Paused
Tap anywhere to resume
Loading…
No guesses. Ever.
Every board is proven solvable before you see it.
Greg dreamed up the modifiers. The mines are his fault.
GregSweeper © 2026 Christopher D. Wells · All rights reserved
Support GregSweeper
(voluntary, helps cover domain + hosting + app-store fees).
Pick a handle for the leaderboard. You can change it anytime in Settings.
Shown on the public leaderboard: a handle, not your real name.
In Daily and Weekly, hitting a mine does not end the game. Your revealed cells stay revealed and a 
marks the mine you hit.
The time cost depends on how much that mine was anchoring the rest of the solve:
So bombing a "throwaway" mine is a cheap shortcut; bombing a critical mine is real time. The popup on each hit shows the breakdown.
The Climb still ends on a mine, so flag carefully.
Every standard board is checked by a solver before you play it. From the marked start square there is always a next square you can prove safe, all the way to the end. You never have to gamble.
The ✓ Certified chip above the board opens that board’s own check, and after a loss every square you could have proved safe is outlined. Chaos is the one exception to the guarantee, and it says so.
Reveal every safe square without hitting a mine. Each number counts the mines in the squares touching it, corners included. Blank squares touch no mines at all; the numbers are your clues.
Not every board is a square grid. Alongside Classic there are six others: Octagons, Honeycomb, Paving Stones, Petals, 3D Cubes and Kites. About half of all dailies use one, and The Climb works through every one of them.
The rules do not change. The only thing a shape changes is which cells count as neighbors, and the first time you meet a shape a card shows you one cell with everything it touches. Cells that meet at just a corner are neighbors too, exactly as diagonals are on a Classic board.
Tap / Click a square to reveal it
Long-press / Right-click to flag a mine you have worked out. Or tap the flag toggle in the top row to switch modes, so every tap places a flag instead.
Tap a revealed number whose mines are all flagged to open its remaining neighbors at once (a chord). Chording with wrong flags detonates.
Hitting a mine in Daily, Weekly or Challenge is not the end. The mine is marked where it blew, a time penalty is added (sized by how much that mine was hiding), and you keep playing. In The Climb, a mine ends the run.
The Daily carries a par time: the predicted clean-solve time for that exact board, refit nightly from real plays. After a win you see the board’s par next to your time, and your personal par drifts toward your own pace as you play more.
Arrows move around the board • Enter / Space reveal (chord on a satisfied number) • F flag
R restart • 1-5 power-ups • Esc close
Daily: one puzzle a day, the same board for everyone, online leaderboard and streaks. The shape varies day to day.
Weekly: one puzzle for the whole week, one try a day, your best time counts
The Climb: 250 levels in blocks of five, working through all seven shapes and introducing the modifiers one at a time, with power-ups and a checkpoint every five levels
Challenge: build a run of one to ten boards, choosing the shapes, modifiers, length and mine count yourself, then race your own clock through them
Greg’s Gym: short practice boards, one deduction pattern at a time. A square only opens when the clues prove it is safe, so you cannot luck through.
Chaos: fast rounds of random modifiers (unlocks at Climb Level 100). The one mode OUTSIDE the no-guess guarantee: boards are not verified solvable, and mines can move.
Some boards add Modifiers: special squares that bend the rules. The first time each one appears, GregSweeper explains it in full; the list below is the quick reference. Mid-game, tap a modifier’s icon above the board to re-read its rule.
Reveal Safe: opens a square the clues already prove safe, a worked example rather than a random gift
Shield: blocks one mine hit
Scan: shows mine count in a row or column
Magnet: extracts every mine in a 3×3 area off the board
X-Ray: reveals mines in a 5×5 area
Lifeline: auto-saves you from one mine
Earned by winning Climb levels, and spent there too. Daily, Weekly and Challenge are played clean.
Greg checks every board before you play it, and keeps notes on how you solve.
His time on the daily is the par you race.
Greg banked you a molt day. Miss a day and it covers you, so your streak lives on.
You can hold two.
GregSweeper is now at gregsweeper.com, so you're in the right place.
Played before on the old address? A signed-in account keeps your streak and stats synced across devices. Sign in any time from Settings.
Play any past board you haven’t finished. A day you’ve already cleared keeps its check and stays put, because a daily is a one‑off. These runs never touch your streak or your daily completion, and Greg still keeps your time.
No past boards in this month.
Replay any past week’s board. A week you already played keeps its check and stays put, because the week’s board is one board. These runs use none of this week’s attempts, never touch your week streak, and never reach the leaderboard.
