gann2 · Monte-Carlo sweep

pool_dial_sweep4_final

12 / 3,000 models done (still running) · ? walk-forward windows · score mode ? · kept top 12 · typical trade count of profitable models on this chart: 1428 · 2026-09-18T18:52:59

Score across every model sampled

best+155.873

top 10%+112.408

median+80.849

bottom 10%+28.772

worst+12.784

The line is how many models scored in each slice of the range. Left tail = the models that lost the most; the bump on the right = the few that came closest to an edge.

Top 10 models — return on their own capital, overlaid

Each line is one model's account if you had traded it straight through every walk-forward validation window, Jan 2023 – Sep 2026. the compounded result, which is what the score now ranks on. Click a line (or its legend entry) to isolate it; click anywhere else to bring them all back — then use ↑/↓ to step through the lines by final return. Click a colour swatch (in the legend or the table below) to hide/unhide that one line on its own, without isolating it -- the top-3-pooled line included, the same way. When isolated, each window shows that model's trade count, rupee P&L, and how much its account grew over the window (hover for the detail). A window can show a real loss even while the line above it keeps rising — the line is one running number shared across every window, so a bad early window can crush it near zero, and later windows then multiply a near-zero number no matter how well they do. Money moved to safety is not stretched that way — it is added straight in real rupees — so it can keep pushing the line up even while the trading side stays crushed. 10 of the lines drawn have capital protection on — their curve already includes money safely set aside along the way, not just what is still being traded.

Return on each model’s own starting capital · 2 Jan 2023 → 11 Sep 2026

#6 pooled entry, shared bank, score +155.86 (1446 trades, 15/15 windows) +14622.74%

#3 pooled entry, shared bank, score +112.38 (1438 trades, 15/15 windows) +10277.17%

#5 pooled entry, shared bank, score +107.40 (1444 trades, 15/15 windows) +9775.11%

#4 pooled entry, shared bank, score +99.65 (1433 trades, 15/15 windows) +9004.02%

#0 pooled entry, shared bank, score +92.89 (1440 trades, 15/15 windows) +8320.12%

#11 pooled entry, shared bank, score +80.83 (1450 trades, 15/15 windows) +7119.63%

#7 pooled entry, shared bank, score +66.62 (1440 trades, 15/15 windows) +5701.95%

#2 pooled entry, shared bank, score +63.04 (1429 trades, 15/15 windows) +5345.58%

#9 pooled entry, shared bank, score +62.06 (1443 trades, 15/15 windows) +5229.12%

#10 pooled entry, shared bank, score +28.77 (1447 trades, 15/15 windows) +1894.39%

#00 Balanced baseline +1007.06%

Top 10 changes

model 11 replaced model 10 in the top 10 — 18 Sep 2026, 18:53

The kept models, in full

Click any column heading to sort by it — first click sorts high → low, click again for low → high. Four ways of ranking the same models — pick a tab.

The scoring described above — the compounded account result, floored for a wipe-out, with small credit for steadiness and real (not one-trade) profit, and small docks for barely trading, trading in only a slice of the years tested, or getting lucky on a couple of trades.

