:sortino-ratio', 'master:max-drawdown', 'master:metric-max-drawdown', 'master:metric-sortino', 'master:sortino-ratio'] · paths=3 · outcome=held · direction=held · limit_share=0.05 · drawdown_share=0.002 · max_drawdown_share=0.002
OPT-21E0074C522026-10-24T12:00:00Z
Ontology-guided optimization: quantitative financial analyst → Maximum Drawdown reached the position size of algorithm family AF-BDFF; drawdown limit rule: held — the measured evidence does not justify a change.
days=28 · path=['quantitative financial analyst', 'Maximum Drawdown'] · rule=Drawdown limit · lever=position size · nodes=['instr:max-drawdown', 'instr:sortino-ratio', 'master:max-drawdown', 'master:metric-max-drawdown', 'master:metric-sortino', 'master:sortino-ratio'] · paths=3 · outcome=held · direction=held · limit_share=0.05 · drawdown_share=0.004 · max_drawdown_share=0.032
OPT-375BEF26CA2026-10-24T12:00:00Z
Ontology-guided optimization: hedge fund manager → Hedge Fund reached the position size of algorithm family AF-BDFF; kelly-sized position rule: proposed under shadow mode (measured on the next day's real-money net): lowered (x0.34).
days=28 · path=['hedge fund manager', 'Hedge Fund'] · rule=Kelly-sized position · wins=10 · fills=30 · lever=position size · nodes=['instr:hedge-fund', 'instr:kelly-criterion', 'instr:quant-systematic', 'master:hedge-fund', 'master:kelly-criterion', 'master:quant-systematic'] · paths=3 · losses=17 · outcome=proposed · direction=lowered · win_share=0.37 · half_kelly=0.134 · payoff_ratio=6.179 · kelly_fraction=0.268 · current_leg_share=0.392
OPT-90677033462026-10-24T03:04:50.776627Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0205 vs 0.0209 base — improved on the base rate.
improved=True · base_error=0.02089 · test_samples=3000 · train_samples=7000 · calibration_error=0.02048
OPT-A360FD9BE52026-10-24T03:04:50.776627Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0239 vs 0.0241 base — improved on the base rate.
improved=True · base_error=0.0241 · test_samples=3000 · train_samples=7000 · calibration_error=0.02387
OPT-505FE3A8672026-10-24T02:49:16.073548Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0205 vs 0.0209 base — improved on the base rate.
improved=True · base_error=0.02089 · test_samples=3000 · train_samples=7000 · calibration_error=0.0205
OPT-8EF3F971402026-10-24T02:49:16.073548Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0239 vs 0.0241 base — improved on the base rate.
improved=True · base_error=0.0241 · test_samples=3000 · train_samples=7000 · calibration_error=0.02387
OPT-E9C6D5331A2026-10-24T02:32:56.464951Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0205 vs 0.0209 base — improved on the base rate.
improved=True · base_error=0.02089 · test_samples=3000 · train_samples=7000 · calibration_error=0.02053
OPT-D4B885327E2026-10-24T02:32:56.464951Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0245 vs 0.0247 base — improved on the base rate.
improved=True · base_error=0.02472 · test_samples=3000 · train_samples=7000 · calibration_error=0.02449
OPT-AD68911F5D2026-10-24T02:17:13.462783Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0200 vs 0.0203 base — improved on the base rate.
improved=True · base_error=0.02026 · test_samples=3000 · train_samples=7000 · calibration_error=0.02
OPT-CF50CE8FD92026-10-24T02:17:13.462783Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0245 vs 0.0247 base — improved on the base rate.
improved=True · base_error=0.02473 · test_samples=3000 · train_samples=7000 · calibration_error=0.02446
OPT-B496D3F2642026-10-24T02:02:05.394424Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0203 vs 0.0206 base — improved on the base rate.
improved=True · base_error=0.02058 · test_samples=3000 · train_samples=7000 · calibration_error=0.02028
OPT-2A72EBA9AD2026-10-24T02:02:05.394424Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0248 vs 0.0250 base — improved on the base rate.