No past weeks yet.
| # | Name | Time | Par |
|---|
No entries yet.
Your friend code
Share it — anyone who enters your code becomes your friend. Codes last 15 minutes.
The Climb
Challenge setupSign in to keep your streak and progress in sync between phone and desktop.
✓ Check your email, then click the link on whichever device you want signed in.
Signed in as …
Your streak and progress are synced across devices.
Used for daily leaderboard submissions. Saves automatically.
Uses distinct colors and shape markers for numbers.
Use the original bomb and flag on every theme. Numbers, backgrounds, and Greg stay themed.
Comfortable and Large can put a board past the screen edge; scroll it, pinch, or use the zoom buttons. Double-tap a number you are finished with to center the view there, and double-tap it again to see the whole board.
Walk through the basics again.
Version and credits.
Pings you when today's daily / weekly is ready.
Extra ping at 8pm ET only if your streak ≥ 3 days and you haven't played today.
Forces a fresh download. Try this first if anything looks stuck or wrong.
Erases all stats, achievements, and leaderboard data on this device.
…
Removes your leaderboard rows, weekly best-times, and progress under your anonymous ID from Firebase. Opens a pre-filled email, and your data are scrubbed within 30 days.
Opens a GitHub issue with your device state pre-filled. No personal data leaves the device until you actually submit the issue.
Sponsor GregSweeper →
Helps cover domain, hosting, and app-store fees (optional, the game is free either way).
Privacy policy → About the par model →Shows what this device is running (useful if something looks off).
Greg’s Time (par) is the typical solve time for a board. Your par is that adjusted by your handicap, your usual pace versus Greg across recent dailies (negative is faster). Delta is how far under or over your par you finished on a given day.
Your typical over/under Greg-par across all your plays. Lower is faster. Solid line is your career average up to each daily; dashed line is your last-10-plays rolling mean. When the dashed line sits below the solid one, you’ve been improving lately.
Each dot is a past daily, positioned by how much you beat or trailed your personal par that day. Dots above the zero line are under par (good); below is over par.
Your mean delta vs. Greg-par on dailies that needed each kind of reasoning. Easy patterns are 1-1, 1-2, and 1-1-1 shapes you spot on sight. Complex patterns are larger overlap deductions where you have to count out the cells. Process of elimination is when patterns don’t apply and you walk through possibilities to find which cells are forced. Boards typically need more than one type of reasoning, so the bars overlap. Sort is by signed effect: reasoning you handle better than expected on the left, reasoning that costs you extra on the right.
Days where you hit at least one mine. Going down on the chart means your flagging discipline is improving.
Your mean delta vs. Greg-par on dailies where each modifier is in play. Greg-par already accounts for each modifier’s baseline cost, so a positive bar means that modifier trips you up more than the model’s built-in adjustment, not just that it’s harder. Sort is by signed effect: modifiers you handle better than expected on the left, modifiers that cost you extra on the right.
Smoothed density of your daily deltas (Gaussian kernel, Silverman bandwidth). The gray dashed line is par; the orange dashed line is the mean of these deltas. The mean isn’t quite the same as your handicap, because handicap is fit only on bomb-free plays, while this line includes every daily you’ve submitted. Mass left of par is under par; right is over.
Where you finished among all players that day. 100th percentile = you won that day; 0th = slowest of the field. With a small field (2 players) the axis is bimodal: you either won (100) or lost (0).
Same puzzle each day. Watch yourself learn it. Your best time across the 7 attempts is what counts on the leaderboard.
Best times from the last few weekly puzzles.
| Week | Best | Played | Rank |
|---|
--
Highest level reached. Checkpoints every 5 levels keep it.
Every board you have finished in a Challenge run.
Most recent successful refit. RMSE is the typical residual between actual completion times and the model’s clean-play predicted par; bias should sit near zero after the intercept correction. Target is the feature tomorrow’s daily will be biased to maximize.
--
RMSE (solid line) is the typical residual between actual completion times and the model’s clean-play prediction; lower over time means the model matches reality better. Bias (dashed line) is the mean residual, and it should hover near zero after each refit’s intercept correction.
Total scores included in each refit. More N = posteriors tighten faster and CVs fall, which is what the experimental design is for.
Same data as the charts above, in raw numbers. Method column: brms = full Bayesian fit, seed = fallback (no fit, residuals against the existing coefficients).
--
Posterior coefficient of variation for the candidate features. The next experiment target is the highest CV not in the last-3 list. CVs near 1.0 mean the data hasn’t pushed the prior much; CVs trending down across refits means the experiment is paying off.
--