# MIROSCOPE ROUND-4 TEMPORAL VALIDATION — REPORT
Date: 2026-09-25. Protocol: `variants/ROUND4-SPEC.md` (locked BEFORE any run;
frozen variant definitions, no parameter changes, EURUSD reference-only,
entry-block trade attribution, V1-RSI45 benched).
Data: 7 validation FX pairs (GBPUSD, USDJPY, AUDUSD, USDCAD, USDCHF, NZDUSD,
EURGBP), Yahoo H1, 2025-01-01 → 2026-09-20 UTC; gold GC=F H1, 2025-01-02 →
2026-09-20 UTC (9,837 bars, validated: monotonic, 24h coverage, 0 malformed).
Engine: `variants/run_variants.py` (stdlib, deterministic; round-4 extension =
full-series runs + entry-block attribution; round-2 EURUSD numbers reproduced
exactly after the change). Analysis: `variants/analyze_r4.py` (verdict logic
asserts EURUSD never enters verdict inputs; proven by counterfactual test).
Accounting: $10k, 1% risk/setup, R-multiple, no spread/commission.
All 27 runs completed: 3 variants x (7 FX + EURUSD ref + GOLD).
Raw trade-level records: `variants/results-r4.jsonl` (3,371 trades).
Verdicts machine file: `variants/verdicts-r4.json`.

## VERDICTS (pre-registered criteria)

| Variant | FX verdict | 7-pair agg E(R) | 7-pair PF | Blocks pass | Pos pairs | Gold verdict |
|---|---|---:|---:|---:|---:|---|
| V3-L5 | **TEMPORALLY ROBUST** | +0.402 | 2.21 | 4/4 | 7/7 | GOLD-REJECTED |
| V3-L3 | **TEMPORALLY ROBUST** | +0.376 | 2.09 | 4/4 | 7/7 | GOLD-INCONCLUSIVE |
| V2-EMA20 | **TEMPORALLY ROBUST** | +0.285 | 1.79 | 4/4 | 7/7 | GOLD-REJECTED |

TEMPORALLY ROBUST required ALL of: 4/4 blocks qualified (>=6/7 pairs with N>=10);
>=3/4 blocks pass (E>=+0.05, PF>=1.15, no qualifying pair-block <= -0.50R);
every block E > -0.25R; no qualifying pair-block E <= -0.50R; >=6/7 pairs positive
aggregate; 7-pair aggregate E >= +0.10R; 7-pair aggregate PF >= 1.30.

## V3-L5 — TEMPORALLY ROBUST
Pair-block cells (n / E(R) / PF), blocks A/B/C/D:

| Pair | A | B | C | D | Pair agg E |
|---|---|---|---|---|---:|
| GBPUSD | 29/+0.391/2.00 | 31/+0.429/2.24 | 34/+0.602/3.64 | 16/+0.311/1.81 | +0.455 |
| USDJPY | 41/+0.224/1.53 | 42/+0.421/2.47 | 42/+0.296/1.78 | 15/-0.122/0.78 | +0.268 |
| AUDUSD | 34/+0.520/3.56 | 32/+0.524/2.78 | 35/+0.548/2.94 | 18/+0.817/5.77 | +0.574 |
| USDCAD | 27/+0.222/1.58 | 25/+0.243/1.55 | 42/+0.364/2.09 | 11/+0.728/5.00 | +0.337 |
| USDCHF | 34/+0.537/2.74 | 32/+0.523/2.67 | 34/+0.200/1.47 | 11/+0.162/1.37 | +0.392 |
| NZDUSD | 36/+0.267/1.60 | 40/+0.660/3.74 | 50/+0.400/2.24 | 17/+0.087/1.20 | +0.402 |
| EURGBP | 29/+0.435/2.65 | 25/+0.379/2.17 | 32/+0.231/1.67 | 16/+0.719/3.53 | +0.402 |

Pooled blocks: A n=230 E+0.368 PF2.06 pass; B n=227 E+0.469 PF2.50 pass;
C n=269 E+0.378 PF2.14 pass; D n=104 E+0.391 PF2.13 pass.
7-pair aggregate: n=830, E=+0.402R, PF=2.21, 7/7 pairs positive.
Worst qualifying pair-block: USDJPY/D -0.122 (above the -0.50 catastrophe guard).
EURUSD REFERENCE ONLY: A 33/+0.322/2.00, B 34/+0.562/3.03, C 42/+0.585/3.20,
D 11/+0.534/2.96. Displayed; never entered verdicts.
Gold: A 29/+0.059/1.13, B 27/+0.205/1.42, C 46/+0.408/2.11, D 21/+0.689/4.71.
Total n=123, E=+0.329, PF=1.86. Verdict GOLD-REJECTED: only 1/4 blocks pass the
pre-registered N>=30 sample bar (A/B/D have 21-29 trades). Note honestly: gold
expectancy was positive in all 4 blocks; the rejection is sample-size, not sign.