model iThe model's number is the order it was tested in the sweep -- it never changes. The table is SORTED best-first, so row 1 is not model 1.	score	account end	set aside	profit factor	at 1.5×	fragility dock	top-3 / profit	fluke dock	thin dock	peak	gave back	raw	ret/dep	win%	trades	windows	overnight	entry	bank	conf_size_low_pct	conf_size_high_pct	conf_size_min_mult	conf_size_max_mult	active_capital_selectivity_strength	active_capital_max_share	active_capital_winner_top_share	throttle_low_winrate	throttle_high_winrate	throttle_min_mult	throttle_max_mult	throttle_window_trades	variant	confidence_source	active_capital_allocation_mode	confidence_sizing	active_capital_sizing	win_rate_throttle_enabled
00	+19.904	+1007.1% 55,35,298	22,39,361	1.37	+622%	—	17%	—	−0.30	67,05,580 17% given back	—	—	+5.5%	52%	1462	15/15	1216	pooled	off	0.3	0.7	0.5	1.5	0	1	0.6						neither	predictor_else_form	proportional	False	False
6	+155.864	+14622.7% 7,36,13,702	4,37,65,630	1.07	+2062%	—	68%	—	−0.30	15,07,74,138 51% given back	−0.01	—	+2.9%	51%	1446	15/15	1200	pooled	shared	0.3053	0.7499	0.2766	2.328	0.9829	0.9232	0.7275	0.269	0.5385	0.635	1.285	11	both	predictor_else_form	winner_take_most	True	True	False
3	+112.377	+10277.2% 5,18,85,867	3,28,09,971	1.06	+871%	—	68%	—	−0.30	11,27,22,269 54% given back	−0.03	—	+2.6%	51%	1438	15/15	1192	pooled	shared	0.2531	0.629	0.4513	1.624	0.9663	0.6575	0.7163	0.2498	0.7083	0.6108	1.084	28	both	predictor_else_form	winner_take_most	True	True	False
5	+107.398	+9775.1% 4,93,75,572	3,01,91,109	1.07	+455%	—	64%	—	−0.30	10,38,13,813 52% given back	−0.02	—	+2.7%	51%	1444	15/15	1198	pooled	shared	0.4281	0.7186	0.761	2.327	0.6531	0.7325	0.8773	0.3412	0.6788	0.758	1.447	32	both	form_only	proportional	True	True	True
4	+99.647	+9004.0% 4,55,20,119	2,87,67,549	1.06	+696%	—	69%	—	−0.30	9,86,20,536 54% given back	−0.03	—	+2.6%	51%	1433	15/15	1187	pooled	shared	0.3825	0.6442	0.7926	2.049	0.3069	0.654	0.9287	0.1844	0.6301	0.4477	1.148	16	both	predictor_else_form	winner_take_most	True	True	False
