Regime Research

System understanding via Markov regimes + strategy diagnostics

Question: which feature states should allow or block strategy actions?

Model readiness

Markouts

392

States w/ markouts

Top-10 occupancy covered

10/10

Pass rate (rows)

100.0% (101/101 pass gate)

Pass rate (weighted)

100.0%

Strategy rows: 101 · Kinds: MAKER_ASK, MAKER_BID, TAKER_BUY, TAKER_SELL

Markov: 7:00 AM · Strategy: 7:00 AM · Reload: 60s

Pass at minCount=20: 0/101 · minCount=5: 38/101 · minCount=1: 67/101

Markov Decision Transparency

Evidence-only view of what worked (positive markouts), what failed (negative markouts), and how current gate thresholds would filter that evidence.

Evidenced rows

67/101

No-evidence rows

Worked

2 rows · 2 fills

Failed

0 rows · 0 fills

Flat

65 rows · 390 fills

Baseline WAvg

+0.05 bps

Baseline WAvg (Notional)

+0.07 bps

Gate-allowed WAvg

+0.05 bps

Gate-blocked WAvg

—

Estimated Uplift

0.00 bps

Allowed evidenced fills: 392/392 · Blocked evidenced fills: 0/392

Blocked reason mix: —

Configuration

Window

Horizon

300000ms

Bins

Features

spread, depth, obi, vol, micro

Gating Policy

Enabled

Min Count

Min WAvg

0 bps

Fail Open

Yes

Min Fill Rate

Reload

60s (60000 ms)

Markov JSON

2/15/2026, 7:00:30 AM

Strategy JSON

2/15/2026, 7:00:30 AM

How We Use This

1. Generate regimes + strategy report daily (or hourly).

2. Gate maker/taker by regime when wavg bps is positive and sample is sufficient.

3. Disable strategies in regimes where markout is consistently negative.

4. Review top/bottom combos here to decide which wallets/strategies to expand or cut.

Most common states

High occupancy doesn't mean high evidence. Most fills happen in a small subset of states.

Occupancy = feature steps in this state. Coverage = occupancy-labeled (markout/occupancy). EV_1step = one-step Markov EV (shrunk μ). Cov% = transition mass into known states.

State	Label	Occupancy	Markout count	Coverage	Cov%
55	spread:0 depth:2 obi:0 vol:0 micro:1	1754	65	3.7%	100.0%
27	spread:0 depth:1 obi:0 vol:0 micro:0	876	25	2.9%	100.0%
56	spread:0 depth:2 obi:0 vol:0 micro:2	876	46	5.3%	100.0%
198	spread:2 depth:1 obi:1 vol:0 micro:0	876	25	2.9%	100.0%
20	spread:0 depth:0 obi:2 vol:0 micro:2	756	25	3.3%	100.0%
182	spread:2 depth:0 obi:2 vol:0 micro:2	682	13	1.9%	99.7%
37	spread:0 depth:1 obi:1 vol:0 micro:1	535	32	6.0%	100.0%
46	spread:0 depth:1 obi:2 vol:0 micro:1	462	31	6.7%	100.0%
9	spread:0 depth:0 obi:1 vol:0 micro:0	438	9	2.1%	100.0%
11	spread:0 depth:0 obi:1 vol:0 micro:2	438	22	5.0%	100.0%

Most evidenced states

Where we have evidence: states that had fills with attributed markouts. Coverage = occupancy-labeled. Cov% = transition mass into known states.

State	Label	Markout count	Notional	Coverage	Cov%
55	spread:0 depth:2 obi:0 vol:0 micro:1	65	6491.9	3.7%	100.0%
56	spread:0 depth:2 obi:0 vol:0 micro:2	46	4412.1	5.3%	100.0%
37	spread:0 depth:1 obi:1 vol:0 micro:1	32	3161.0	6.0%	100.0%
46	spread:0 depth:1 obi:2 vol:0 micro:1	31	253.9	6.7%	100.0%
27	spread:0 depth:1 obi:0 vol:0 micro:0	25	2490.5	2.9%	100.0%
20	spread:0 depth:0 obi:2 vol:0 micro:2	25	2197.3	3.3%	100.0%
198	spread:2 depth:1 obi:1 vol:0 micro:0	25	1511.2	2.9%	100.0%
54	spread:0 depth:2 obi:0 vol:0 micro:0	23	2207.5	5.3%	100.0%
11	spread:0 depth:0 obi:1 vol:0 micro:2	22	2176.0	5.0%	100.0%
64	spread:0 depth:2 obi:1 vol:0 micro:1	19	1804.0	4.3%	100.0%
207	spread:2 depth:1 obi:2 vol:0 micro:0	16	324.3	3.7%	100.0%
181	spread:2 depth:0 obi:2 vol:0 micro:1	16	246.0	3.8%	100.0%
182	spread:2 depth:0 obi:2 vol:0 micro:2	13	229.3	1.9%	99.7%
47	spread:0 depth:1 obi:2 vol:0 micro:2	13	50.3	10.7%	100.0%
9	spread:0 depth:0 obi:1 vol:0 micro:0	9	899.9	2.1%	100.0%

