NumberSight — Monster Forecaster (Front-End Only)

1) Specimen Intake

Upload CSV, confirm the 7 number columns, then run training. Defaults are safe. Deep knobs are tucked away.

Tip: last ~1000 rows is enough.

CSV file

Expected: Lotto Max draws with 7 ascending numbers per row (1..50). Any extra columns are fine.

Delimiter

Header row

Data rows to use

Window size

Tolerance (hit)

Column mapping (auto-detected, editable)

Detected columns (preview)

Number columns (0-based indexes, 7 values)

Example: 2,3,4,5,6,7,8. Must be exactly 7 indexes.

Skip rows (top)

Min value / Max value

Advanced training knobs (optional)

Hedge learning rate (η)

EMA rate (β)

SES α candidates

Microtendency threshold (θ)

Enforce uniqueness

Keep legacy Monster

Speed

Left = cinematic updates. Right = near-instant (still safe on mobile).

Academic disclaimer: Lottery draws are effectively random; this is a learning tool, not gambling advice.

2) Live Vitals

Mission control. Tabs below show detail when you want it.

front-end only

Rows (usable)

—

Cycles

—

rows − window

Hit rate (±tol)

—

Avg error

—

mean |forecast−actual|

Waiting for data.

Forecast

Per-Position Winners

PHE Weights

Microtendency

Exports

After training, you get PHE and PHE-MC (microtendency corrected) forecasts, plus optional legacy Monster & Pure Raw.

No forecast yet.

3) Lab Journal (last cycles)

Readable log. Forensics live in exports.

last 8 cycles

No data yet.