0	+92.887	+8320.1% 4,21,00,605	2,37,06,467	1.09	+496%	—	60%	—	−0.30	8,13,62,440 48% given back	—	—	+2.8%	51%	1440	15/15	1194	pooled	shared	0.3048	0.662	0.304	2.274	0.1719	0.6524	0.6381	0.1665	0.7651	0.6958	1.397	17	both	form_only	proportional	True	True	True
11	+80.826	+7119.6% 3,60,98,157	2,24,28,534	1.07	+155%	—	66%	—	−0.30	7,68,23,018 53% given back	−0.02	—	+2.5%	52%	1450	15/15	1204	pooled	shared	0.3197	0.6842	0.6652	2.428	0.9172	0.2067	0.8771	0.3897	0.5397	0.6584	1.066	33	both	predictor_else_form	equal	True	True	True
7	+66.618	+5701.9% 2,90,09,746	1,82,58,305	1.06	+171%	—	69%	—	−0.30	6,26,01,505 54% given back	−0.03	—	+2.4%	51%	1440	15/15	1194	pooled	shared	0.2236	0.8121	0.6617	1.773	0.7934	0.9002	0.8768	0.3595	0.6289	0.7059	1.308	23	both	predictor_else_form	equal	True	True	False
2	+63.037	+5345.6% 2,72,27,919	1,69,32,542	1.06	+105%	—	68%	—	−0.30	5,82,20,330 53% given back	−0.03	—	+2.4%	51%	1429	15/15	1183	pooled	shared	0.2324	0.5776	0.5325	1.488	0.369	0.5582	0.8134	0.1629	0.7813	0.6288	1.391	28	active_only	predictor_else_form	equal	False	True	False
9	+62.061	+5229.1% 2,66,45,598	1,19,90,662	1.19	+4530%	—	52%	—	−0.30	3,89,10,810 32% given back	—	—	+4.8%	51%	1443	15/15	1198	pooled	shared	0.3503	0.6901	0.2041	1.491	0.3729	0.8102	0.4861	0.2904	0.6955	0.3051	1.092	38	both	predictor_else_form	winner_take_most	True	True	True
10	+28.772	+1894.4% 99,71,954	39,45,009	1.36	+1646%	—	22%	—	−0.30	1,24,99,400 20% given back	—	—	+5.8%	51%	1447	15/15	1201	pooled	shared	0.3363	0.7384	0.7977	2.371	0.9344	0.3846	0.9087	0.2433	0.6703	0.6839	1.391	34	confidence_only	form_only	winner_take_most	True	False	True
1	+12.784	+298.1% 19,90,669	8,61,543	1.50	+155%	—	12%	—	−0.35	21,47,816 7% given back	—	—	+7.0%	47%	1301	15/15	1058	pooled	shared	0.2993	0.6041	0.3258	1.101	0.672	0.7356	0.673	0.252	0.5706	0.7772	1.215	13	confidence_only	form_only	equal	True	False	False