improved=True · base_error=0.02504 · test_samples=3000 · train_samples=7000 · calibration_error=0.02477
OPT-E106FAF11A2026-10-24T01:45:50.487144Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0203 vs 0.0206 base — improved on the base rate.
improved=True · base_error=0.02058 · test_samples=3000 · train_samples=7000 · calibration_error=0.02028
OPT-34896E27272026-10-24T01:45:50.487144Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0248 vs 0.0250 base — improved on the base rate.
improved=True · base_error=0.02505 · test_samples=3000 · train_samples=7000 · calibration_error=0.02477
OPT-7CFFDD37C42026-10-24T01:29:47.770466Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0203 vs 0.0206 base — improved on the base rate.
improved=True · base_error=0.02058 · test_samples=3000 · train_samples=7000 · calibration_error=0.02027
OPT-DD599B41862026-10-24T01:29:47.770466Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0254 vs 0.0257 base — improved on the base rate.
improved=True · base_error=0.02567 · test_samples=3000 · train_samples=7000 · calibration_error=0.02538
OPT-9898901EE42026-10-24T01:13:45.459367Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0205 vs 0.0209 base — improved on the base rate.
improved=True · base_error=0.02089 · test_samples=3000 · train_samples=7000 · calibration_error=0.0205
OPT-E9A313C7AC2026-10-24T01:13:45.459367Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0254 vs 0.0257 base — improved on the base rate.
improved=True · base_error=0.02567 · test_samples=3000 · train_samples=7000 · calibration_error=0.02538
OPT-4ADBE725502026-10-24T00:58:33.691295Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0206 vs 0.0209 base — improved on the base rate.
improved=True · base_error=0.02089 · test_samples=3000 · train_samples=7000 · calibration_error=0.02061
OPT-9A762531FD2026-10-24T00:58:33.691295Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0254 vs 0.0257 base — improved on the base rate.
improved=True · base_error=0.02567 · test_samples=3000 · train_samples=7000 · calibration_error=0.02538
OPT-F3256FF7CF2026-10-24T00:42:21.794713Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0200 vs 0.0203 base — improved on the base rate.
improved=True · base_error=0.02026 · test_samples=3000 · train_samples=7000 · calibration_error=0.01996
OPT-D59CADCFDD2026-10-24T00:42:21.794713Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0254 vs 0.0257 base — improved on the base rate.
improved=True · base_error=0.02567 · test_samples=3000 · train_samples=7000 · calibration_error=0.02537
OPT-2AC4B680FD2026-10-24T00:26:13.397151Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0193 vs 0.0196 base — improved on the base rate.
improved=True · base_error=0.01963 · test_samples=3000 · train_samples=7000 · calibration_error=0.01932
OPT-3CB610C22B2026-10-24T00:26:13.397151Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0254 vs 0.0257 base — improved on the base rate.
improved=True · base_error=0.02567 · test_samples=3000 · train_samples=7000 · calibration_error=0.02537
OPT-4334423C0A2026-10-24T00:09:54.116882Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0194 vs 0.0196 base — improved on the base rate.
improved=True · base_error=0.01963 · test_samples=3000 · train_samples=7000 · calibration_error=0.01939
OPT-C296B25E372026-10-24T00:09:54.116882Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0257 vs 0.0260 base — improved on the base rate.
improved=True · base_error=0.02598 · test_samples=3000 · train_samples=7000 · calibration_error=0.02569
OPT-A480B0B4562026-10-24T00:09:11.474145Z
Optimization of the position size for algorithm family AF-BDFF: set; proposed under shadow mode (measured, not applied).
mode=shadow · status=proposed · direction=set
OPT-E2DA2EA3D52026-10-23T23:54:48.663632Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0204 vs 0.0206 base — improved on the base rate.
improved=True · base_error=0.02057 · test_samples=3000 · train_samples=7000 · calibration_error=0.02044
OPT-A4C37B2DA42026-10-23T23:54:48.663632Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0247 vs 0.0250 base — improved on the base rate.
