In short: this section does not predict "lucky" numbers and does not change the ticket's odds. It builds a curated set out of the 49 numbers, sized 18–48, using normal-distribution statistics (the ball-draw process itself is a random process with a normal distribution).
The historical archive and the analysis-depth parameters [200, 100, 50] split the roles between long-term and short-term statistics. The short-term statistic is "hotness": how many times a number hit in the last 50, 100, or 200 draws.
1. What the method does
- Each of the 49 numbers gets a Score from the archive (foundation + local hotness).
- A dome is built from the scores (a band around the mean ± k·σ on a normal distribution).
- The dome's width is tuned so that exactly N numbers fall inside (pool size 18–48).
- This set can be viewed, backtested, and sent to the basket.
The odds of any 6-number combination are the same. Only the set's composition changes, and the "crowd" you share the prize with if the set hits.
Algorithm picture (in the Analytics tab): "Dome analysis" (dome_dashboard_{size}_{lang}.png).
2. Setup
- On the left, pick a draw snapshot and, if needed, the archive from year in the sidebar.
- Open the Curated Pools section.
- At the top, choose the backtest depth: 10 / 20 / 30.
- 10 (tier 1 / guest)
- 20 (tier 2)
- 30 (Explorer tier)
The on-screen rating is scaled to a "per 10" baseline.
3. How a number's score is computed (the set algorithm)
The whole historical archive
The full available draw array (from the very first draw to the one being analyzed) drives the number's fundamental metrics. It feeds 3 of the 4 components of the final score:
| Component | What it is | Weight |
|---|---|---|
Overall frequency (overall_frequencies) | how many times the number hit across all available history | 3.0 |
Pair frequency (pair_frequencies) | how often the number appeared together with other balls across all history | 1.5 |
Skip / idle count (skip_counts) | how many draws in a row the number has not hit as of now | 1.0 |
Pair frequency is a "synergy" indicator for a number. It rewards balls that don't just hit on their own, but regularly show up alongside other frequently appearing numbers.
Local "hotness" over the last 50–200 draws
Short-term statistic: how many times a number hit in the last 50, 100, or 200 draws (recent_frequencies, weight 2.5).
Final Score
Each of the 49 numbers gets a score as the sum of weighted, normalized (÷Max) coefficients:
Score = 3.0 · (Freq_overall / Max) + 2.5 · (Freq_recent / Max) + 1.5 · (Freq_pair / Max) + 1.0 · (Skip / Max)
The dome (normal distribution)
The mean μ and standard deviation σ are computed over all 49 scores. Numbers whose score falls within the band [μ − k·σ, μ + k·σ] enter the set. The coefficient k is tuned via the golden-section method so the band contains exactly N numbers (the pool size). If there are slightly too many/few, they are trimmed or topped up by score ranking.
This is exactly the band shown in the "Dome analysis" picture.
What's the practical difference between 50, 100, and 200?
| Depth (RECENT) | What the algorithm evaluates | Model behaviour |
|---|---|---|
| 50 draws | The most "aggressive" trend of the last few weeks. | Reacts quickly to spikes in number activity, sensitive to current "streaks". |
| 100 draws | A balanced medium-term trend. | Smooths out random outliers from recent draws while staying current. |
| 200 draws | Long-term trend stability. | Picks stable numbers that are less affected by short-term fluctuations. |
4. All section features (by access level)
A. Backtest depth 10 / 20 / 30
Picker at the top of the section.
(tier 1 → 10 · tier 2 → up to 20 · Explorer tier → up to 30)
B. "Archive from year (precalc)" slider
The archive start year for the "best pool" rankings. The precalc step is usually 10 years.
(the widget is visible to everyone; useful meaning and "honest" values from tier 1; for guests the metrics/tables are often blurred)
C. "Best pool (out of 28–42) by archive year" card
Shows the best result by set size. A larger set doesn't always (but very often does) give higher prediction efficiency.
(available from guest level with partial blur on table results. Full information from tier 1)
D. "Use best size and depth" button
Fills the UI with the best size and RECENT depth from the precalc.
(tier 1)
E. "Best pools per each precalc year" summary
Table: year → size, recent, rating, efficiency, 6 / 5 / 5+C.
(tier 2; for guest/tier 1 — structure + blur)
F. Top-3 pools by year
The three best size×RECENT combinations for the selected year.
(tier 1)
G. Full ranking table "Top pools for this year" (expander)
The extended list of precalc candidates.
(tier 2)
H. Pool-size slider 48…18
Choosing N numbers in the set.
- Guest: only 18–19 and 43–48
- All sizes 18–48 (tier 1 and above)
The "Tier: Open / I / II / Archive" label on the size is a marketing tag for the set style, not a membership gate.
I. RECENT analysis depth (50 / 100 / 200)
How many recent draws are heavily weighted in the score as local "hotness" (see the table above).
(available together with the chosen size; sizes themselves — see H)
J. Metric cards for the selected size
Size, combination coverage, 6-hit efficiency, hits6/N at the chosen BT depth.
(for the allowed size — everyone; efficiency = Lift×100%, 100% ≈ a random set of the same size)
K. "Pool numbers" tab
The balls of the current set.
(within the allowed size — everyone)
L. "Add to basket"
Sends the current set.
- Guest: limit of 1 set in the basket
- tier 1+: several sets
M. "Analytics" tab
- Backtest rating + per-draw table
(tier 1 / any paid tier; for guests rating/hits are often blurred)
- Dome expander — a PNG of the normal distribution of scores with the selection zone
(if the file exists in the snapshot — it can be shown to everyone)
- Table from the pool file: Distance / Recent / 6 / 5 / 5+C
N. Related tools outside the Curated Pools screen
Auto Pick, intersections/unions, the rich basket — layered on top of pools.
(Explorer tier)
5. "Best pool by year" precalc (a separate loop)
After generating sets, the system separately compares which size and which RECENT worked best on history:
- Only sizes 28–42 take part in the ranking (not the extreme 18 and 48).
- Sweep: size × RECENT × archive start year (~10-year step).
- Walk-back backtest with 10 / 20 / 30 slices.
- Rating: 6→500, 5→100, 4→10, 3→1, then normalized to a base of 10.
- Efficiency:
(hits6/N) / C(size,6)/C(49,6) × 100%. - Sort order: rating → efficiency → hits6 → hits5 → smaller size.
That's why the card is called "best out of 28–42".
6. Typical scenarios
Guest
- Size from the available zone (18–19 or 43–48).
- RECENT 50 / 100 / 200 (see the difference table).
- Numbers tab → optionally 1 set to the basket.
- Year rankings — showcase (blurred).
Tier 1
- BT depth 10.
- Precalc year → best-pool card → "Use best size and depth".
- Compare the Top-3.
- Numbers → to the basket.
- Analytics: backtest table + dome.
Tier 2
Same, plus full year tables + BT 20.
Explorer tier
Same, plus BT 30 → basket → Auto Pick / Compare / intersections with the pool.
Important to remember
- A pool does not make a combination "more likely".
- Score and the dome order archive statistics on a normal distribution — this is not a forecast of tomorrow's draw.
- Backtest efficiency > 100% ≠ a guarantee for tomorrow; it's history relative to a random set of the same size.
- The precalc ranking covers only 28–42; check the extreme sizes separately via live/file BT.
- Pair frequency rewards "synergy" (numbers appearing together), not solo frequency.
- The dome picture is a visual of the Score → μ±kσ → N-number set steps.