The same scoring, but with NO penalty for trading rarely — the "barely traded" and "only traded in a slice of the years tested" docks are switched off. A model that found one genuinely good, rare setup can rank well here even though the ranking above would mark it down for thinness. Getting lucky on a couple of trades is still docked, and a real wipe-out still sinks to the bottom — those aren't about trade count.

model iThe model's number is the order it was tested in the sweep -- it never changes. The table is SORTED best-first, so row 1 is not model 1.	score	account end	set aside	profit factor	at 1.5×	fragility dock	top-3 / profit	fluke dock	thin dock	peak	gave back	raw	ret/dep	win%	trades	windows	overnight	entry	bank	conf_size_low_pct	conf_size_high_pct	conf_size_min_mult	conf_size_max_mult	active_capital_selectivity_strength	active_capital_max_share	active_capital_winner_top_share	throttle_low_winrate	throttle_high_winrate	throttle_min_mult	throttle_max_mult	throttle_window_trades	variant	confidence_source	active_capital_allocation_mode	confidence_sizing	active_capital_sizing	win_rate_throttle_enabled
6	+146.282	+14622.7% 7,36,13,702	4,37,65,630	1.07	+2062%	—	68%	—	−0.30	—	—	—	+2.9%	51%	1446	15/15	1200	pooled	shared	0.3053	0.7499	0.2766	2.328	0.9829	0.9232	0.7275	0.269	0.5385	0.635	1.285	11	both	predictor_else_form	winner_take_most	True	True	False
3	+102.817	+10277.2% 5,18,85,867	3,28,09,971	1.06	+871%	—	68%	—	−0.30	—	—	—	+2.6%	51%	1438	15/15	1192	pooled	shared	0.2531	0.629	0.4513	1.624	0.9663	0.6575	0.7163	0.2498	0.7083	0.6108	1.084	28	both	predictor_else_form	winner_take_most	True	True	False
5	+97.801	+9775.1% 4,93,75,572	3,01,91,109	1.07	+455%	—	64%	—	−0.30	—	—	—	+2.7%	51%	1444	15/15	1198	pooled	shared	0.4281	0.7186	0.761	2.327	0.6531	0.7325	0.8773	0.3412	0.6788	0.758	1.447	32	both	form_only	proportional	True	True	True
4	+90.088	+9004.0% 4,55,20,119	2,87,67,549	1.06	+696%	—	69%	—	−0.30	—	—	—	+2.6%	51%	1433	15/15	1187	pooled	shared	0.3825	0.6442	0.7926	2.049	0.3069	0.654	0.9287	0.1844	0.6301	0.4477	1.148	16	both	predictor_else_form	winner_take_most	True	True	False
0	+83.245	+8320.1% 4,21,00,605	2,37,06,467	1.09	+496%	—	60%	—	−0.30	—	—	—	+2.8%	51%	1440	15/15	1194	pooled	shared	0.3048	0.662	0.304	2.274	0.1719	0.6524	0.6381	0.1665	0.7651	0.6958	1.397	17	both	form_only	proportional	True	True	True
11	+71.244	+7119.6% 3,60,98,157	2,24,28,534	1.07	+155%	—	66%	—	−0.30	—	—	—	+2.5%	52%	1450	15/15	1204	pooled	shared	0.3197	0.6842	0.6652	2.428	0.9172	0.2067	0.8771	0.3897	0.5397	0.6584	1.066	33	both	predictor_else_form	equal	True	True	True
7	+57.058	+5701.9% 2,90,09,746	1,82,58,305	1.06	+171%	—	69%	—	−0.30	—	—	—	+2.4%	51%	1440	15/15	1194	pooled	shared	0.2236	0.8121	0.6617	1.773	0.7934	0.9002	0.8768	0.3595	0.6289	0.7059	1.308	23	both	predictor_else_form	equal	True	True	False
2	+53.470	+5345.6% 2,72,27,919	1,69,32,542	1.06	+105%	—	68%	—	−0.30	—	—	—	+2.4%	51%	1429	15/15	1183	pooled	shared	0.2324	0.5776	0.5325	1.488	0.369	0.5582	0.8134	0.1629	0.7813	0.6288	1.391	28	active_only	predictor_else_form	equal	False	True	False
9	+52.372	+5229.1% 2,66,45,598	1,19,90,662	1.19	+4530%	—	52%	—	−0.30	—	—	—	+4.8%	51%	1443	15/15	1198	pooled	shared	0.3503	0.6901	0.2041	1.491	0.3729	0.8102	0.4861	0.2904	0.6955	0.3051	1.092	38	both	predictor_else_form	winner_take_most	True	True	True
10	+19.072	+1894.4% 99,71,954	39,45,009	1.36	+1646%	—	22%	—	−0.30	—	—	—	+5.8%	51%	1447	15/15	1201	pooled	shared	0.3363	0.7384	0.7977	2.371	0.9344	0.3846	0.9087	0.2433	0.6703	0.6839	1.391	34	confidence_only	form_only	winner_take_most	True	False	True
1	+3.138	+298.1% 19,90,669	8,61,543	1.50	+155%	—	12%	—	−0.35	—	—	—	+7.0%	47%	1301	15/15	1058	pooled	shared	0.2993	0.6041	0.3258	1.101	0.672	0.7356	0.673	0.252	0.5706	0.7772	1.215	13	confidence_only	form_only	equal	True	False	False

Nothing but where the account ended up, compounded — no floor for a wipe-out, no credit for steadiness, no docking for luck or thin trading. If a model made the most money, even by one huge trade or a near-wipe-out elsewhere, it's at the top here.

