blindgrid

Lottery grids from cryptographic randomness,
inside a budget you cannot exceed.

Install · What it refuses to do · Households · Tablets · Language · Add your own lottery · Responsible gambling

What this is

A small CLI that answers two questions once a month:

• Which draws should I play? Given a budget and a set of lotteries, it allocates the money by weight, picks real calendar dates at random, and stops. No pattern, no habit, no "I always play on Saturdays".
• Which numbers? Drawn from secrets.SystemRandom, then filtered against a handful of anti-pattern rules so the grid does not look like something a human would have chosen.

The plan is printed as a table and exported to a Markdown file.

What this is not

It does not predict anything, and it never will. Lottery draws are independent events. The numbers drawn last week carry no information about next week's draw; a number that has not come up in two years is not "due". This is not an opinion about lotteries, it is what independence means.

So the following are permanently out of scope, and pull requests adding them will be declined:

What is left, then? Two honest things: randomness without human bias, and a budget ceiling that does not move.

Why filter the numbers at all, if every combination is equally likely?

Because the payout is not equally likely. Every combination has the same probability of being drawn, but combinations humans favour — birthdates below 32, sequences, neat patterns — are picked by thousands of other players at the same time. When such a combination wins, the jackpot is split. Filtering does not improve your odds of winning; it improves what you would receive if you did.

The rule set stays deliberately small for the same reason. Every additional rule shrinks the sample space, and a heavily filtered "random" pick is just a different kind of predictable.

Install

Runs on Linux, macOS, Windows, and iOS through a-Shell.

pip install blindgrid        # or: uv tool install blindgrid

That is the whole of it if you already have Python 3.11 or newer — the package is on PyPI. The scripts below add a little more: they pick an install method for you, write a starter configuration, and tell you what to put on your PATH.

macOS and Linux

curl -fsSL https://raw.githubusercontent.com/adrnbttr/blindgrid/main/install.sh | bash

Windows (PowerShell)

irm https://raw.githubusercontent.com/adrnbttr/blindgrid/main/install.ps1 | iex

Either one installs into your own user profile, writes a starter config, and tells you what to add to your PATH if anything is missing. Neither asks for sudo or administrator rights, and neither writes outside your home directory. Prefer to read the script first? Download it and open it — both are short, and both are linted in CI (shellcheck, PSScriptAnalyzer).

Python 3.11 or newer is required, except with uv, which brings its own. 3.11 is a floor, not a default: it is the newest Python a-Shell ships, and raising it would drop iPad support — a test says so. Uninstalling is install.sh --uninstall / install.ps1 -Uninstall, and your config survives it.

iPad and iPhone, in a-Shell (free, App Store):

pip install blindgrid

Then python3 -m blindgrid generate. Every dependency is pure Python, which is what makes this work on a device with no compiler. If pip is not on your path there, curl -sL https://raw.githubusercontent.com/adrnbttr/blindgrid/main/install.py | python3 does the same and sets up a starter config — see On a tablet or a phone.

Full installation guide — per-platform details, other install methods, file locations, troubleshooting, updating, uninstalling.

Use

blindgrid config show      # where your config lives and what is in it
blindgrid generate         # plan the current month

generate asks for two things — the budget for this month, and which lotteries to include — then prints the plan and writes plan.md. It plans the current month by default; --month is only there for planning ahead.

A month is drawn once

Run generate again and you get the same plan back, not a new one:

$ blindgrid generate
Plan already drawn on 2026-09-01. Showing it again — pass --force to draw a new one.

This is partly convenience — you can come back to it while you fill your grids — but mostly it is the point. A tool that redrew on every run would let you reroll until the numbers looked right, and picking the draw you like best is exactly the bias the filters exist to remove.

--force is there for the genuine mistake, a mistyped budget or the wrong lotteries. It says what it is replacing before it does it.

The plan lives in one file in your state directory (see file locations), replaced when the month turns. It is a self-contained snapshot, so editing your config afterwards never changes a plan you already hold. Draws whose date has passed are struck through when the plan is shown again, so what is left to play is obvious.

On a tablet or a phone

The plan is measured against your terminal and laid out to fit it, rather than assuming a width.

The table is deliberately narrow. The month is already in the title, so each row carries the day alone instead of repeating 2026-09-, and the weekday is abbreviated the way each language writes it — German uses Mi, not a truncated Mit. That is twelve columns saved, which is what puts a two-player plan inside the ~90 columns an iPad gives you in portrait. Rotating the device no longer changes the layout.

