Models

Learning & Model Performance

Versioned model registry, time-split OOS promotion, Yahoo 1d/3d/5d settlement and probability calibration.

Bu sayfa ne işe yarıyor?

Model kayıt defteri, kalibrasyon, promotion. Challenger otomatik production olmaz.

Hangi soruya cevap veriyor?

Hangi model canlı, hangisi reddedildi, hangisi bozuluyor?

Kararda nasıl kullanılmalı?

REJECTED adaptive sinyal AL üretmez. Promotion ayrı onay ister.

Hangi motora veri sağlar?

Strategy registry · Live scorecard (sonraki dalga)

Learning Loop

Model Governance

Thirty-day retraining clock, operational drift versus held-out TEST, and explicit champion/challenger promotion. Training never promotes automatically.

Model	Role	Retrain	Drift	Approval
Adaptive Meta ASELS vadaptive.2026.08.17T010726	OTHER	DUE 2026-09-16	NEEDS_SAMPLE at least 30 operational outcomes required	—
Direction Logistic v2026.08.16T144022	OTHER	DUE 2026-09-15	NEEDS_SAMPLE at least 30 operational outcomes required	—
LSTM v0.1.0	OTHER	DUE no successful run	NEEDS_SAMPLE at least 30 operational outcomes required	—
Macro Model v0.1.0	OTHER	DUE no successful run	NEEDS_SAMPLE at least 30 operational outcomes required	—
Momentum v0.1.0	OTHER	DUE no successful run	NEEDS_SAMPLE at least 30 operational outcomes required	—
News Model v0.1.0	OTHER	DUE no successful run	NEEDS_SAMPLE at least 30 operational outcomes required	—
Transformer v0.1.0	OTHER	DUE no successful run	NEEDS_SAMPLE at least 30 operational outcomes required	—
XGBoost v0.1.0	OTHER	DUE no successful run	NEEDS_SAMPLE at least 30 operational outcomes required	—

Model

Role

Retrain

Drift

Approval

Adaptive Meta ASELS

vadaptive.2026.08.17T010726

OTHER

DUE

2026-09-16

NEEDS_SAMPLE

at least 30 operational outcomes required

—

Direction Logistic

v2026.08.16T144022

OTHER

DUE

2026-09-15

NEEDS_SAMPLE

at least 30 operational outcomes required

—

LSTM

v0.1.0

OTHER

DUE

no successful run

NEEDS_SAMPLE

at least 30 operational outcomes required

—

Macro Model

v0.1.0

OTHER

DUE

no successful run

NEEDS_SAMPLE

at least 30 operational outcomes required

—

Momentum

v0.1.0

OTHER

DUE

no successful run

NEEDS_SAMPLE

at least 30 operational outcomes required

—

News Model

v0.1.0

OTHER

DUE

no successful run

NEEDS_SAMPLE

at least 30 operational outcomes required

—

Transformer

v0.1.0

OTHER

DUE

no successful run

NEEDS_SAMPLE

at least 30 operational outcomes required

—

XGBoost

v0.1.0

OTHER

DUE

no successful run

NEEDS_SAMPLE

at least 30 operational outcomes required

—

Model Success

Calibration

Adaptive Meta ASELS

adaptive-meta-logit · vadaptive.2026.08.17T010726

3 predictions0 stock linksREJECTED

accuracy: —
brier: —
log loss: —
metric sample: 1
operational: 5d · n=1
operational brier/logloss: 0.374 / 0.946
sharpe: —
cal buckets: 2
artifact: adaptive-meta-logit-v1
latest train: SUCCESS

1d:0.5-0.6: pred 0.54 → act 0.00
5d:0.0-0.4: pred 0.39 → act 1.00

Direction Logistic

trained-logit · v2026.08.16T144022

0 predictions0 stock linksREJECTED

accuracy: 0.535
brier: 0.252
log loss: 0.697
metric sample: 572
operational: —
operational brier/logloss: — / —
sharpe: —
cal buckets: 5
artifact: direction-logit-v1
latest train: SUCCESS

TEST:1d:0.0-0.4: pred 0.37 → act 0.52
TEST:1d:0.4-0.5: pred 0.44 → act 0.47
TEST:1d:0.5-0.6: pred 0.53 → act 0.66
TEST:1d:0.6-0.7: pred 0.62 → act 1.00

LSTM

heuristic-sequence · v0.1.0

122 predictions30 stock linksHEURISTIC

accuracy: 0.380
brier: 0.318
log loss: 0.849
metric sample: 92
operational: 5d · n=92
operational brier/logloss: 0.318 / 0.849
sharpe: —
cal buckets: 12
artifact: none
latest train: none

1d:0.0-0.4: pred 0.30 → act 0.56
1d:0.4-0.5: pred 0.46 → act 0.80
1d:0.5-0.6: pred 0.55 → act 0.77
1d:0.6-0.7: pred 0.65 → act 0.36

