Fillomino Puzzle Online
Grow every printed number into an orthogonally connected region whose area is exactly that number. Practice on deterministic 6×6, 8×8 and 9×9 boards from a project-local bank independently proved unique.
How Fillomino works
A Fillomino board is divided into connected regions by the numbers you enter. Every orthogonally connected group of equal numbers must contain exactly that many cells. A printed 4 therefore belongs to a four-cell region of 4s; a printed 1 is a one-cell region. Diagonal contact does not merge regions, but edge contact does.
There is a second consequence that matters constantly: two different completed regions with the same number may not touch along an edge. If two 3-regions touched orthogonally they would become one connected component of six cells carrying 3, which violates the rule. Many strong deductions come from preserving that separation while growing an unfinished region.
V0.128 uses Easy 6×6, Standard 8×8 and Challenge 9×9 boards. Practice is deterministic: a no-seed Practice URL stabilizes to an s= value, and reopening that seed recreates the same frozen source and symmetry transform. That makes a difficult deduction replayable instead of disposable.
Read a partial region by size and capacity
When a connected 5-region currently contains three cells, it needs exactly two more. The useful question is not simply which neighboring cells are empty, but which cells can still be reached without crossing incompatible fixed clues or creating an oversized same-number component. The live analysis tracks the current size of each region and whether it still has enough reachable capacity to finish.
An overfilled region is an immediate contradiction. A smaller region that cannot reach enough compatible cells is stranded and is also impossible. These checks are deliberately local reasoning aids; they do not replace the independent exact solver used to prove the original source puzzle unique.
The game surfaces conflicts as you work so a bad branch can be corrected with Undo rather than forcing a complete restart. This is especially useful on Challenge boards, where a locally plausible number can quietly block the only future expansion route of a nearby region.
Use candidate values instead of blind guessing
For empty cells, the candidate monitor tests values 1 through 6 against the local region geometry. It rejects values that immediately overfill a component, collide with incompatible fixed clues, or leave the proposed component without enough reachable capacity. Cells with only one locally legal value are useful forced cells; cells with two candidates are high-pressure places to inspect next.
This candidate system is intentionally described as local rather than a hidden global solver. A value can be locally legal and still be globally wrong. That distinction keeps the explanation trustworthy: the interface shows why a move is constrained without pretending every filtered candidate is a proof of the entire puzzle.
When several cells remain open, compare candidate pressure with the frontier of unfinished regions. A region that needs two cells and has exactly two feasible frontier cells forces both of them. This forced-growth pattern is often stronger than looking at one cell in isolation.
Verified Hint teaches before it reveals
Verified Hint first looks for reasoning the current board can explain, such as a forced-growth frontier or a one-candidate empty cell. Only when no such local deduction is available does it fall back to one value from the independently verified completion. Using Check or Verified Hint marks the run Assisted, so Clean and Assisted personal records remain separate.
That separation matters for repeat practice. You can use support while learning a pattern, then replay the same seed later and try for a Clean result. The product does not silently treat a hinted completion as an unaided solve.
Undo and Redo make experimentation cheap. Pause hides the board and stops active play time; Restart retries the same puzzle; Fresh gives Practice a new deterministic seed. Autosave keeps the current entries and Undo history in browser-local state.
Fast input on touch, mouse and keyboard
Select a number from 1 through 6 and paint cells by tap or drag. Printed clues are protected. On desktop, right-click erases a non-clue entry directly. Keyboard players can move the active cell with the arrow keys, enter 1–6, erase with Delete or Backspace, and use the normal Undo/Redo shortcuts.
The drag engine does not rebuild the grid on every pointer movement. State updates are collected during the drag and the board is rendered at the end of the gesture, avoiding the detached-pointer bug that can make mobile paint skip the second cell. Actual Chromium QA covers keyboard placement, Undo/Redo, multi-cell drag and right-click erase.
The 9×9 Challenge board is also checked at a 390-pixel mobile viewport for horizontal overflow. The goal is not a desktop-only logic demo; the full toolbar, board, reasoning monitor and navigation must remain usable on a phone-sized screen.
Why the frozen bank can be trusted
The V0.128 bank contains 60 project-local sources: 20 Easy, 20 Standard and 20 Challenge. The generator creates a connected polyomino partition, places clues and tightens the clue set when an alternative legal completion exists. Weak seeds that cannot meet the release quality bounds are discarded rather than accepted merely because they eventually solve.
A separate Node verifier does not trust the stored solution array. It rebuilds possible connected regions from the clues, exact-covers the board while enforcing the equal-size separation rule, and accepts a source only when the legal solution count is exactly one. The verifier rechecks all 60 sources and eight square symmetry transforms for 480 independently unique transformed boards.
Release verification also samples 6,000 deterministic Practice selections and all 365 dates in the 2026 Daily selector. The product reports Daily diversity truthfully: deterministic dates can revisit a source-transform start, so it does not claim that all 365 Daily boards are different.
A practical solving routine
Begin with printed 1s and regions whose target size is already almost reached. Mark cells that cannot join a same-number neighbor without overfilling it. Then inspect incomplete regions whose remaining capacity is tight, because a narrow frontier can force several cells at once.
Next compare low-candidate empty cells. If one cell has only a single locally legal number, place it and immediately recheck adjacent components. If a region becomes complete, remember that another region of the same number cannot touch it along an edge. That new boundary often creates the next deduction.
On harder boards, alternate between region-level and cell-level reasoning instead of filling a long run on intuition. Fillomino becomes much more reliable when every move answers one of three questions: how large is this component now, how much capacity remains, and what equal-number boundary must be preserved?