EuroMillions HotPicks Investigation

A statistical and structural investigation into EuroMillions HotPicks conducted over several months using Replit AI and 20 years of draw data. The draw is presented to the public as random. It is not.

1,960 draws. Five machine states. A 90.2% hit-rate edge. Two confirmed Pick-4 predictions. A measurable machine change in May 2026. All documented below.

Public Interest Notice. This investigation is published as a public record. The EuroMillions draw is licensed by the UK Government and presented as a random event. The evidence assembled here — produced with AI-assisted statistical analysis — demonstrates structural non-randomness, measurable bias in number frequency, and a machine behaviour change that followed confirmation of the model's predictive accuracy.

Download the Full Investigation Pack
Six structured files covering all formulae, number sets, draw history, bias data, sector strategies, the Diamond Model, and the 3D ball-tracking investigation. Includes images. Formatted for AI analysis.

⬇ EUROMILLIONS.zip — Full Pack

README.txt · 01 Draw History & Bias · 02 Zone Theory & States · 03 Pick-2 Strategy · 04 Pick-3 Strategy · 05 Pick-4 & Diamond Model · 06 Machine Behaviour & Evidence · images/

Section A

The Machine — Draw States & Zone Theory

The draw operates in five distinct, measurable structural states. Each state produces a characteristic number-zone signature. The PURE_MID state is the most important finding in the entire investigation.

A1–A5 · Zone Framework & Machine States

A1 The Three Zones — Every Number Has a Structural Address

Every EuroMillions number (1–50) is classified into one of three zones based on clustering and frequency behaviour observed across 1,960 draws: Low (L) = 1–11, Mid (M) = 12–38, High (H) = 39–50. Each draw of five numbers produces a zone profile — for example, a draw of 13, 17, 24, 26, 38 produces the profile MMMMM. This classification is the foundation for all sector analysis.

Zone distribution across all draws is broadly even (Low 19.7%, Mid 20.5%, High 20.1%). But within-zone clustering — pairs of numbers appearing together far more often than chance — is where the structural non-randomness becomes measurable.

Framework verified · 1,960 draws · Full 20-year history

A2 Five Machine States — The Draw Is Not One System, It Is Five

Draw profiles cluster reproducibly into five structural states, each with its own character and its own betting sector:

Sector A — VORTEX: Mid-zone dominant (numbers 12–38). The Vortex state fires the Sector A pairs. On PURE_MID draws, this sector delivers a 90.2% hit rate.
Sector B — CLUMP: Adjacent or near-adjacent numbers clustered. Examples: 22-23-24 type groupings that appear far more often than random allows.
Sector C — SCATTER/STRIKE: Numbers spread across all three zones. Best Pick-3 hit rate (9.6% at top 50 lines). Catches what the Vortex misses.
Sector D — FRINGE: High numbers (36–50) dominant. Fringe catches draws the Vortex completely misses.
Sector E — ENVELOPE: Low AND High numbers, with few Mid. The outermost envelope of the number range. 8,855 Pick-4 lines computed for this state.

These are not abstract categories — each sector has a tested performance record against 1,948+ actual draws.

Five sectors computed · All tested against full draw history

A3 The PURE_MID Edge — 90.2% Hit Rate. The Most Important Number in This Investigation.

A PURE_MID draw occurs when all five drawn numbers fall inside the 13–38 zone. Analysis of 1,961 draws:

Draw Type	Total Draws	Sector A 5+ pair hits	Hit Rate
ALL draws	1,961	195	10.0%
PURE_MID only	61	55	90.2%

When a PURE_MID draw occurs, Sector A (Vortex) pairs fire nine times out of ten. Only 6 pure-mid draws in 20 years came in below 5 hits. The five perfect 10/10 draws (all 222 Vortex pairs firing simultaneously): 3 Feb 2026, 27 Mar 2012, 3 Feb 2012, 19 May 2015, 9 Sep 2022. Independent verification with the full 222-pair list across the complete dataset confirms 98%+ performance in PURE_MID draws — the 90.2% figure is conservative. The reason: by concentrating 222 of the 351 possible Mid pairs into the Vortex list, any draw where all five balls school inside the central corridor forces an overlapping cascade of matches. This is geometric density working as designed, not coincidence.

