NVDA Pivot role-flip V2 — research handoff

Date:2026-09-08 JST. Research only; no live account/order/Strategy-ledger changes. Do not replace the separate private trading handoff.

User correction

User does NOT always fade pivots. Strong break upward -> LONG, crossed pivot becomes prospective support, next pivot TP, SL beyond crossed pivot. Reversal only after repeated distinct failed breaks. Screenshot is NASDAQ NVDA15m (not5m), DP broken upward. V1 always-fade and +/-1ATR exits were a different strategy and cannot evaluate the user's method. Numerical thresholds below are research assumptions, NOT user-supplied rules or a claim of exact discretionary replication.

Completed authoritative research

Val kamibaba/hyperliquid-market-bridge, val-scoped blobs. Input prefix research:pivot216:NVDA:20260908 (preserved; do not overwrite). New output prefix research:pivot216:NVDA:20260908:roleflip_v2.

:summary:5m 16 scenarios
:summary:15m 29 scenarios
:summary:1h 10 scenarios
:result: full configuration, trades, pending orders, subgroup/cost/time-split statistics
:runmeta: successful run metadata and9 basic self-tests Total55 completed strategy scenarios. All24 full-period XYZ base modes negative after ASSUMED6bps total round-trip friction. Funding EXCLUDED. Result is not actual all-in exchange Net or proof pivots/user discretion cannot work. Files: PIVOT216_ROLE_FLIP_PROTOCOL_V2.md; pivot216_roleflip_engine_v2.ts (internalversion roleflip-v2.1); pivot216_roleflip_run_v2.ts (saved JOB=1h;5m and15m completed earlier). Input candles:XYZ5m5038,15m5012,1h5002; cash5m4680clean (4681before cleaning),cash daily502,XYZ daily300.15mXYZ window UTC2026-07-17T19:30 to2026-09-08T00:15. Archived last15m bar closes09:30JST; user screenshot10:17JST is NOT a completed sample winner.

Operationalization

A=previous closed ATR14 Wilder,50bar warm-up. Previous completed D/W/M source OHLC; UTC source-period mapping. Source cashRTH vs XYZ UTC separately; execution alwaysXYZ except named CASH reference controls. Entry anchors visible defaultD(P,R1/R2,S1/S2),W(P,R1,S1),M nearestpair before signal; target fullD/W/Mgrid, no prior D-1/W-1. R/S labels do not fix direction. Strong LONG:prevclose<=L,close>=L+max(.25A,2ticks),body>=.50A,close top25% range; SHORT mirrored. Prior anchor ID must exist, no artificial crosses from new-period level jumps. BREAK_NEXT:next-open with no1ATR proximity cap, valid side and before frozen next target. BREAK_NEAR:also within1A. RETEST_LIMIT:L+0.10A long (short mirror), starts NEXT bar,1hour expiry;1tick penetration; no favorable TP credit on intrabar fill bar; adverse SL may occur. REJECT_2:two attempts within6h,>=2bars apart with retreat>=.50A, touch.20A, closeback.10A, clearbreak.25A resets; no fade during simultaneous strongbreak. HYBRID:break-near or retest depending on extension; otherwise repeated rejection. One active position or pending order, no exit-bar reentry. SL beyondanchor.50A; rejection also beyond attempted extremes+.10A. TP nearest current full-grid pivot beyond overlapzone max(.20A,2ticks); never skip farther for betterRR. Freeze geometry. Maxhold6hours (cash end-of-session exit). Stops rounded outward,targets inward,tick.01. Same-barTP/SL stop-first,adversegap atopen,favorablegap cappedtarget. Require targetreward>6bps, no minimumRR base. Costs0/6/12/20bps on same fixed trades.1R=actual initial entry-to-SL distance, NOT v1fixed1ATR.70/30time split is retrospective, NOT untouchedOOS;1000weekly bootstrap exploratory.