Below that it sheds what it can spare, one thing at a time, rather than giving up on the table at the first squeeze:

Those widths are the point at which each layout is guaranteed; a column or two either side it depends on the draw, since a row of single-digit numbers is narrower than one full of two-digit ones. Nothing is measured in advance — the layout is chosen against your actual terminal, every run.

What never goes, at any width, is the answer to what do I play, and when: the date, the lottery, and the numbers.

The summaries follow the same rule — squeezed, they become plain lines rather than a table with Lottery cut down to Lotte….

Narrower still, on a phone or in a split view, the table would shred the numbers across lines, so it becomes a list:

September 2026

 8 Tue  EuroMillions  Adrien  2.50 EUR
    10 11 30 34 47  ·  stars  8 12

 9 Wed  Loto  Adrien  2.20 EUR
     7 20 22 34 35  ·  lucky  2

The numbers are the one thing you copy onto a paper slip, so they are never wrapped and nothing is ever cut with an ellipsis, at any width. This stays readable down to about 40 columns.

--compact and --table override the choice if you would rather decide yourself. The Markdown export is unaffected: it keeps full dates and full weekday names, since it is read away from the terminal with no title above it to supply the month.

Playing as a household

Two people, one month, one command. Everyone keeps their own ceiling and their own games:

blindgrid player add     # name, ceiling, which lotteries, what weights
blindgrid generate       # asks each person for their budget, then plans

player add takes options too, which is what you want in a shell that cannot draw a checkbox — the iOS ones cannot:

blindgrid player add --name Marie --ceiling 25 --play Loto --play EuroDreams=1

╭──────┬─────┬────────┬──────────────┬────────────────────────────────┬──────────╮
│ Date │ Day │ Player │ Lottery      │ Numbers                        │     Cost │
├──────┼─────┼────────┼──────────────┼────────────────────────────────┼──────────┤
│   11 │ Fri │ Adrien │ EuroMillions │ 10 11 22 35 36  ·  stars  8 12 │ 2.50 EUR │
│   19 │ Sat │ Marie  │ Loto         │  4  5 10 38 42  ·  lucky  3    │ 2.20 EUR │
│   21 │ Mon │ Marie  │ EuroDreams   │  7  9 18 32 34  ·  dream  5    │ 2.50 EUR │
╰──────┴─────┴────────┴──────────────┴────────────────────────────────┴──────────╯

Three things hold:

• Nobody spends against anyone else's ceiling. Each person has their own max_monthly_budget, and one player's budget never changes another's share. There is no household cap, on purpose: if two people together want to spend more than they should, that is a conversation, not something software gets to arbitrate.
• Everyone plays only their own games. Weights are per person; a weight of zero, or simply leaving a lottery out, means they never play it.
• Draws are spread out. The household covers as many different draws as it can before any date is played twice.

On spreading draws

Spreading does not improve anyone's odds, and the tool will never claim it does. Two grids are two independent chances whether they sit on the same draw or on two different ones, and the probability that at least one of them wins is identical either way.

What it does buy is exposure to more distinct jackpots — some of which have rolled over and are larger — and the certainty that two people in one house are not holding near-duplicate tickets for a single draw. That is diversification, not an edge.

When there are fewer draws left than grids to play, dates are shared rather than grids dropped, and the output says which. Losing a grid someone budgeted for, to satisfy a rule that does not change the odds, would be the wrong trade.

[[player]]
name = "Adrien"
max_monthly_budget = 40.00
  [player.weight]
  EuroMillions = 1.0
  Loto = 1.0
  EuroDreams = 0.4

[[player]]
name = "Marie"
max_monthly_budget = 25.00
  [player.weight]
  Loto = 1.0
  EuroDreams = 1.0

Declare nobody and blindgrid stays in single-player mode, exactly as before.

Usage: blindgrid [OPTIONS] COMMAND [ARGS]...

  generate    Build this month's plan: how much to spend, which draws, which numbers.
  config      Inspect and edit the configuration.
  lottery     Manage lottery definitions.
  player      Manage the people who play.
  version     Print the installed version.

Useful options on generate:

Every prompt has an option that skips it, so the whole tool works without an interactive terminal — in a script, over SSH, or in a shell that cannot draw prompts.

Language

The interface speaks English, French, Spanish and German, and follows your system locale by default — so there is usually nothing to set at all.

Order of precedence: --lang, then BLINDGRID_LANG, then language in the config file, then your system locale, then English.