improved=True · base_error=0.02505 · test_samples=3000 · train_samples=7000 · calibration_error=0.02475
OPT-86CE59A3E52026-10-23T23:38:32.586588Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0204 vs 0.0206 base — improved on the base rate.
improved=True · base_error=0.02057 · test_samples=3000 · train_samples=7000 · calibration_error=0.02043
OPT-AEF456D7EA2026-10-23T23:38:32.586588Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0247 vs 0.0250 base — improved on the base rate.
improved=True · base_error=0.02505 · test_samples=3000 · train_samples=7000 · calibration_error=0.02475
OPT-C04D1B7A432026-10-23T23:33:44.632385Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0202 vs 0.0203 base — improved on the base rate.
improved=True · base_error=0.02026 · test_samples=3000 · train_samples=7000 · calibration_error=0.02019
OPT-7155684A602026-10-23T23:33:44.632385Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0247 vs 0.0250 base — improved on the base rate.
improved=True · base_error=0.02505 · test_samples=3000 · train_samples=7000 · calibration_error=0.02475
OPT-C7CAFF85112026-10-23T23:17:40.906101Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0184 vs 0.0184 base — no gain over the base rate.
improved=False · base_error=0.01835 · test_samples=3000 · train_samples=7000 · calibration_error=0.01836
OPT-1A201A55472026-10-23T23:17:40.906101Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0247 vs 0.0250 base — improved on the base rate.
improved=True · base_error=0.02505 · test_samples=3000 · train_samples=7000 · calibration_error=0.02475
OPT-C8D773B3D32026-10-23T23:01:56.981504Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0180 vs 0.0180 base — improved on the base rate.
improved=True · base_error=0.01805 · test_samples=3000 · train_samples=7000 · calibration_error=0.01797
OPT-AAD2F978DF2026-10-23T23:01:56.981504Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0244 vs 0.0247 base — improved on the base rate.
improved=True · base_error=0.02474 · test_samples=3000 · train_samples=7000 · calibration_error=0.02443
OPT-C99899DD292026-10-23T22:57:18.491710Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0182 vs 0.0184 base — improved on the base rate.
improved=True · base_error=0.01836 · test_samples=3000 · train_samples=7000 · calibration_error=0.01818
OPT-60A23897552026-10-23T22:57:18.491710Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0244 vs 0.0247 base — improved on the base rate.
improved=True · base_error=0.02474 · test_samples=3000 · train_samples=7000 · calibration_error=0.02443
OPT-CAE5D447E62026-10-23T22:47:02.548842Z
Post-training promotion of a new candidate parameter set for algorithm family AF-4023 after out-of-sample review (n = 11); fitness score +0.30.
n=11 · fitness=0.297
OPT-6D7D5869F32026-10-23T22:40:56.023750Z
Calibration of the BlackBox-611 (algorithm family AF-19F3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0185 vs 0.0187 base — improved on the base rate.
improved=True · base_error=0.01868 · test_samples=3000 · train_samples=7000 · calibration_error=0.01848
OPT-7B6A02F95D2026-10-23T22:40:56.023750Z
Calibration of the BlackBox-611 (algorithm family AF-54A3): monotone fit on 7,000 training / 3,000 held-out samples; calibration error 0.0244 vs 0.0247 base — improved on the base rate.
improved=True · base_error=0.02474 · test_samples=3000 · train_samples=7000 · calibration_error=0.02443
OPT-BCBECFF2632026-10-23T22:36:01.423703Z
Optimization of the position size for algorithm family AF-BDFF: set; proposed under shadow mode (measured, not applied).
mode=shadow · status=proposed · direction=set
OPT-6D7D5C91952026-10-23T22:30:27.583938Z
Hyperparameter search over the parameter space of algorithm family AF-4023: 40 generations x 48 candidates, 1,743 evaluations on 324 training / 99 held-out deals.
population=48 · evaluations=1743 · generations=40 · train_deals=324 · holdout_deals=99
JSON feed: https://cymetica.com/api/v1/fund-optimizations · summary: https://cymetica.com/api/v1/fund-optimizations/summary · ontology: https://cymetica.com/api/v1/fund-optimizations/ontology