Date: 2026-09-08 JST. User corrected v1: strong breakout -> trade with the break near the pivot, next pivot TP, invalidation beyond anchor pivot; fade only after repeated rejection. User example is a 15m NVDA cash chart with D P broken upward. Resistance becomes prospective SUPPORT after an upward break. V1 always-fade / +/-1ATR exits did not test this method.

Research only. No account state, orders, signatures, schedules or paid data. Preserve v1 inputs/results. Input blob prefix research:pivot216:NVDA:20260908. New output prefix research:pivot216:NVDA:20260908:roleflip_v2.

Previous completed source-period Traditional D/W/M; discard first partial W/M source periods. CASH_RTH independent Yahoo OHLC vs XYZ_UTC OHLC separately; execute XYZ only except clearly labeled cash control. Actual TradingView feed/period mapping not certified. Reuse frozen historical inputs; do not refresh into the original prefix. Entries use default displayed D P/R1/R2/S1/S2, W P/R1/S1, M nearest above/below PRECEDING close. Target search uses full D R1-R5/S1-S5, W R1-R3/S1-S3, M R1-R5/S1-S5 as original next-target code. R/S names do not fix trade direction; ANY pivot can flip roles. Levels within max(0.20 ATR,2 ticks) of anchor are treated as one zone for target search, not independent obstacles. Target = first full-grid level beyond that zone in intended direction, fixed at signal time. It is never moved farther away to improve RR. Price tick=0.01 for these NVDA approximations.

ATR14 Wilder of execution candles; A=previous completed candle ATR; 50-bar warm-up. Strong cross long: preceding close<=L, current close>=L+0.25A, bullish body>=0.50A, close in upper25% of current range; short mirrored. New-period level shifts alone never count as crosses: anchor must exist on preceding bar. BREAK_NEAR: after strong cross, buy/sell next candle open only if still on break side and within1.0A of anchor. No hypothetical fill at already-passed pivot price. RETEST_LIMIT: after strong cross, rest limit L+0.10A (long), L-0.10A (short), beginning NEXT bar; expires in1 hour or when source anchor expires. Require1 tick penetration for intrabar fills, not mere touch. A gap across the stop is filled conservatively at limit then stopped at adverse open, not magically canceled before the gap. No TP credit on an intrabar entry bar because ordering is unknown; SL can occur. If target hit without fill, cancel only after that bar closes. REJECT_2: within6h, >=2 distinct failed attempts at same active pivot from same side, separated by at least2 candles AND retreat close>=0.50A away. Touch range within0.20A, close back>=0.10A on approach side. A decisive close beyond0.25A resets attempts. Enter next open on approach side, <=1A from anchor. One continuous cluster is not several rejections. No fade against a simultaneously qualified strong breakout at another nearby visible pivot. HYBRID: strong break has priority; if signal close<=1A from pivot, BREAK_NEAR, otherwise RETEST_LIMIT; if no strong cross, REJECT_2. One position or pending entry at a time. No same-exit-bar reentry. Never combine independently tested modes by adding their trades. SL: anchor minus0.50A for long / plus0.50A for short. REJECT_2 also places SL beyond observed attempt extremes by0.10A. TP = next pivot, not1ATR. SL rounded outward; TP rounded toward entry. Require target reward greater than assumed6bps roundtrip cost, valid risk>0, no minimum RR in base. Pending orders/trades freeze anchor/TP/SL/A. Max holding6 clock hours for ALL execution timeframes. Cash-only closes before session end; no overnight cash entry. Signals near data end excluded so full entry+holding horizon is observable. Both SL/TP within same execution OHLC => stop-first. Adverse gaps at open; favorable gaps capped at target. Structural levels do NOT move after future daily rollover. Historical Mark/L2/queue/partial fills/size capacity/funding not recreated. No claim these are realized exchange Net profits.

Base friction6bps total entry+exit; sensitivities0/12/20bps on same trades. Report both fixed-risk R and equal-notional bps, profit factors for each, target success/SL/timeouts, trade win rate, maxR drawdown, actual RR, pending fill counts, ambiguous bars, long/short and D/W/M and signal-time session breakdowns. No daily dollar target sizing. Main28 scenarios: XYZ 5m/15m/1h x CASH_RTH/XYZ_UTC x BREAK_NEAR/RETEST_LIMIT/REJECT_2/HYBRID=24, plus CASH5m four modes. Secondary17: RTH-only HYBRID5m/15m both sources(4); D P-only LONG HYBRID5m/15m both sources(4); CASH_RTH XYZ15m HYBRID one-factor SL0.25/0.75, body0.35/0.75, netRR>=1.0(5); common5m-overlap HYBRID15m/1h both sources(4). Total45 reported strategy scenarios. Additionally20 deterministic shifted-grid placebo HYBRIDs for each15m source=40. Changes after results must be logged, not hidden. Split each full dataset70/30 by timestamp, excluding boundary-spanning trades. This is retrospective temporal stability, NOT untouched out-of-sample: v1 already examined these dates. Weekly block bootstrap1000 draws; intervals exploratory, few weekly clusters and multiple testing caveats explicit. Same-date secondary removes date confounding but not bar-resolution/intrabar ambiguity differences. Result-selected attractive subgroups are hypotheses, not validated strategies.

Pure engine takes normalized bars/daily inputs/config; new symbols must explicitly specify source instrument, price scale, tick, execution/session clock, holidays/DST and fees. NY cash classifier is NVDA-specific. Keep strategy research separate from live ledger/handoff.

Corrected screenshot timeframe:15m, not5m. Add BREAK_NEXT: same strong-cross definition but no arbitrary1ATR chase cap; enter the next open if still on the break side and before the already-frozen next pivot. This prevents the extra near-entry assumption from silently excluding the user's large-break candle example. Additional four15m scenarios: each source with DWM BOTH and D_P_ONLY LONG. Total49 strategy scenarios plus40 placebo runs. Main interpretation emphasizes15m CASH_RTH levels, XYZ trade candles, and these explicit differences from the screenshot's NASDAQ24h candles. Bar-close confirmation is a conservative observable proxy, not proof the user waits for a close. Frozen input sample ends09:30JST; the screenshot10:17JST is not a completed trade or a winner in these samples. Full input/target-level checks and synthetic execution tests must pass before interpreting results. Additional pre-outcome source control: because the supplied screenshot explicitly uses15m NASDAQ, add six CASH_RTH15m controls (four existing modes plus BREAK_NEXT DWM BOTH and DP LONG), resampled only from complete RTH5m bars. This makes55 strategy scenarios. These cash-price fills are reference results, never presented as XYZ fills; the NASDAQ overnight feed remains unavailable. No outcomes inspected before these additions.