Semiconductor pivot fee V3 — completed research handoff

2026-09-08 JST. Research only; no orders, account mutations, Strategy ledger changes or schedules. Do not replace the separate private live-account handoff.

User request and governing version

User supplied Taker0.0081% / Maker0.0027%, asked whether the cited profitable cash NVDA15m result meant it won, and requested identical-condition SK hynix/MU/SNDK backtests. Correct cited authority is the LAST55-condition TypeScript roleflip-v2.1 study, CASH_15m_BREAK_NEXT_DWM:77 trades,63.636% win,+0.06331844837892621R after assumed6bps. Do not mix older75-case JS results. Exact engine copied to pivot216_semiconductors_frozen_engine_v3.ts. All entry/exit thresholds and old reward>6bps selection gate preserved; lower fees only change scoring. The6bps gate is a retained selection buffer, not the actual fee. Six modes BREAK_NEXT/BREAK_NEAR/RETEST_LIMIT/REJECT_2/HYBRID/DP_LONG.15m, prior ATR14, SL0.50ATR beyond pivot,next-pivot TP,6h holding,cash session close,one pending/position,stop-first ambiguities. User's full discretion/RSI/volume/regime not replicated.

Persistent authority

Val kamibaba/hyperliquid-market-bridge; val-scoped blobs. New prefix P=research:pivot216:semiconductors:20260908:fee_v3. Old NVDA archive research:pivot216:NVDA:20260908 preserved. P:dataset: / P:raw:; P:normalized::; P:result: full trades/orders/statistics; P:summary:NVDA/MU/SNDK/SKHX; P:runmeta:; P:export:summary.csv; P:all_summaries; P:final_audit; P:primary_diagnostics; P:candidate_diagnostics; P:FX_conversion_SKHX. Scripts: pivot216_semiconductors_support_v3.ts, pivot216_semiconductors_run_v3.ts (saved SYMBOL=SKHX), pivot216_semiconductors_export_v3.ts. Protocol PIVOT216_SEMICONDUCTORS_FEE_V3_PROTOCOL.md includes pre-outcome data-quality amendment. Immutable guards reject rerun into occupied result keys. Use a new version/prefix for extensions.

Completed checks

72 cases=4symbols3price paths6modes.6774 simulated rows across overlapping cases, NOT unique independent observations.26positive fee-only meanR cases.9basic engine tests passed. NVDA77 count and old6bps mean matched exactly. All72trade order/risk/target/nonoverlap checks passed, no censored trades.72fee-formula recomputations matched. Local72-row CSV SHA256 matched cloud:6d7dbe1d1952f32ec18c49b31fe5d3a93664c5a9a3e2af31dee53743ef97027d (19500bytes). Excel formula error scan0; visual preview inspected; fee-rescore CLI tested on72rows. These checks do not prove profitable live execution.

Fee assumptions

Taker decimal.000081,Maker.000027. Baseline exact fee per one-unit trade=.000081*(Entry+Exit), approximately1.62bps roundtrip. Separate passive assumption: limit entry maker,TP maker,other entry and stop/time exits taker. Additional combined spread/slippage2/4/8bps are sensitivities, not measured historical friction. Funding excluded throughout. Cash price paths use hypothetical HL fees for transfer reference, not actual broker equity fees,tax or short-borrow cost.