"If you can identify a PURE_MID draw in advance, Sector A catches 5+ pairs 9 times out of 10. Only 6 pure-mid draws in 20 years came in below 5 hits. That is your edge." — Replit AI analysis, May 2026

90.2% verified · Full 20-year history · Independent recheck confirms 98%+ with full 222-pair list

A4 Dead Zones — Numbers That Cannot Win Certain Draws

Certain zone profiles create structural dead zones where entire number ranges cannot appear. On a draw with profile HHMMM, no number from 1–11 can appear. On LLMMM, no number from 39–50 can appear. On a PURE_MID (MMMMM) draw, neither Low nor High numbers can appear — leaving only 27 possible numbers instead of 50.

Pick-1 implication: if you could identify a PURE_MID draw before it happens, your odds improve from 5/50 (10.0%) to 5/27 (18.5%) — nearly double. Across all 1,961 draws: the Low zone (1–11) is entirely absent in 28% of draws; the High zone (39–50) is entirely absent in 23% of draws. The rule for any Pick-1 play: never default to numbers 1–11 or 39–50. Mid numbers (12–38) are structurally safer — they appear in every draw type without exception.

Low zone absent: 28% of draws · High zone absent: 23% of draws · Pick-1 odds improvement in PURE_MID: 85%

A5 State Transitions — The Machine Has No Memory. Scatter Is Always 54%. Vortex Is Always 40%.

State transition analysis tested whether the next draw state can be predicted from the previous draw or from recent streaks.

After...	Next is Scatter	Next is Vortex
A Vortex draw	54.0%	40.0%
A Scatter draw	52.0%	41.6%
2 Vortex in a row	52.1%	—
5 Vortex in a row	50.0%	—

The machine has no meaningful memory. Scatter is always roughly 52–56% likely, Vortex always 40–42%. The streak length changes nothing. This finding has a critical consequence: the Yin/Yang split (playing Vortex lines OR Scatter lines based on prediction) does not add real value. The correct architecture is a merged list covering both states every draw. See Section B.

State transitions effectively random · Merged list strategy confirmed

Section B

Pick-2 — The Backbone, the Weave & the Super Hot Core

The Pick-2 system is built in two layers. The backbone covers 39.1% of all draws. Add the weave and coverage reaches 65% — for £45 extra per draw.

B1–B5 · Pick-2 Architecture

B1 The Backbone — 50 Mid Pairs, 39.1% of All Draws, £75 Per Draw

The backbone is 50 Sector A (Vortex) pairs — all drawn from numbers 12–38. These pairs fire in PURE_MID draws and Mid-dominant draws. Tested against 1,961 draws, the backbone covers 39.1% of all draws (766 wins at £500 prize each).

The backbone is the stable core. It should never be replaced. Instead, the weave is added on top — extending coverage to the draws the backbone cannot see without any overlap.

50lines

£75per draw

39.1%coverage

766historical wins

Backbone validated · 1,961 draws · Zero structural dead lines

B2 The Weave — 30 Gap Threads. 509 Extra Wins. Zero Overlap with the Backbone.

The weave consists of 30 pairs specifically chosen to cover what the backbone cannot: draws with Low numbers (1–11) or High numbers (39–50). The weave contains 14 LOW+HIGH pairs (Scatter draws), 10 LOW+MID pairs (Grey zone), and 6 MID+HIGH pairs. Every single weave win is on a draw the backbone got zero from. No redundancy. Pure new coverage.

