jnu / crucible Goto Github PK

Validate all components of the grid are connected.

Import/export puzzles in .puz format

Add an icon to show whether the grid is valid. Can be placed with the other icons (like the grid lock icon).

Implement a new storage client that persists over the network.

Currently the only symmetry option is 180deg rotational symmetry or no symmetry at all. Mirror symmetry should also be possible, though it's much less common.

Integrate with something like Firebase to allow users to have their own accounts. Relates to #13

• Tab should move to next word
• Backspace should keep backing up through next word

UI kind of breaks down with large grids -- support this case better

Allow mouse interactions on the WordWizard component:

• Click on a word to fill it into the grid at the current cursor position
• Context menu to disable a word from being used in the list (see #6)
• Context menu to enter a new word into the list (see #6)

Can generate reasonable clue suggestions with an LLM call.

Suggestions can probably all be batched into a single prompt.

The user could also add a puzzle description to suggest the LLM a certain vibe/theme for the clues.

Some way to load custom lists.

Grid print view should format according to the NYT submission guidelines.

Rebus squares are currently not supported and should be. They should add a visual indicator (usually a circle) to the UI to mark their presence.

Grid "lock" icon should be clickable to lock/unlock the grid.

Also, add a title to explain what this means.

Autofill can be made N times faster by delegating to a pool of workers.

Word lists should support metadata, including:

• Score (included in some major wordlists)
• Whether word has appeared in a puzzle
• Clues used for this word in past puzzles

The first two of those can be used to weight words during auto-fill / suggestion.

In the implementation, there's not currently a way to freeze any metadata in the encoded DAWG from tiny-trie. So metadata would have to be stored and queried separately, unless tiny-trie was extended to store values (which wouldn't be that hard, but definitely a bit of work).

Allow user to add / remove words from the word lists. While the custom patches can't be incorporated into the underlying DAWG, they can be patched onto the wordlist in the caching layer that sits above the DAWG itself.

The customizations can be saved in the storage layer (which currently is just LocalStorage via the BrowserStorage client).

Currently validation fails at certain unchecked crosses in the grid. Need to catch these and mark them as invalid.

Add a simple solvability indicator to the UI that shows the results of a smoke-test to determine whether the grid is likely solvable.

The smoke test can be performed by running a timeboxed auto-fill search. If the auto-fill runs for longer than, say, 5 seconds, we can assume there is probably an answer out there somewhere. The search usually returns relatively quickly in cases that aren't solvable that a human might not be aware of (such as with sparsely filled crossings, such as __p____j____j -- while the user is plunking things down in the cross, this word has become unsatisfiable, and the computer can detect that very quickly).

This could replace/enhance the current suggestion analysis, which recurses into crossings one level deep to try to catch unsolvable patterns. The smoke-test is a more robust approach, but doesn't lend itself to visualization as easily.

The UI can show a spinner while the smoke-test is ongoing, then a check or x in the result when finished. This can be placed alongside the "gridlock" icon.

UI feature / actions / reducers to toggle smoke test on and off

Follow up from #21

The included free wordlist (broda) has a lot of shortcomings. Could probably generate a reasonable wordlist with an LLM. The results might have less crosswordese / nonsensical acronyms. (Then again, they might not, and might have other problems, like reddit usernames, etc.)