model iThe model's number is the order it was tested in the sweep -- it never changes. The table is SORTED best-first, so row 1 is not model 1.	score	account end	set aside	profit factor	at 1.5×	fragility dock	top-3 / profit	fluke dock	thin dock	peak	gave back	raw	ret/dep	win%	trades	windows	overnight	entry	bank	conf_size_low_pct	conf_size_high_pct	conf_size_min_mult	conf_size_max_mult	active_capital_selectivity_strength	active_capital_max_share	active_capital_winner_top_share	throttle_low_winrate	throttle_high_winrate	throttle_min_mult	throttle_max_mult	throttle_window_trades	variant	confidence_source	active_capital_allocation_mode	confidence_sizing	active_capital_sizing	win_rate_throttle_enabled
6	+58.696	+14622.7% 7,36,13,702	4,37,65,630	1.07	+2062%	—	68%	—	−0.30	—	—	—	+2.9%	51%	1446	15/15	1200	pooled	shared	0.3053	0.7499	0.2766	2.328	0.9829	0.9232	0.7275	0.269	0.5385	0.635	1.285	11	both	predictor_else_form	winner_take_most	True	True	False
5	+37.369	+9775.1% 4,93,75,572	3,01,91,109	1.07	+455%	—	64%	—	−0.30	—	—	—	+2.7%	51%	1444	15/15	1198	pooled	shared	0.4281	0.7186	0.761	2.327	0.6531	0.7325	0.8773	0.3412	0.6788	0.758	1.447	32	both	form_only	proportional	True	True	True
3	+37.152	+10277.2% 5,18,85,867	3,28,09,971	1.06	+871%	—	68%	—	−0.30	—	—	—	+2.6%	51%	1438	15/15	1192	pooled	shared	0.2531	0.629	0.4513	1.624	0.9663	0.6575	0.7163	0.2498	0.7083	0.6108	1.084	28	both	predictor_else_form	winner_take_most	True	True	False
0	+35.788	+8320.1% 4,21,00,605	2,37,06,467	1.09	+496%	—	60%	—	−0.30	—	—	—	+2.8%	51%	1440	15/15	1194	pooled	shared	0.3048	0.662	0.304	2.274	0.1719	0.6524	0.6381	0.1665	0.7651	0.6958	1.397	17	both	form_only	proportional	True	True	True
4	+32.505	+9004.0% 4,55,20,119	2,87,67,549	1.06	+696%	—	69%	—	−0.30	—	—	—	+2.6%	51%	1433	15/15	1187	pooled	shared	0.3825	0.6442	0.7926	2.049	0.3069	0.654	0.9287	0.1844	0.6301	0.4477	1.148	16	both	predictor_else_form	winner_take_most	True	True	False
9	+28.310	+5229.1% 2,66,45,598	1,19,90,662	1.19	+4530%	—	52%	—	−0.30	—	—	—	+4.8%	51%	1443	15/15	1198	pooled	shared	0.3503	0.6901	0.2041	1.491	0.3729	0.8102	0.4861	0.2904	0.6955	0.3051	1.092	38	both	predictor_else_form	winner_take_most	True	True	True
11	+26.339	+7119.6% 3,60,98,157	2,24,28,534	1.07	+155%	—	66%	—	−0.30	—	—	—	+2.5%	52%	1450	15/15	1204	pooled	shared	0.3197	0.6842	0.6652	2.428	0.9172	0.2067	0.8771	0.3897	0.5397	0.6584	1.066	33	both	predictor_else_form	equal	True	True	True
7	+20.503	+5701.9% 2,90,09,746	1,82,58,305	1.06	+171%	—	69%	—	−0.30	—	—	—	+2.4%	51%	1440	15/15	1194	pooled	shared	0.2236	0.8121	0.6617	1.773	0.7934	0.9002	0.8768	0.3595	0.6289	0.7059	1.308	23	both	predictor_else_form	equal	True	True	False
2	+19.591	+5345.6% 2,72,27,919	1,69,32,542	1.06	+105%	—	68%	—	−0.30	—	—	—	+2.4%	51%	1429	15/15	1183	pooled	shared	0.2324	0.5776	0.5325	1.488	0.369	0.5582	0.8134	0.1629	0.7813	0.6288	1.391	28	active_only	predictor_else_form	equal	False	True	False
10	+11.054	+1894.4% 99,71,954	39,45,009	1.36	+1646%	—	22%	—	−0.30	—	—	—	+5.8%	51%	1447	15/15	1201	pooled	shared	0.3363	0.7384	0.7977	2.371	0.9344	0.3846	0.9087	0.2433	0.6703	0.6839	1.391	34	confidence_only	form_only	winner_take_most	True	False	True
1	+1.258	+298.1% 19,90,669	8,61,543	1.50	+155%	—	12%	—	−0.35	—	—	—	+7.0%	47%	1301	15/15	1058	pooled	shared	0.2993	0.6041	0.3258	1.101	0.672	0.7356	0.673	0.252	0.5706	0.7772	1.215	13	confidence_only	form_only	equal	True	False	False