## V3-L3 — TEMPORALLY ROBUST
| Pair | A | B | C | D | Pair agg E |
|---|---|---|---|---|---:|
| GBPUSD | 33/+0.521/2.52 | 40/+0.276/1.67 | 40/+0.572/3.35 | 20/+0.398/2.29 | +0.444 |
| USDJPY | 49/+0.324/1.82 | 47/+0.464/2.67 | 52/+0.369/2.01 | 17/+0.048/1.10 | +0.350 |
| AUDUSD | 38/+0.419/2.79 | 40/+0.388/2.15 | 41/+0.504/2.74 | 19/+0.780/5.81 | +0.485 |
| USDCAD | 36/+0.085/1.18 | 35/+0.228/1.53 | 49/+0.339/1.98 | 15/+0.766/4.83 | +0.290 |
| USDCHF | 39/+0.541/2.69 | 37/+0.386/2.10 | 47/+0.170/1.39 | 16/+0.157/1.37 | +0.330 |
| NZDUSD | 43/+0.305/1.70 | 46/+0.474/2.53 | 54/+0.468/2.57 | 19/+0.111/1.25 | +0.384 |
| EURGBP | 37/+0.517/2.98 | 33/+0.363/2.06 | 40/+0.071/1.17 | 18/+0.624/3.32 | +0.353 |

Pooled blocks: A n=275 E+0.383 PF2.09 pass; B n=278 E+0.376 PF2.08 pass;
C n=323 E+0.357 PF2.04 pass; D n=124 E+0.411 PF2.26 pass.
7-pair aggregate: n=1000, E=+0.376R, PF=2.09, 7/7 pairs positive.
Every one of the 28 pair-block cells is positive — the most uniform result of the program.
EURUSD REFERENCE ONLY: A 41/+0.313/1.89, B 39/+0.631/3.36, C 50/+0.439/2.41,
D 12/+0.554/3.22.
Gold: A 33/+0.089/1.19 pass, B 37/+0.548/2.55 pass, C 51/+0.390/2.05 pass,
D 23/+0.756/5.46 fail (N<30). Total n=144, E=+0.420, PF=2.18.
Verdict GOLD-INCONCLUSIVE: 3/4 blocks pass; block D one sample-bar short.

## V2-EMA20 — TEMPORALLY ROBUST
| Pair | A | B | C | D | Pair agg E |
|---|---|---|---|---|---:|
| GBPUSD | 26/+0.424/2.07 | 35/+0.340/2.32 | 30/+0.478/2.98 | 17/+0.243/1.68 | +0.384 |
| USDJPY | 22/+0.133/1.30 | 35/+0.526/2.71 | 39/+0.368/1.96 | 12/-0.273/0.57 | +0.300 |
| AUDUSD | 28/+0.325/2.01 | 39/+0.220/1.59 | 29/+0.242/1.75 | 15/-0.026/0.95 | +0.219 |
| USDCAD | 38/+0.053/1.12 | 45/+0.274/1.65 | 43/+0.279/1.82 | 18/+0.763/6.78 | +0.278 |
| USDCHF | 29/+0.266/1.70 | 32/+0.523/3.02 | 48/+0.167/1.35 | 14/-0.083/0.84 | +0.254 |
| NZDUSD | 24/+0.056/1.14 | 33/+0.142/1.36 | 33/+0.406/2.16 | 9/+0.392/2.55 | +0.232 |
| EURGBP | 30/+0.498/3.13 | 34/+0.266/1.81 | 31/+0.245/1.75 | 17/+0.299/1.73 | +0.327 |

Pooled blocks: A n=197 E+0.249 PF1.67 pass; B n=253 E+0.323 PF1.95 pass;
C n=253 E+0.303 PF1.84 pass; D n=93 (6/7 pairs; NZDUSD/D n=9) E+0.195 PF1.48 pass.
7-pair aggregate: n=805, E=+0.285R, PF=1.79, 7/7 pairs positive.
Weakest cells are all in the short block D (USDJPY/D -0.273, AUDUSD/D -0.026,
USDCHF/D -0.083); none breaches the -0.50 catastrophe guard, and pooled D passes.
EURUSD REFERENCE ONLY: A 25/+0.566/3.69, B 33/+0.387/2.26, C 38/+0.307/1.95,
D 11/+0.174/1.39.
Gold: A 30/+0.133/1.32 pass, B 23/+0.247/1.57 fail (N<30),
C 31/+0.399/2.44 pass, D 16/+0.581/4.10 fail (N<30). Total n=100, E=+0.313,
PF=1.92. Verdict GOLD-REJECTED: 2/4 blocks pass the N>=30 bar. Expectancy
positive in all 4 gold blocks; rejection is sample-size, not sign.

## PLAIN-LANGUAGE BOTTOM LINE
All three variants are temporally robust on FX: positive in every one of the four
calendar blocks, positive on all seven validation pairs, ~800-1000 trades each,
with the frozen definitions untouched since round 2. V3-L3's 28-of-28 positive
pair-block cells are the cleanest temporal result in the program. V3-L5 and V3-L3
remain nearly identical (branch 4: a divergence-threshold plateau, not a magic
number). Gold does not confirm: expectancy was positive in 11 of 12 gold blocks,
but the pre-registered 30-trade block bar rejects V3-L5 and V2-EMA20 and leaves
V3-L3 inconclusive — the verdicts stand as written; gold is not used to rescue
or condemn the FX result.

## ROUND-5 BRANCHES FIRING (not started)
- Branch 1: V3-L5 + V2-EMA20 robust → forward-data validation on genuinely new
  dates, parameters permanently frozen.
- Branch 4: V3-L5 + V3-L3 both robust and similar → threshold-plateau comparison
  (3pt vs 5pt) on new data; no 4pt invention.
- Branch 8: V2-EMA20 FX robust, gold rejects → V2 stays an FX-specific hypothesis;
  forward-test FX only, do not refit for gold.

## CAVEATS (unchanged, still binding)
No spread/commission/slippage modeled. 7 pairs are not 7 independent experiments.
Same calendar window on all pairs (macro-regime risk). Rounds 1-2 selection bias:
round 4 tests persistence of selected hypotheses, not unbiased discovery. Nothing
here authorizes paper trading; that remains Michael's explicit decision.