Current manual method · completed v1.3c baseline · v1.5.3b RHP live research

BD Hedge Separation PiX Strategy

A three-layer authority page: the current BD manual hedge-separation method, the completed narrower v1.3c replay evidence, and the in-progress v1.5.3b RHP stage-atomic research run.

Real BTC/ETH/SOL 1m replay, 1 January–14 June 2026, virtual fills only. No live-order path.

Current BD manual method — separate from the arena seeds

This is the present human operating logic. It is not identical to the completed v1.3c engine and it is not a claim that v1.5.3b has already validated every manual action. The arena is being used to test and decompose this method under matched controls.

Base clip parity

1 / .001 / .04

1 SOL, 0.001 BTC or 0.04 ETH per leg.

Initial state

LONG + SHORT

Open both directions on the same instrument.

Working leverage

300×

Cross-margin / wallet-silo context with deliberately small clips.

Primary objective

Harvest excess

Realize rescue and winner profit while preserving useful cores.

Rounded cross-asset base quantities

Instrument	Base quantity per LONG leg	Base quantity per SHORT leg	Role
SOLUSDT	1 SOL	1 SOL	Canonical manual reference clip
BTCUSDT	0.001 BTC	0.001 BTC	Rounded SOL-comparable clip
ETHUSDT	0.04 ETH	0.04 ETH	Rounded SOL-comparable clip

These are intentionally rounded asset quantities, not continuously price-rebalanced equal notionals. The v1.5.3b primary leaderboard remains a separate fixed-notional lane.

Winner sidelet trend earn size

Losing sidewait for adverse threshold

→

Structure checktrend, corridor, stabilization

→

Add / re-anchortimed, not blind

→

Close excessretain useful core

Open the hedge. Start with the same rounded base quantity LONG and SHORT. The hedge supplies time and optionality; it is not the edge by itself.

Separate the books. Manage the profitable and losing sides independently. Do not automatically flatten the winner merely because the other side needs repair.

First adverse add. Around −200% ROE, the usual manual candidate is an add of roughly 2× the current losing-side size, but only after reading trend, volume, extension and available margin.

Escalate by distance, not panic. Later adverse thresholds and add sizes may approximately double (−400%, −800% … and 2× size steps). This is a discretionary ladder, not a requirement to add blindly.

Harvest the rescue clip. When the newly added excess earns roughly +2%, close that excess and preserve the intended base or residual position.

Harvest the winner progressively. Around +200% ROE, realize or protect roughly half while locking about +100%; around +400% and +500%, progressively lock additional profit near +300% and +400%. Exact fractions remain situational.

Grow a clean winner selectively. On a rapid, well-confirmed continuation, add roughly 2–3× to the profitable leg. Across successful harvest cycles, the retained core may step from roughly 2 → 4 → 8 SOL rather than resetting to the original base every time.

Re-anchor when separation becomes useful. When the losing side is badly displaced—often near a roughly 500% ROE separation—close and reopen it at a better level rather than carrying a permanently poor entry.

Use a large rescue only after stabilization. A typical SOL example is adding 10–20 SOL, then closing most of it at +1–2% on the rescue clip—for example, close 18 and retain about 2 SOL as core.

What makes the method distinct

Both directions remain available.
The winner may be grown and harvested instead of sacrificed.
The loser is repaired through timing, re-anchoring and excess-close cycles.
Risky variants remain testable and are classified by capital-at-risk rather than deleted.

Hard reality

300× leverage magnifies execution and liquidation errors.
A constant hedge does not remove gross exposure or margin risk.
Zero trading fee does not mean zero funding, spread or slippage.
A one-way trend can outrun every rescue ladder if sizing and wallet limits are wrong.

Delayed-add stabilization gate — experimental risk governor

Instead of adding immediately after an adverse threshold, wait a short causal 1m observation window and classify what happened: snapback, stabilization, deeper extension or adverse acceleration. The result may allow, delay, reduce, deepen or veto the add.