Return divided by the worst real drawdown along the way — doubling every bet roughly doubles both the return AND the drawdown together, so a model that simply sizes bigger stays roughly FLAT here even though it tops the "current ranking" tab. A model that climbs here instead genuinely cut its losses relative to its gains — the win-rate throttle, in particular, working as intended — not just leverage.

model iThe model's number is the order it was tested in the sweep -- it never changes. The table is SORTED best-first, so row 1 is not model 1.	score	account end	set aside	profit factor	at 1.5×	fragility dock	top-3 / profit	fluke dock	thin dock	peak	gave back	raw	ret/dep	win%	trades	windows	overnight	entry	bank	conf_size_low_pct	conf_size_high_pct	conf_size_min_mult	conf_size_max_mult	active_capital_selectivity_strength	active_capital_max_share	active_capital_winner_top_share	throttle_low_winrate	throttle_high_winrate	throttle_min_mult	throttle_max_mult	throttle_window_trades	variant	confidence_source	active_capital_allocation_mode	confidence_sizing	active_capital_sizing	win_rate_throttle_enabled
6	+77.441	+14622.7% 7,36,13,702	4,37,65,630	1.07	+2062%	—	68%	—	−0.30	—	—	—	+2.9%	51%	1446	15/15	1200	pooled	shared	0.3053	0.7499	0.2766	2.328	0.9829	0.9232	0.7275	0.269	0.5385	0.635	1.285	11	both	predictor_else_form	winner_take_most	True	True	False
9	+61.662	+5229.1% 2,66,45,598	1,19,90,662	1.19	+4530%	—	52%	—	−0.30	—	—	—	+4.8%	51%	1443	15/15	1198	pooled	shared	0.3503	0.6901	0.2041	1.491	0.3729	0.8102	0.4861	0.2904	0.6955	0.3051	1.092	38	both	predictor_else_form	winner_take_most	True	True	True
0	+47.143	+8320.1% 4,21,00,605	2,37,06,467	1.09	+496%	—	60%	—	−0.30	—	—	—	+2.8%	51%	1440	15/15	1194	pooled	shared	0.3048	0.662	0.304	2.274	0.1719	0.6524	0.6381	0.1665	0.7651	0.6958	1.397	17	both	form_only	proportional	True	True	True
5	+47.013	+9775.1% 4,93,75,572	3,01,91,109	1.07	+455%	—	64%	—	−0.30	—	—	—	+2.7%	51%	1444	15/15	1198	pooled	shared	0.4281	0.7186	0.761	2.327	0.6531	0.7325	0.8773	0.3412	0.6788	0.758	1.447	32	both	form_only	proportional	True	True	True
3	+44.562	+10277.2% 5,18,85,867	3,28,09,971	1.06	+871%	—	68%	—	−0.30	—	—	—	+2.6%	51%	1438	15/15	1192	pooled	shared	0.2531	0.629	0.4513	1.624	0.9663	0.6575	0.7163	0.2498	0.7083	0.6108	1.084	28	both	predictor_else_form	winner_take_most	True	True	False
4	+38.289	+9004.0% 4,55,20,119	2,87,67,549	1.06	+696%	—	69%	—	−0.30	—	—	—	+2.6%	51%	1433	15/15	1187	pooled	shared	0.3825	0.6442	0.7926	2.049	0.3069	0.654	0.9287	0.1844	0.6301	0.4477	1.148	16	both	predictor_else_form	winner_take_most	True	True	False
11	+31.639	+7119.6% 3,60,98,157	2,24,28,534	1.07	+155%	—	66%	—	−0.30	—	—	—	+2.5%	52%	1450	15/15	1204	pooled	shared	0.3197	0.6842	0.6652	2.428	0.9172	0.2067	0.8771	0.3897	0.5397	0.6584	1.066	33	both	predictor_else_form	equal	True	True	True
10	+25.075	+1894.4% 99,71,954	39,45,009	1.36	+1646%	—	22%	—	−0.30	—	—	—	+5.8%	51%	1447	15/15	1201	pooled	shared	0.3363	0.7384	0.7977	2.371	0.9344	0.3846	0.9087	0.2433	0.6703	0.6839	1.391	34	confidence_only	form_only	winner_take_most	True	False	True
7	+23.089	+5701.9% 2,90,09,746	1,82,58,305	1.06	+171%	—	69%	—	−0.30	—	—	—	+2.4%	51%	1440	15/15	1194	pooled	shared	0.2236	0.8121	0.6617	1.773	0.7934	0.9002	0.8768	0.3595	0.6289	0.7059	1.308	23	both	predictor_else_form	equal	True	True	False
2	+21.456	+5345.6% 2,72,27,919	1,69,32,542	1.06	+105%	—	68%	—	−0.30	—	—	—	+2.4%	51%	1429	15/15	1183	pooled	shared	0.2324	0.5776	0.5325	1.488	0.369	0.5582	0.8134	0.1629	0.7813	0.6288	1.391	28	active_only	predictor_else_form	equal	False	True	False
1	+2.739	+298.1% 19,90,669	8,61,543	1.50	+155%	—	12%	—	−0.35	—	—	—	+7.0%	47%	1301	15/15	1058	pooled	shared	0.2993	0.6041	0.3258	1.101	0.672	0.7356	0.673	0.252	0.5706	0.7772	1.215	13	confidence_only	form_only	equal	True	False	False