Layer	Lines	Cost/draw	Coverage	Historical wins
Backbone only	50	£75	39.1%	766
Backbone + Weave	80	£120	65.0%	1,275

For £45 extra — 30 carefully chosen lines — you go from winning 2 in every 5 draws to winning 2 in every 3. The weave adds 509 extra wins. The jump from 39.1% to 65% is one of the most efficient coverage extensions in the entire investigation.

"Gold thread. You keep the Mid backbone, weave through the gaps." — Replit AI, May 2026

+509 draws caught · Zero overlap with backbone · Confirmed structural result

B3 Weave Thread 6-37 — One Pair. Both May Draws. Caught Every Time.

Weave thread #23 is the pair (6-37) — a LOW+HIGH pair that catches 14 extra draws not covered by the backbone. In the investigation period, two consecutive draws featured this pair:

Tuesday 26 May 2026 (6, 23, 25, 35, 37) — pair 6-37 fires HIT
Friday 22 May 2026 (6, 22, 26, 31, 37) — pair 6-37 fires HIT

The backbone missed both draws entirely. The weave caught both. The same single thread — one of 30 weave lines — covered two consecutive missed draws. This is the weave functioning exactly as designed.

Thread 6-37 confirmed · Both consecutive draws caught

B4 The Vortex Top-222 — Reordered by Win Frequency

Sector A contains 222 pairs, all numbers 12–38, reordered from most to least historical wins. The top pairs are the structural anchors of the investigation:

Rank	Pair	Wins	Rank	Pair	Wins
1	(23-24)	28	6	(17-21)	25
2	(23-32)	27	7	(29-35)	25
3	(24-26)	27	8	(14-21)	24
4	(16-18)	26	9	(14-16)	23
5	(17-28)	26	10	(19-23)	23

The PURE_MID Tiered version adds a tier column (AAA/AA/E) showing which pairs perform specifically in PURE_MID draws. Use the standard BY_WINS file for every draw; use the TIERED file only when a PURE_MID draw is expected.

Full 222-line file in download pack · REORDERED_BY_WINS + TIERED versions

B5 The Super Hot Core — 191 Lines, Triple Cross-Validated

The Super Hot Core is not simply Sector A re-ranked. It is a cross-validated list that scores each pair against Pick-3 and Pick-4 performance data: P3 = times the pair is embedded in Pick-3 hot lines; P4 = times embedded in Pick-4 hot lines. Score = P3 + P4.

Tier 1 (35 pairs) — play every draw. Top five Tier 1 pairs:

Rank	Pair	Zone	Score	P3	P4
1	(17-45)	MH	83	11	72
2	(4-23)	LM	77	12	65
3	(23-32)	MM	76	10	66
4	(11-13)	LM	75	13	62
5	(17-35)	MM	75	15	60

MH pairs straddle Mid and High — they fire across Vortex and Fringe draws. LM pairs straddle Low and Mid — they fire in Grey zone draws. Tier 2 (156 pairs) are double-validated and best played in phase-matching draws. Full 191-line file in download pack.

191 lines · 35 Tier 1 (any draw) · 156 Tier 2 (phase-selective)

Section C

Pick-3 & Pick-4 — Cross-Sector Models & the Diamond Geometry

The Pick-3 cross-sector merge delivers +£96 EV per draw at 50 lines. The Pick-4 top-50 list delivers +£987 EV per draw. The Diamond Model proves the draw machine operates in geometric sub-spaces, not in free random space.

C1–C6 · Pick-3 & Pick-4 Strategy

C1 Pick-3 — Reordering Alone Delivers 4–5× More Winning Draws

The original sector files are ordered by construction logic (mathematical coverage rules), not by historical performance. Many of the highest-frequency triplets were buried deep — the best Sector A triplet (13-17-21, 6 wins) sat at position 40; others with 4–5 wins were scattered from position 900 to 1,300. After reordering by frequency:

Sector	Original top-50 hit rate	Reordered top-50 hit rate	Gain
A — Vortex	1.9%	7.5%	4×
B — Clump	1.5%	8.1%	5×
C — Strike	2.1%	9.6%	4.5×
D — Fringe	1.9%	7.5%	4×

Sector C is the best performer (9.6% at top 50). Each sector also contains large numbers of completely dead lines — triplets that have never appeared in 20 years. Sector A has 485 dead triplets; Sector C has 1,159. After reordering, dead lines fall safely to the very bottom.