Current evidence boundary: delayed-add remains a risk-governor watch branch, not a proven profit engine. It must beat instant-add, fixed-wait, random-wait, no-add and same-ladder/no-sensor controls before stronger promotion.

Manual method, v1.3c and v1.5.3b are different authorities

Layer	Contains	Evidence status
Current BD manual method	Rounded 1 SOL / 0.001 BTC / 0.04 ETH bases; adverse ladder; selective 2–3× winner adds; staged winner protection; dynamic re-anchor; 10–20 SOL rescue clips.	Current operating method and research seed. Not yet fully encoded or matched-control validated.
Completed v1.3c engine	Equal hedge; losing-side re-anchor; bounded fixed-multiple rescue; excess close; retained residual. No active winner pyramiding or staged winner ladder.	Completed 300,000-task real-1m replay baseline.
v1.5.3b RHP arena	Corrected barscope and sizing; stage-atomic retry/resume; typed AI families; exact legacy lane; controls; walk-forward and locked holdout ladder.	In progress. Discovery leaders are not final winners before locked holdout.

What the v1.3c strategy actually was

Initial state

Equal hedge

Open LONG and SHORT at the same notional.

Loss response

Re-anchor

Close and reopen the deeply losing side when its trigger and trend gate pass.

Recovery engine

Mega rescue

Increase the re-anchored losing side to a bounded multiple of the original base.

Profit realization

Excess close

When the rescued side recovers beyond its weighted entry, close the excess and keep a residual.

Not tested in v1.3c: adding to the profitable side, staged winner exits, winner trailing, or force-closing the winner when mega rescue fires. Older manual text implying those behaviors has been removed.

The tested state machine

LONG baseequal notional

SHORT baseequal notional

→

Trigger + gatelosing side only

→

Re-anchorclose + delayed reopen

→

Mega rescuetarget total multiple

→

Excess closeretain residual hedge

Open both sides. The base size is expressed as fixed notional derived from base_sol × $150, then converted to asset quantity at the first close.

Identify the losing side. Re-anchor is eligible when its ROE crosses the configured negative trigger, the opposite side is positive, and the trend gate passes.

Close and reopen the loser. Reopen after the configured number of completed 1m bars and at the configured base multiple.

Allow mega rescue after re-anchor. If the same side again crosses the trigger, add until the configured total multiple is reached, capped by max_total_side_multiplier.

Harvest the rescue. When that rescued side moves the configured percentage beyond its weighted average in the profitable direction, close everything above the residual keep multiple.

Respect event caps. Each policy limits re-anchor and mega-rescue counts; the ending residual hedge remains open unless the replay stop ends the path.

ROE approximation

price move % ≈ |ROE %| ÷ leverage

Leverage

ROE trigger

How profit taking worked

What happened

The enlarged formerly losing side recovered past its weighted average entry.
The engine closed the excess quantity in one event.
A configured 1×, 2×, 3× or 5× base residual remained.
Realized rescue profit and ending unrealized hedge PnL were both counted in final MTM.

What did not happen

No adding to the already profitable side.
No +200%/+400% winner ladder.
No multi-stage winner scale-out.
No trailing winner stop.
No execution of the stored close_winner_on_mega flag.

Completed evidence

Tasks

300,000

complete

Candidates

10,000

wide deterministic sample

Control-beating rows

4,785

7.98% of candidate rows

Full 6/6

119

1.19% of configurations

Why this matters: the result was selective. Roughly 92% of candidate tests did not beat their best matched control, yet a persistent frontier survived all six asset×cost comparisons.

P0 Survival Max

full 6/6 configurations

P1 Conservative

full 6/6 configurations

P2 Balanced

full 6/6 configurations

P3 Aggressive

full 6/6 configurations

P4 Moonshot

full 6/6 configurations

P5 Tiny-Wallet Blast

full 6/6 configurations

Risk lanes and representative robust seeds

These are not one universal ranking. Each profile has a different utility function. High PnL and high risk remain visible rather than being deleted by a single wallet preference.