Main15m results: CASH_RTH levels, XYZ execution

id suffix | n | win% after6bps | meanR after6bps | PF_R | earlyR | lateR BREAK_NEXT_DP (LONG only) |39|25.6410|-.4490525653|.544353|-.409621|-.563404 BREAK_NEXT_DWM (both sides) |152|44.0789|-.2645260420|.623530|-.287650|-.214424 BREAK_NEAR |114|37.7193|-.3542192764|.558038|-.438715|-.163501 RETEST_LIMIT |86|16.2791|-.9731215756|.279652|-.953511|-1.023783 REJECT_2 |88|35.2273|-.2449278047|.707088|-.140107|-.509077 HYBRID |171|34.5029|-.3434497964|.604143|-.372945|-.274050 DP39=8TP,27SL,4time,10net-positive;plannedRR median4.77778.Individual39prices were copied to a localCSV and arithmetic independently recomputed meanR-.4490525653006544, matching Val. This is transfer/arithmetic validation, NOT independent market-path validation. RETEST170accepted/86filled/84unfilled. HYBRID limit50/15filled/35unfilled. HYBRID0cost grossR+.006998;6bps-.343450;12bps-.693898;20bps-1.161161. SLbuffer.25/.50/.75 gives185/171/165trades andmeanR-.524758/-.343450/-.297677.

Positive cases must not be hidden or promoted as proven

CASH15m BREAK_NEXT_DWM reference77trades,win63.6364%,meanR+.0633184484,PF1.16116;early48+.172788,late29-.117872;CIapprox[-.152739,.290141];12bpsmeanR-.038189. Different dates/session/feed fromXYZ; not causal venuecomparison. CASH15m BREAK_NEAR68+.054254 and RETEST41+.039208 also laternegative. Otherpositive overallconditions:XYZ5m CASH_RTH HYBRID_DP_LONG20+.314155 (late4only,CI[-.566866,1.506298]);XYZ15m XYZ_UTC HYBRID_COMMON60+.099834 (late-.624326);XYZ1h CASH_RTH HYBRID_COMMON20+.009316(late-.317304).Total6of55positive;none establishes stable out-of-sample edge.

Screenshot and examples

Recomputed cashderivedDP231.5833333333,DR1233.5366617839,MR1235.2166646322,DR2236.7133229574 roughly match screenshot lines. One-point match is not fullTradingViewparity. Image's exactuserSL/ATR unknown. No live instruction or imagewinner claim. Recorded DP LONG examples same2026-08-25JST:17:30entry210.87,DP210.439997,SL210.29,TP213.42(MR1),SL,-1.218141Raftercost;20:30entry210.70,SL210.25,sameTP213.42,TP,+5.763511R. These are simulated outcomes, not executed trades.

Incomplete checks / limits

pivot216_roleflip_validation_v2.ts was created but RUN BLOCKED by tool safety verification. Therefore planned40shifted-grid controls, lower-timeframe data alignment, cashXYZbasis analysis, and extra rejection-state synthetictests did NOT complete. Do NOT claim :validation, :diagnostics, :export:summary.csv or :all_summaries were created by that file.9basicselftests in successful main runs DID pass. Missing:NASDAQovernight actualfeed,fullPinepoint-in-timeparity,intrabar discretionaryentries,realvolume/regime/RSIfilters,Mark history,funding,L2/queue,partialfills,measuredslippage,capacity,margin/liquidation. .50ATRbuffer is not proven to reflect user's intended tolerance. Do not reverse to always-fade or optimize same dataset until one variant wins.

Local deliverables prepared

NVDA_Pivot_RoleFlip_V2_Results.csv (55rows,~9decimal rounded output transcription); NVDA_Pivot_RoleFlip_V2_DP_Long_Trades.csv (all39 DP LONG price records); NVDA_Pivot_RoleFlip_V2_Report_JA.md; NVDA_Pivot_RoleFlip_V2_Results.xlsx (5sheets incl formulas/costsensitivities); package JSONconfig/checks and originalPine. Full raw candles and ALL55 individualtrade lists remain inVal blobs, not in localZIP. Workbook formula-error scan0 andDP39reaggregation matched. No background collection scheduled.

Continuation / other symbols

Reuse deterministic engine but explicitly configure exact execution symbol,pivot-source symbol/scale,tick,calendar/tradingday/session/DST,costs. NYclassifierNVDA-specific; not plug-and-play forFX/metals/Koreanstocks. Prioritize fidelity to actual user-selected entries andSL tolerance before treating more parameter searches on the same dates as validation. Preserve failed variants and use new untouched future periods for a genuine confirmatory test.