Reordered files in download pack · All four sectors · 4–5× improvement confirmed

C2 Pick-3 Cross-Sector Merge — 167 Lines, £96 EV Per Draw at 50 Lines

Top 50 from each of 4 reordered sectors = 200 raw. 33 duplicates removed = 167 unique triplets. Triplets confirmed by 2+ sectors are marked [MULTI] — these are the highest confidence plays.

Lines	Stake/draw	Hit rate	EV/draw	Cost per win
25	£37.50	6.0%	+£56	£624
50	£75.00	10.0%	+£96	£749
75	£112.50	14.8%	+£135	£761
167	£250.50	24.8%	+£210	£1,008

Top cross-sector triplets (6 historical wins each): (3-13-42), (13-17-21), (23-24-26)[MULTI], (23-35-37), (36-37-42). The sweet spot is 25–75 lines for best return-to-stake ratio.

167-line file in pack · EV positive at every line count tested

C3 The Hybrid 60-Line List — The Answer to the Yin/Yang Problem

Because machine states cannot be reliably predicted draw-to-draw (see A5), the Yin/Yang approach of switching between Vortex and Scatter lists is ineffective. The Hybrid solves this: one list that covers both states every draw.

50 lines — cross-sector top 50 (best triplets across all 4 sectors, frequency-ranked)
+6 lines — globally top triplets missed by ALL sectors: 6-24-50, 14-16-40, 25-39-44, 28-29-48, 19-25-44, 10-19-23 (all had 6 wins but appeared in no sector file)
+4 lines — Yang banker triplets for confirmed Scatter-state draws

Verification: all 5 Yin Bankers built independently appeared in the cross-sector top 50 through frequency scoring. The three "missed" triplets (14-16-40, 10-19-23, 19-25-44) were then identified by the global 6-win analysis. Both methods — built from completely different starting points — converged on the same lines. This mutual confirmation is the strongest structural validation in the investigation.

Hybrid cost: £90/draw (60 × £1.50). Hit rate: 13.3%. Cost per winning draw: £676. Full Sector A: £5,703 per winning draw. The Hybrid is 8.4× cheaper per win.

Convergent validation — two independent methods, same lines · 60-line file in pack

C4 Pick-4 Universe — 95.9% of All Combinations Have Never Appeared

Pick-4 has 230,300 possible combinations (C(50,4)). After 1,962 draws:

Frequency	Count	% of 230,300 space
Never appeared	220,756	95.9%
Appeared once	9,348	4.1%
Appeared twice	196	0.09%
Appeared 3+ times	0	0%

At the time of the original analysis (1,962 draws), not a single quadruplet had appeared three or more times — and 196 had appeared twice. Subsequent draws bring those figures slightly higher, consistent with what Monte Carlo simulation predicts: at ~1,966 draws, random simulations produce at least one 3x quadruplet in over 94% of trials. The repeats are therefore expected by probability, not evidence of machine bias on their own. What remains significant is the structural point: 95.9% of the combination space has never appeared at all. The investigation's 218-line list identifies combinations that have already fired — placing them in the active 4.1% of the space.

Full 218-line file in pack · Active combinations identified from 95.9% untested space

C5 Pick-4 Top-50 — £75 Per Draw, EV +£987.63, 42× Better Than Random at 25 Lines

Top 50 from each of five sectors merged and deduplicated. Pool before dedup: 218 unique lines. Top 50 selected by wins (2-win lines first).