Which dials the winners agree on

Every model in the sweep is a different set of these dials, picked at random. This table asks, for each dial: did the best models tend to settle on a different value than the average model? The grey dot is where all the models sit on that dial; the blue dot is where the best models sit (10% of the survivors once the sweep is big enough; the best few while it's still small). When the two dots are far apart, that dial matters — the winners agree on it. When they sit on top of each other, that dial makes little difference. If the winners keep pinning a dial to one end of its range, the sweep opens that end a little and keeps searching — those dials show a small blue ↔ with the new range.

Dial	best 10 avg iThe average value the best 10 models by score landed on for this dial (survivors only).	agreement iHow far apart the two dots are, as a percentage of the dial's full range. Bigger = the winners agree on this dial more strongly.
active_capital_sizing iWhether idle capital is re-allocated toward the models actually in a trade, instead of every model keeping a fixed share.	100% on	17%
throttle_max_mult i	1.25 · 5 live	5%
conf_size_max_mult iThe largest bet multiplier a high-confidence trade can be raised to (2 = double the normal size).	2.037 · 10 live	4%
throttle_low_winrate i	0.2969 · 5 live	4%
confidence_sizing iWhether bets are sized up/down by trade confidence at all (same trades either way, different amounts).	80% on	3%
throttle_min_mult i	0.6043 · 5 live	3%
throttle_window_trades i	30 · 5 live	3%
win_rate_throttle_enabled i	40% on	2%
conf_size_min_mult iThe smallest bet multiplier a low-confidence trade can be cut to (0.5 = half the normal size).	0.5146 · 10 live	2%
conf_size_high_pct iConfidence at or above this level gets the LARGEST bet size.	0.6988 · 10 live	2%
conf_size_low_pct iConfidence below this level gets the SMALLEST bet size. Between the low and high anchors the bet scales up in a straight line.	0.3209 · 10 live	0%
throttle_high_winrate i	0.6698 · 5 live	0%
active_capital_winner_top_share iIn winner-take-most mode: how much of the idle pot the single top-ranked model takes outright before the rest is shared.	0.7147 · 4 live	0%
active_capital_selectivity_strength iHow strongly idle money is tilted toward the pool's pickier, better-performing models rather than split evenly.	0.6162 · 10 live	0%
active_capital_max_share iThe most of the shared pot any ONE model may be handed at once.	0.7073 · 10 live	0%