Lane	Representative 6/6 seed	Avg MTM	Worst MTM	Avg score lift	Worst score lift
P0 Survival Max	`deep_p0_t500_sep300_m50_b1p1_same3_83dc7823`	$220.69	$166.04	+127.49	+76.68
P1 Conservative	`deep_p1_t2000_sep2000_m40_b1p0125_flat3_ce1751a4`	$202.12	$147.60	+120.72	+84.39
P2 Balanced	`deep_p2_t2000_sep1100_m50_b1p15_none_9cb5d2d0`	$240.14	$201.91	+168.55	+26.12
P3 Aggressive	`deep_p3_t1600_sep700_m50_b1p5525_none_71657874`	$468.31	$399.48	+436.67	+379.24
P4 Moonshot	`deep_p4_t2000_sep1100_m60_b3p6_none_4d8476d0`	$922.13	$772.98	+451.52	+21.06
P5 Tiny-Wallet Blast	`deep_p5_t2000_sep500_m60_b2p76_flat5_68e39a6f`	$853.98	$626.96	+582.30	+465.72

Historical v1.3c P2 reference — not the current manual protocol

deep_p2_t500_sep1500_m50_b1p15_same3_e55b312a

Zero-cost SOL MTM

+$156.68

Realized

+$60.39

Max drawdown

8.43%

Cross-asset status

6/6

This seed used the narrower v1.3c state machine: losing-side re-anchor, one bounded rescue and one excess close. It did not execute the current manual winner-add, staged winner-harvest or large discretionary rescue logic.

Historical v1.3c P5 reference — explicit all-risk lane

deep_p5_t1300_sep2000_m50_b3p24_same3_b478159a

Zero-cost SOL MTM

+$1,673.27

Realized

+$1,386.71

Max drawdown

68.79%

Historical rescue target

≈195 SOL

Reference only: this seed demonstrated why raw PnL cannot be read without capital-at-risk. It is preserved as a high-risk research surface, not presented as the current manual default.

Wallet size is an output, not a strategy gate

v1.3c used a fixed virtual $500 wallet. That baseline is useful for comparison, but it must not decide whether a strategy exists. A trader may intentionally allocate $100, $1,000, $10,000 or much more to different risk lanes.

v1.4 must report the minimum tested wallet for full-path survival, probability of total loss, PnL by horizon and capital efficiency—while retaining even strategies that fail a chosen wallet.

Extreme P4 watchlist

deep_p4_t1000_sep300_m30_b5p4_same5_37ae672d reached +$2,233.19 MTM zero-cost and +$2,175.88 realistic-cost on SOL before the fixed-$500 drawdown rule stopped the replay on 6 February. Positive equity remained. Only wallet-neutral full-path replay can determine the required capital and final outcome.

Known limits of the current evidence

Engine semantics

reanchor_target_separation_roe was named and stored but never checked in the replay.
mega_rescue_close_winner_on_trigger was stored but never executed.
Winner-side pyramiding and trailing were absent.
The fixed-$500 drawdown stop truncated some high-PnL paths.

Validation scope

One historical Jan–Jun 2026 interval.
Virtual fills, not exchange execution.
Cross-margin and liquidation mechanics remain simplified.
6/6 is a profile-score win against controls, not guaranteed raw-PnL dominance.
Zero-cost is a diagnostic scenario, not a universal live-cost assumption.

v1.5.3b RHP live extract — stage-atomic run

Latest live checkpoint: LIVE_EXTRACT_BDTA_V153B_RHP_SLIM_FOR_GPT_20260625_131601.zip. This is still an in-progress research run, not a final winner report.