Top Transitions

From	To	Count	Prob
55	55	1750	100.0%
27	27	874	100.0%
56	56	874	100.0%
198	198	874	100.0%
20	20	754	100.0%
182	182	678	99.7%
37	37	533	99.8%
46	46	460	99.8%
9	9	437	100.0%
11	11	437	100.0%
54	54	437	100.0%
64	64	437	100.0%

Top non-self transitions

From	To	Count	Prob
186	188	64	43.5%
188	186	64	17.3%
170	168	16	47.1%
168	170	15	17.6%
179	177	12	42.9%
177	179	8	5.1%
177	188	6	3.8%
168	177	5	5.9%
177	168	5	3.2%
187	186	4	57.1%
186	179	3	2.0%
186	187	3	2.0%

Best Evidenced Combos

Top 15 by WAvg among rows with attributed markouts (count > 0). This is what actually worked.

State	Kind	Wallet	Count	WAvg	Fill Rate
177	TAKER_SELL	3	1	10.35 bps	100.0%
177	TAKER_SELL	5	1	10.35 bps	100.0%
47	TAKER_BUY	2	7	0.00 bps	100.0%
47	TAKER_BUY	4	3	0.00 bps	100.0%
46	TAKER_BUY	2	10	0.00 bps	100.0%
46	TAKER_BUY	3	9	0.00 bps	100.0%
46	TAKER_BUY	4	3	0.00 bps	100.0%
198	TAKER_BUY	3	9	0.00 bps	100.0%
198	TAKER_BUY	4	3	0.00 bps	100.0%
171	TAKER_BUY	3	5	0.00 bps	100.0%
9	TAKER_BUY	3	5	0.00 bps	100.0%
55	TAKER_SELL	2	12	0.00 bps	100.0%
55	TAKER_SELL	3	15	0.00 bps	100.0%
55	TAKER_SELL	5	9	0.00 bps	100.0%
182	TAKER_SELL	3	9	0.00 bps	100.0%

Worst Evidenced Combos

Bottom 15 by WAvg among rows with attributed markouts (count > 0). This is what did not work.

State	Kind	Wallet	Count	Fill Rate
10	TAKER_BUY	2	1	100.0%
9	TAKER_SELL	3	4	100.0%
171	TAKER_SELL	3	4	100.0%
46	TAKER_SELL	3	4	100.0%
46	TAKER_SELL	2	5	100.0%
47	TAKER_SELL	2	3	100.0%
55	TAKER_BUY	6	4	100.0%
64	TAKER_BUY	6	4	100.0%
37	TAKER_BUY	3	4	100.0%
27	TAKER_BUY	3	4	100.0%
11	TAKER_BUY	3	4	100.0%
207	TAKER_BUY	3	4	100.0%
20	TAKER_BUY	3	4	100.0%
182	TAKER_BUY	3	4	100.0%
55	TAKER_BUY	3	12	100.0%

No-Markout Combos (Unknown Outcome)

Orders seen for these (state, kind, wallet) combos, but no attributed markouts in this window yet. Treat as unknown, not win/loss.

Unknown rows: 34 / 101

State	Kind	Wallet	WAvg	Fill Rate
170	MAKER_ASK	14	—	100.0%
168	MAKER_ASK	14	—	100.0%
178	MAKER_BID	14	—	100.0%
188	TAKER_BUY	13	—	100.0%
186	TAKER_BUY	13	—	100.0%
168	TAKER_BUY	13	—	100.0%
177	TAKER_BUY	13	—	100.0%
177	TAKER_BUY	3	—	100.0%
179	TAKER_BUY	13	—	100.0%
188	TAKER_BUY	3	—	100.0%
186	TAKER_SELL	13	—	100.0%
188	TAKER_SELL	13	—	100.0%
168	TAKER_SELL	13	—	100.0%
170	TAKER_SELL	13	—	100.0%
177	TAKER_SELL	13	—	100.0%