Macro Model

heuristic-macro · v0.1.0

122 predictions30 stock linksHEURISTIC

accuracy: 0.370
brier: 0.337
log loss: 0.888
metric sample: 92
operational: 5d · n=92
operational brier/logloss: 0.337 / 0.888
sharpe: —
cal buckets: 12
artifact: none
latest train: none

1d:0.0-0.4: pred 0.27 → act 0.60
1d:0.4-0.5: pred 0.44 → act 0.62
1d:0.5-0.6: pred 0.55 → act 0.67
1d:0.6-0.7: pred 0.66 → act 0.50

Momentum

heuristic-technical · v0.1.0

122 predictions30 stock linksHEURISTIC

accuracy: 0.467
brier: 0.290
log loss: 0.786
metric sample: 92
operational: 5d · n=92
operational brier/logloss: 0.290 / 0.786
sharpe: —
cal buckets: 12
artifact: none
latest train: none

1d:0.0-0.4: pred 0.30 → act 0.50
1d:0.4-0.5: pred 0.45 → act 0.67
1d:0.5-0.6: pred 0.54 → act 0.75
1d:0.6-0.7: pred 0.65 → act 0.55

News Model

heuristic-nlp · v0.1.0

122 predictions30 stock linksHEURISTIC

accuracy: 0.446
brier: 0.254
log loss: 0.702
metric sample: 92
operational: 5d · n=92
operational brier/logloss: 0.254 / 0.702
sharpe: —
cal buckets: 6
artifact: none
latest train: none

1d:0.4-0.5: pred 0.48 → act 0.56
1d:0.5-0.6: pred 0.53 → act 0.66
1d:0.6-0.7: pred 0.64 → act 0.67
5d:0.4-0.5: pred 0.48 → act 0.59

Transformer

heuristic-sequence · v0.1.0

122 predictions30 stock linksHEURISTIC

accuracy: 0.380
brier: 0.318
log loss: 0.849
metric sample: 92
operational: 5d · n=92
operational brier/logloss: 0.318 / 0.849
sharpe: —
cal buckets: 12
artifact: none
latest train: none

1d:0.0-0.4: pred 0.30 → act 0.56
1d:0.4-0.5: pred 0.46 → act 0.80
1d:0.5-0.6: pred 0.55 → act 0.77
1d:0.6-0.7: pred 0.65 → act 0.36

XGBoost

heuristic-tabular · v0.1.0

122 predictions30 stock linksHEURISTIC

accuracy: 0.435
brier: 0.293
log loss: 0.792
metric sample: 92
operational: 5d · n=92
operational brier/logloss: 0.293 / 0.792
sharpe: —
cal buckets: 12
artifact: none
latest train: none

1d:0.0-0.4: pred 0.32 → act 0.60
1d:0.4-0.5: pred 0.44 → act 0.56
1d:0.5-0.6: pred 0.55 → act 0.73
1d:0.6-0.7: pred 0.64 → act 0.50