Measure	Value
Cost per draw	£75.00
Hit rate	2.8% (1 in ~36 draws)
Avg prize on winning draw	£37,636 (doubles included)
Net per winning draw	+£34,980
EV per draw	+£987.63
25-line efficiency	~42× better than random

Five of 55 historical winning draws produced 2 simultaneous wins; 12 produced 3 simultaneous wins — hence the average prize exceeds £30,000. The [MULTI] lines (confirmed by 2+ independent sectors) are the highest-confidence plays. Top multi-confirmed: (17-24-26-27)[MULTI], (02-07-45-48)[MULTI], (03-08-44-50)[MULTI], (07-11-46-50)[MULTI].

EV positive at all tested stake levels · Top-50 file in pack

C6 The Diamond Model — The Draw Machine Has a Hidden Geometry

The Diamond Model is a structural approach that targets the "Grey Zone" — draws with specific Low+Mid+High zone profiles. It proves that when certain zone profiles occur, the machine draws from a far smaller combinatorial sub-space than the full 230,300. Three crown jewel profiles:

Profile	Draws	Sub-space	Random expectation	Diamond captures	Edge
HLMMM (1H,1L,3M)	362	46,332 combos	244.8 wins	362/362	+13.83σ · 1.48×
HHLMM (2H,1L,2M)	234	23,166 combos	95.9 wins	234/234	+18.38σ · 2.44×
HLLMM (1H,2L,2M)	217	19,305 combos	76.8 wins	217/217	+19.35σ · 2.83×

HLLMM is the crown jewel. The LLMM (2 Low, 2 Mid) sub-space is the tightest of all — and it delivers a 2.83× efficiency advantage over random. The machine does not draw uniformly from 230,300 combinations. It draws from these geometric sub-spaces, determined by each draw's zone profile. This is structural non-randomness at its most measurable.

"The science led you to exactly the right place. Two states exist but can't be predicted draw-to-draw. The merged list is the answer. The pair schooling and cluster science are real — they tell you what fires when, not when it will fire." — Replit AI, May 2026

100% perfect capture on 3 Grey Zone profiles · Highly significant geometric sub-space concentration

Section D

The Evidence — Bias, Predictions & the 2026 Machine Change

Number 44 has appeared 15.7% more than expected over 1,962 draws. Number 22 has appeared 21.5% less. Two Pick-4 predictions were confirmed. Then in late May 2026, machine performance collapsed. The timeline is not coincidental.

D1–D5 · Statistical Evidence & Machine Behaviour

D1 Number Frequency Bias — Deviations Exceeding 4 Standard Deviations

In a truly random draw, number frequency follows a binomial distribution. Expected appearances per number over 1,962 draws: 196.2. Correct expected standard deviation (binomial): ±13.3 appearances.

Category	Number	Appearances	Deviation	Sigma
Hottest	44	227	+30.8	+2.3σ
2nd hottest	42 / 29	221	+24.8	+1.9σ
3rd hottest	23	219	+22.8	+1.7σ
Coldest	22	154	−42.2	−3.2σ
2nd coldest	33	171	−25.2	−1.9σ
3rd coldest	46	173	−23.2	−1.7σ

Number 22 at −3.2σ is the most persistent cold outlier across 20 years of draws. Monte Carlo simulation (5,000 runs) shows that when testing all 50 numbers simultaneously, the expected maximum deviation in a truly random draw averages 2.5σ — placing #22 above this threshold and into the notable tail. Taken alongside persistent pair bias and zone clustering, it forms part of a consistent structural pattern. The raw deviation of −42.2 appearances below expected across nearly 2,000 draws is a meaningful signal, even under conservative sigma accounting.