Current stage

Daily24h

08_DAILY24H_DISCOVERY running

Completed tasks

74,728

0 task-attempt errors in extract

Daily active time

6.75h

28.1% of 24h target observed

Sizing contract

1 / .001 / .04

SOL / BTC / ETH asset-parity lane

What is better now: v1.5.3b fixes the v1.5.3a stage-skip/MemoryError problem in this extract. Smoke, Standard, Strict Controls and Deep6h are complete; Daily24h is running; no retry error is visible.

Stage progress

Stage	Status	PASS tasks	Active time
04_SMOKE_DISCOVERY	complete	5,112	0.08h
05_STANDARD_DISCOVERY	complete	43,092	5.47h
06_STRICT_CONTROLS	complete	4,104	1.80h
07_DEEP6H_DISCOVERY	complete	10,414	6.00h
08_DAILY24H_DISCOVERY	running	12,006	6.75h

Corrected test contract

Smoke	6,000-bar scope
Standard	50,000-bar scope
Strict controls	full 236,161-row scope
Deep/Daily	180,000-bar scope
Primary ranking	fixed-notional 100 USDT per leg
Asset-parity diagnostic	1 SOL, 0.001 BTC, 0.04 ETH

Control caution: the root candidate_vs_matched_controls.csv in this extract reflects the latest completed Deep6h discovery report and uses global v1.4.2a controls. Candidate-specific controls are present in the strict-control stage, but final promotion should wait for the locked holdout/control-aware verdict.

Daily24h partial watch-list

Family	Program	Rows	Median PnL	Worst PnL	Survived
clock_phase_control	`AI15-CLOCK_PHASE_CONTRO-1405551790`	17	$1,189.03	$-112.99	76%
volatility_budgeted_rescue	`AI15-VOLATILITY_BUDGETE-80befdc946`	17	$508.83	$236.26	47%
clock_phase_control	`AI15-CLOCK_PHASE_CONTRO-e1e4f05a0b`	17	$180.63	$-321.37	65%
volatility_budgeted_rescue	`AI15-VOLATILITY_BUDGETE-1d417ab897`	17	$86.47	$-115.61	100%
fractal_coherence_switch	`AI15-FRACTAL_COHERENCE_-5060b177af`	17	$41.49	$-89.02	88%

Deep6h robust-score leaders

Family	Program	Score	Median FC	Worst FC
trend_exhaustion_handoff	`AI15-TREND_EXHAUSTION_H-db9986c439`	169.66	$804.78	$45.54
multi_horizon_pressure	`AI15-MULTI_HORIZON_PRES-330ab0df2a`	152.96	$128.75	$1.61
multi_horizon_pressure	`AI15-MULTI_HORIZON_PRES-9c24860cb1`	132.32	$30.88	$1.50
pi_phase_gate	`AI15-PI_PHASE_GATE-15987285bb`	109.43	$903.28	$28.47
inventory_corridor_rescue	`AI15-INVENTORY_CORRIDOR-f7371b8ba5`	105.69	$308.00	$-51.16

Do not overclaim yet: high force-close PnL remains a hypothesis until Daily24h, Weekly168h, Walkforward and Locked Holdout finish. Treat these rows as seed candidates, not proven live edges.

Next research gate after v1.5.3b

Finish the evidence ladder

Complete Daily24h, Weekly168h, Walkforward and Locked Holdout before final promotion.

Encode manual parity

Translate the −200% ladder, 2–3× winner adds, staged protection, dynamic re-anchor and bounded 10–20 SOL rescue into causal arena programs.

Capital and control truth

Require candidate-specific controls and report minimum surviving wallet, gross notional, drawdown, liquidation buffer and open-tail dependency.

current manual seed → causal ProgramIR translation
                         ↓
matched controls + exact legacy comparison
                         ↓
Daily / Weekly / Walkforward / Locked Holdout
                         ↓
P0–P5 and Pareto frontiers by capital-at-risk