Main findings

NVDA quoted cash77:meanR+.1374199725,PF1.3805079,win63.636%;early48+.2435433,late29-.0382324. Extra4bps+.0697486. Historical sample profitable, not demonstrated stable future edge. Same cash BREAK_NEXT:SKHX68-.0136616R;MU75-.0694696R;SNDK64-.1138858R. NVDA cash success did not generalize to all3cash names. XYZ_CASH_LEVELS BREAK_NEXT:NVDA152-.0644746;MU141+.1867998(PF1.3561,early105+.146748,late36+.303618,extra4+.119197,extra8+.051594);SNDK145-.0242982;SKHX159+.0573810(FXapproximation,extra4-.0130351). XYZ_UTC_LEVELS BREAK_NEXT:NVDA159-.2077895;MU170-.0879084;SNDK131-.0110524;SKHX171+.1034650(PF1.184,early121+.066835,late50+.192110,extra4+.031604,extra8-.040256). SKHX nativeXYZ HYBRID210+.123841R,PF1.196,early+.130960,late+.108647;extra4+.032660,extra8-.058522. SNDK nativeXYZ DP_LONG27+.452806R,PF1.785,late6only;not same main condition,CI crosses0 and top3 removal gives-.115494. Main MU withoutbest1+.100786,withoutbest3+.002399. SKHX nativebreak without3-.023073;nativeHYBRID without3+.038331. Removing winners is concentration sensitivity, not an unbiased estimator or reason alone to discard tail-return strategies. MU nativeXYZ signals using cash pivots gain mainly outside cash RTH:48RTH-.045746,79outside+.307247,14cashclosed+.304433. SKHX nativebreak:55RTH-.123928,86outside+.162416,30closed+.351359. Subgroup attribution is NOT separately rerun time-filter strategy. Historical execution friction/funding matters especially there. R=equal initial stop risk;bps=equal entry notional. Some signs differ. Do not claim every sizing method loses from negative meanR (e.g.SNDK XYZcash breakout-.024298R but+3.279526bps).

Data and fidelity

Cutoff2026-09-08T00:30Z. CashUS each1560x15m,60sessions June11-Sep4. SKcash1378x15m,53complete sessions June15-Sep7. XYZ NVDA5012,MU/SNDK4994each,SKHX4993;roughlyJuly18-Sep8,zero gaps. NewXYZ grid.10USD conservatively rounded;oldNVDA.01USD,cashUS.01USD,cashKRW1000KRW. SK common stock000660.KS,not ADRSKHY. Regular-only Yahoo wrongly ended15:00KST; extended fetch recovered official09:00-15:30.15:30auction folded into15:15bar; no invented15:20/25prices. Exclude six incomplete CASH execution sessions:June23,June26,July7,July13,July28,July29. Their dailyOHLC remains available for pivots. Missing-day exclusion may bias cash inference; XYZ retains full own candles. SKXYZ cash-source levels are explicitly approximate FX translation: each cash dayOHLC divided by latestFX known before that day open, then prior-period aggregation. Not exact USD oracle/Pine price path. MaxrecentFXlag50.33h includingweekend. Never apply today'sFX to all history. NativeXYZ pivots require no cashFX mapping. CommonwindowUTCJuly20 00:00-Sep5 00:00 is subset of existing trades fully contained, NOT a reset/re-run. MainMU137+.178639R(extra4+.114800);SKHXnativebreak156+.076782(extra4+.009516). Samecalendar,not same local trading hours.70/30 split retrospective;weekly bootstrap1000exploratory,no multiple-comparison adjustment. NVDA already inspected;new symbols not tuned.

Local deliverables actually created

/mnt/data/Semiconductor_Pivot_V3/Results_72.csv /mnt/data/Semiconductor_Pivot_V3/Results_FeeAdjusted.xlsx (5sheets) /mnt/data/Semiconductor_Pivot_V3/Report_JA.md /mnt/data/Semiconductor_Pivot_V3/MANIFEST.json /mnt/data/Semiconductor_Pivot_V3/audit_fees.py /mnt/data/Semiconductor_Pivot_Fee_V3_20260908.zip ZIP contains these5files, NOT raw candles/all6774trades/fullValstrategy code. Those remain inVal. Python utility rescales aggregate taker fee/friction using coefficients; cannot regenerate market trades,PF,win rates or confidence intervals. Excel averageR fee formulas update; overview/PF/win/CI/passive columns remain static baseline and are labeled as such.

Next research constraints

Evidence prioritizes MU exact cash-source/XYZ-breakout and SKHX nativeXYZ setups for further execution validation, not immediate increased live risk. SNDKDP candidate remains exploratory. Preserve all tested variants; do not choose winners on these dates and call them OOS. No guarantees dailyNet target, no inferred user consent to risk caps. Add symbols only with explicit execution source,pivot source,price scale,calendar,DST,tick,fee and data-quality mapping. No fullPine parity,Mark/L2/queue/partialfill/funding/large-size/margin/liquidation simulation completed.