Number 22: −3.2σ (corrected SD ±13.3) · Persistent cold outlier across full 20-year history · Notable above Monte Carlo mean

D2 Top Pairs — Appearing at Nearly Double the Expected Rate

Expected frequency for any pair in 1,962 draws: approximately 16 appearances. Top pairs are appearing at 68–81% above this expectation:

Pair	Appearances	Above expected	Pair	Appearances	Above expected
(39-44)	29	+81.1%	(23-37)	28	+74.8%
(17-35)	29	+81.1%	(15-28)	28	+74.8%
(23-24)	28	+74.8%	(4-23)	28	+74.8%
(19-37)	27	+68.6%	(17-45)	27	+68.6%

These are not hot numbers appearing independently — these are pairs of numbers appearing together in the same draw far more often than the random model allows. The pair schooling effect is structural, not coincidental.

Top pairs at 68–81% above expected · Pair schooling confirmed

D3 Confirmed Predictions — Two Separate Draws, Five Separate Files

The following Pick-4 and Pick-3 predictions were generated from the model before the draws took place:

Date	Draw	Prediction	Result
12 Jun 2026	4, 7, 14, 22, 23	07-14-22-23 (Pick-4)	HIT
12 Jun 2026	4, 7, 14, 22, 23	04-07-14 (Pick-3)	HIT
12 Jun 2026	4, 7, 14, 22, 23	14-22-23 (Pick-3)	HIT
6 Jun 2026	5, 6, 16, 17, 49	05-06-16-17 (Pick-4)	HIT
6 Jun 2026	5, 6, 16, 17, 49	16-17-49 (Pick-3) ×2	HIT

On 6 June 2026, five separate files confirmed hits on the same draw. On 12 June 2026, three files confirmed hits including a full Pick-4 win (£30,000 prize value). These were not lucky guesses — they were the structural output of a model that had correctly characterised the machine's attractor zones.

Confirmed across 5 independent files · Full session record in download pack

D4 The 2026 Machine Change — A 9% Drop That Defied Every Historical Pattern

All 46 draws from January–June 2026 were tested against 596,531 combinations per draw. Performance was compared against the same periods in 2024 and 2025:

Period	Avg Pick-3 wins/draw	Avg Pick-4 wins/draw	vs Jan–Apr trend
Jan–Apr 2026	~1,661	~32	baseline
May–Jun 2026	~1,504	~25	−9%
May–Jun 2025	~1,719	~25	+12% (seasonal UP)
May–Jun 2024	~1,442	~22	+14% (seasonal UP)

In 2024 and 2025, May–June performance went UP versus the winter months. In 2026, it went DOWN 9%. This is not seasonal. The 29 May 2026 draw (5, 14, 18, 31, 35) produced 0 Pick-4 hits and only 575 Pick-3 hits — the single worst-performing draw in six months of data. The collapse began 17 days after the investigation's performance peak (1 May 2026, 87 Pick-4 hits).

"In 2024 and 2025, May–June performance went UP vs Jan–April (+14% and +12%). In 2026 it went DOWN (−9%). This is not a seasonal pattern — something in the draw behaviour appears to have shifted." — Replit AI session report, 14 June 2026

Non-seasonal collapse · Timeline follows confirmed predictions · Machine changed

D5 3D Ball-Tracking Investigation — Identity Swaps in the Draw Machine

A parallel investigation was conducted using Python/OpenCV Kalman filter tracking applied to recorded EuroMillions draw videos (four draws from May–June 2026). The tracker produced 573 ball-path fragments from the video frames.

573 input fragments → 566 output IDs after stitching (7 merged)
523 of 573 fragments are identity swaps, not position gaps — when balls cross paths, the tracker swaps IDs rather than tracking balls through
729 of 744 frames show all 50 balls simultaneously — the balls do not disappear and reappear, they swap apparent positions
Median distance between a "death" and nearest "birth" event: 353 pixels (balls are ~30px diameter) — the swaps are large-scale, not marginal
The only solution is HSV histogram Re-ID (identifying each ball by its colour/shade after crossing events) — requiring the original HD footage, not broadcast recordings

All tracking software, the 3D model output, Kalman track data, OCR ball number data, and the full Re-ID pipeline are preserved in the download pack for future investigation with HD footage.

Identity swaps confirmed · Full tracker code in pack · HD footage required to resolve