Directional Probability Buckets

Model	Range	Predicted	Actual	Gap
XGBoost	1d:0.0-0.4	0.315	0.600	0.285
Transformer	1d:0.0-0.4	0.300	0.561	0.261
Momentum	1d:0.0-0.4	0.298	0.500	0.202
Macro Model	1d:0.0-0.4	0.274	0.596	0.322
LSTM	1d:0.0-0.4	0.300	0.561	0.261
XGBoost	1d:0.4-0.5	0.443	0.563	0.119
Transformer	1d:0.4-0.5	0.458	0.800	0.342
News Model	1d:0.4-0.5	0.480	0.564	0.084
Momentum	1d:0.4-0.5	0.451	0.667	0.215
Macro Model	1d:0.4-0.5	0.442	0.615	0.173
LSTM	1d:0.4-0.5	0.458	0.800	0.342
XGBoost	1d:0.5-0.6	0.550	0.727	0.177
Transformer	1d:0.5-0.6	0.547	0.769	0.222
News Model	1d:0.5-0.6	0.534	0.660	0.126
Momentum	1d:0.5-0.6	0.535	0.750	0.215
Macro Model	1d:0.5-0.6	0.549	0.667	0.118
LSTM	1d:0.5-0.6	0.547	0.769	0.222
Adaptive Meta ASELS	1d:0.5-0.6	0.542	0.000	-0.542
XGBoost	1d:0.6-0.7	0.641	0.500	-0.141
Transformer	1d:0.6-0.7	0.649	0.364	-0.285
News Model	1d:0.6-0.7	0.643	0.667	0.023
Momentum	1d:0.6-0.7	0.650	0.545	-0.105
Macro Model	1d:0.6-0.7	0.662	0.500	-0.162
LSTM	1d:0.6-0.7	0.649	0.364	-0.285
XGBoost	1d:0.7-0.8	0.736	0.667	-0.069
Transformer	1d:0.7-0.8	0.767	0.750	-0.017
Momentum	1d:0.7-0.8	0.743	0.533	-0.210
Macro Model	1d:0.7-0.8	0.738	0.750	0.012
LSTM	1d:0.7-0.8	0.767	0.750	-0.017
XGBoost	1d:0.8-1.0	0.820	0.714	-0.106
Transformer	1d:0.8-1.0	0.819	0.667	-0.152
Momentum	1d:0.8-1.0	0.818	0.692	-0.126
Macro Model	1d:0.8-1.0	0.820	0.625	-0.195
LSTM	1d:0.8-1.0	0.819	0.667	-0.152
XGBoost	5d:0.0-0.4	0.315	0.500	0.185
Transformer	5d:0.0-0.4	0.300	0.585	0.286
Momentum	5d:0.0-0.4	0.298	0.375	0.077
Macro Model	5d:0.0-0.4	0.274	0.596	0.322
LSTM	5d:0.0-0.4	0.300	0.585	0.286
Adaptive Meta ASELS	5d:0.0-0.4	0.388	1.000	0.612
XGBoost	5d:0.4-0.5	0.443	0.625	0.182
Transformer	5d:0.4-0.5	0.458	0.800	0.342
News Model	5d:0.4-0.5	0.480	0.590	0.109
Momentum	5d:0.4-0.5	0.451	0.714	0.263
Macro Model	5d:0.4-0.5	0.442	0.692	0.250
LSTM	5d:0.4-0.5	0.458	0.800	0.342
XGBoost	5d:0.5-0.6	0.550	0.591	0.041
Transformer	5d:0.5-0.6	0.547	0.385	-0.163
News Model	5d:0.5-0.6	0.534	0.480	-0.054
Momentum	5d:0.5-0.6	0.535	0.625	0.090
Macro Model	5d:0.5-0.6	0.549	0.417	-0.132
LSTM	5d:0.5-0.6	0.547	0.385	-0.163
XGBoost	5d:0.6-0.7	0.641	0.333	-0.308
Transformer	5d:0.6-0.7	0.649	0.455	-0.194
News Model	5d:0.6-0.7	0.643	0.333	-0.310
Momentum	5d:0.6-0.7	0.650	0.364	-0.287
Macro Model	5d:0.6-0.7	0.662	0.000	-0.662
LSTM	5d:0.6-0.7	0.649	0.455	-0.194
XGBoost	5d:0.7-0.8	0.736	0.444	-0.292
Transformer	5d:0.7-0.8	0.767	0.375	-0.392

Recent Outcomes

Stock	Model	Hz	Actual	Direction	Error
PETKM	LSTM	5d	0.017	Miss	0.024
PETKM	Macro Model	5d	0.017	Miss	0.024
PETKM	Momentum	5d	0.017	Miss	0.024
PETKM	News Model	5d	0.017	Miss	0.024
PETKM	Transformer	5d	0.017	Miss	0.024
PETKM	XGBoost	5d	0.017	Hit	0.024
QUAGR	LSTM	5d	-0.103	Hit	0.082
QUAGR	Macro Model	5d	-0.103	Hit	0.082
QUAGR	Momentum	5d	-0.103	Hit	0.082
QUAGR	News Model	5d	-0.103	Hit	0.082
QUAGR	Transformer	5d	-0.103	Hit	0.082
QUAGR	XGBoost	5d	-0.103	Hit	0.082
RALYH	LSTM	5d	-0.021	Miss	0.026
RALYH	Macro Model	5d	-0.021	Miss	0.026
RALYH	Momentum	5d	-0.021	Hit	0.026
RALYH	News Model	5d	-0.021	Hit	0.026
RALYH	Transformer	5d	-0.021	Miss	0.026
RALYH	XGBoost	5d	-0.021	Miss	0.026
TKFEN	LSTM	5d	-0.047	Miss	0.114
TKFEN	Macro Model	5d	-0.047	Miss	0.114
TKFEN	Momentum	5d	-0.047	Miss	0.114
TKFEN	News Model	5d	-0.047	Miss	0.114
TKFEN	Transformer	5d	-0.047	Miss	0.114
TKFEN	XGBoost	5d	-0.047	Miss	0.114
TUKAS	LSTM	5d	-0.045	Miss	0.050
TUKAS	Macro Model	5d	-0.045	Miss	0.050
TUKAS	Momentum	5d	-0.045	Miss	0.050
TUKAS	News Model	5d	-0.045	Miss	0.050
TUKAS	Transformer	5d	-0.045	Miss	0.050
TUKAS	XGBoost	5d	-0.045	Miss	0.050
TUPRS	LSTM	5d	0.027	Hit	0.021
TUPRS	Macro Model	5d	0.027	Hit	0.021
TUPRS	Momentum	5d	0.027	Hit	0.021
TUPRS	News Model	5d	0.027	Hit	0.021
TUPRS	Transformer	5d	0.027	Hit	0.021
TUPRS	XGBoost	5d	0.027	Hit	0.021
TUREX	LSTM	5d	-0.095	Hit	0.074
TUREX	Macro Model	5d	-0.095	Hit	0.074
TUREX	Momentum	5d	-0.095	Hit	0.074
TUREX	News Model	5d	-0.095	Hit	0.074