The whole interface moves, including month and weekday names:

$ blindgrid generate --lang fr

              Tirages à jouer — septembre 2026
╭──────┬──────┬──────────────┬────────────────────────────────┬──────────╮
│ Date │ Jour │ Loterie      │ Numéros                        │     Coût │
├──────┼──────┼──────────────┼────────────────────────────────┼──────────┤
│    4 │ ven  │ EuroMillions │  5 19 28 32 50  ·  stars  1  2 │ 2.50 EUR │
╰──────┴──────┴──────────────┴────────────────────────────────┴──────────╯

  Poids · Alloué · Engagé · Grilles · Non dépensé

Terms follow each country's own vocabulary rather than a literal translation: a played grid is an apuesta in Spanish and a Tipp in German, which is what the operators there call it.

Only the interface is translated. Lottery labels, pool names and player names are printed exactly as you wrote them, and the weekday keys in the config file stay English (monday, tuesday, …) — those are a file format, not text for reading.

Adding a language is one file in src/blindgrid/i18n/ plus one line in its __init__.py. Tests check that every catalogue carries exactly the same keys and placeholders as the English one, so a half-finished translation fails the build rather than leaking English sentences into someone's session.

Configuration

Two files:

• config.example.toml — versioned, three French games as an illustration.
• config.toml — yours, gitignored, holding your own ceiling.

The ceiling is the point of the file:

max_monthly_budget = 40.00

generate refuses any budget above it. There is no override flag. A cap you can raise in the moment while looking at a jackpot headline is not a cap.

Adding a lottery from any country

Nothing in the code knows about France, or about any specific game. A lottery is a price, a set of draw days, and one or more pools of numbers. That is the whole model, and it covers every draw-style lottery I am aware of.

Here is the US Powerball — five numbers from 69, plus one from 26, drawn on Mondays, Wednesdays and Saturdays at $2 a play:

[[lottery]]
label = "Powerball"
currency = "USD"
price_per_grid = 2.00
draw_days = ["monday", "wednesday", "saturday"]
weight = 1.0

  [[lottery.pool]]
  name = "numbers"
  count = 5
  max = 69

  [[lottery.pool]]
  name = "powerball"
  count = 1
  max = 26

Drop that into config.toml and it works. Or run blindgrid lottery add and answer the prompts.

The filters adapt on their own: the 1-from-26 Powerball pool is too small for a parity or spread rule to mean anything, so those rules switch themselves off for it rather than searching forever for a combination that cannot exist.

Weights

Weights are relative shares of one envelope, not draw counts.

weight = 1.0   # EuroMillions
weight = 1.0   # Loto
weight = 0.4   # EuroDreams — played, but a smaller slice
weight = 0.0   # disabled, without deleting the definition

With a €40 budget and those weights, EuroMillions and Loto receive €16.66 each and EuroDreams €6.66. Each share then buys as many grids as it can afford, and the leftover in each share stays unspent. It is never pooled, never redistributed, never rounded up into one more grid.

If a share cannot cover a single grid, that lottery is skipped for the month and the output says so explicitly. It never borrows from another share to make itself viable.

The randomness

Every number comes from secrets.SystemRandom, which draws from the operating system's entropy pool. The random module is not imported anywhere in the package — its Mersenne Twister is seedable and reproducible, which is a liability here — and a test enforces that.

The same source picks which dates to play, because a schedule chosen by habit is as predictable as numbers chosen by birthdate.

Development

git clone https://github.com/adrnbttr/blindgrid.git
cd blindgrid
uv venv && uv pip install -e ".[dev]"

pytest
ruff check . && ruff format --check .

CI runs the suite on Linux (3.11, 3.12, 3.13), macOS and Windows, and installs the project with each of the three installers on a clean runner before uninstalling it again.

CONTRIBUTING.md covers the rest — including what will not be merged, which is worth reading first — and CHANGELOG.md records what changed in each release.

Responsible gambling

This tool exists because a budget written down and capped is better than a budget decided in the moment. It does not make gambling profitable. Over any meaningful number of draws, the expected return of a lottery ticket is negative — that is how lotteries fund themselves — and no arrangement of numbers changes that.

Play with money you have already decided to lose. If gambling has stopped being a small monthly cost and started being something else, these services are free and confidential:

• France — Joueurs Info Service, 09 74 75 13 13, https://www.joueurs-info-service.fr
• United Kingdom — GamCare, 0808 8020 133, https://www.gamcare.org.uk
• United States — 1-800-GAMBLER, https://www.ncpgambling.org
• International — https://www.gamblingtherapy.org

License

MIT — see LICENSE.