SD1H4H — Value-Area Migration Strategy

Results across all 15 markets this project has tested (50% target fraction)

6 of 14 cost-surviving markets pass this project's 3-fold walk-forward bar -- against roughly 1.75 expected by pure chance at this many comparisons (same math this project used for its session-filter multiple-comparisons check). A real, elevated hit rate, but three of the robust hits (ETH, SOL, LINK) are crypto assets that move together and are not independent confirmations of anything -- the honest read is closer to 4 independent hits (gold, Nasdaq, EUR/USD, "crypto" as one correlated bloc) than 6. Gold's own profile looked the most conservative on the full sample. The 2026-09-16 holdout re-check changes that read: gold value-area 50%/ATR-stop is only 1/3 pre-holdout (net avg R +0.031, n=856) while Nasdaq, EUR/USD, ETH, SOL, and LINK remain 3/3. Gold is therefore the weakest of the six under the holdout-respecting bar, not the most trustworthy. Source: backtest/results/holdout_review.json. Gold's own profile is the most conservative and arguably most trustworthy of the passers, being both the market the idea was observed on and the one with the least cross-market correlation risk.

Gold's target-fraction sweep (net of costs, stop_mode=atr)

50% and 78.6% were 3/3 robust; 61.8% and 100% were 2/3 -- a believable, mixed pattern (not a suspiciously clean sweep), which is itself evidence the corrected version isn't hiding the same bug in a subtler form.

Full fraction sweep, all 14 cost-surviving markets

19 of 56 market×fraction combinations pass 3/3 walk-forward (vs. roughly 7 expected by pure chance) and 9 of 14 markets have at least one passing fraction -- but per the note above, the more informative read is the pattern: EUR/USD, SOLUSDT, and USATECHIDXUSD show a broad region (majority of fractions passing); USA500IDXUSD shows an adjacent pair at the high end. Everything else is either a single isolated point or an alternating pattern (pass/fail/pass/fail) -- treated as noise, the same way this project has rejected similarly unstable parameter sweeps before (e.g. the stop-distance H1 hypothesis).

Period concentration, 4H timeframe, and Phase 2 confluence factors (spec Section 11)

All three follow-ups from the prior "not done yet" list are now closed out -- two negative, one differentiating:

• Per-year concentration check (differentiating): within the six 50%-fraction robust markets, USATECHIDXUSD is positive every single year 2022-2026 (cleanest of the six); SOLUSDT and EURUSD are consistent with one weak/negative year each. ETHUSDT's entire edge is 63% concentrated in 2023 alone, and LINKUSDT shows a declining trend every year since 2023 (34.7 → 21.8 → 14.8 → 4.1 net R) -- both already flagged as correlated crypto, now shown to be the weaker two of the three specifically.
• 4H timeframe check (negative): re-running the identical 50%/ATR-stop config on 4H bars flips 5 of 6 markets flat-to-negative (XAUUSD -0.018, USATECHIDXUSD -0.048, EURUSD -0.025, ETHUSDT -0.009, SOLUSDT -0.049 net avg R); only LINKUSDT stayed positive (+0.105), read as noise given it's already the weakest market above.
• Phase 2 confluence factors (negative): wick/body ratio, volume spike (with its tick-proxy caveat), and ATR contraction were each tested individually on gold at one pre-specified threshold. All three made results WORSE than the unfiltered baseline while also cutting the trade count substantially -- the opposite of what an "earlier, higher-confidence entry" factor should do. Not extended to other markets since there's no positive result on gold to justify it.

15min timeframe check (spec Section 12)

Same config re-run on 15min bars for the six 1H-validated markets. Gross is positive on all six, but net-of-costs fails for 4 of 6 (XAUUSD, ETHUSDT, SOLUSDT, LINKUSDT) -- the same cost_in_R = fee / stop_width mechanic already documented elsewhere in this project: trade frequency roughly triples at 15min, and the ATR-based stop shrinks at finer granularity while the fixed per-trade cost doesn't, so the same fee eats a much larger fraction of R. USATECHIDXUSD is a genuine finding: net avg R +0.037, 3/3 walk-forward robust, 86.2% Monte Carlo probability of profit -- the same market that was already the cleanest of the six on 1H, now shown to hold on a second, independent timeframe. This is a meaningfully stronger form of corroboration than anything else checked so far (every other cross-check -- the fraction sweep, the 4H check, Phase 2 factors -- either weakened or failed to extend a 1H-only result). EUR/USD's 15min result looked thin (net avg R +0.005) and failed walk-forward outright (1/3), confirming it wasn't real at this timeframe. The other 9 markets that didn't validate on 1H haven't been checked on 15min, to avoid reopening the same multiple-comparisons risk the fraction sweep already flagged.