UNDER THE HOOD
JEV proposes.
The controller proves.
Our solver turns Picross into a sequence of tightly bounded classification decisions. JEV never receives the puzzle name, finished picture, record ID, training split, ROM data, or a hidden answer. It sees public clues and marks. The server chooses one unknown square, checks every proposed move, and marks the board only after a proof passes.
THE WORDS JEV RECEIVES
The current playing rules.
These are the exact rules used by the live solver, published from the same source as its requests. The full set accompanies each board. Each question also repeats the rules most relevant to its own line.
Rules version: compact-two-line-20260922
Full and empty lines first
FIRST: A single clue equal to the line length fills every square. An empty clue list empties every square. UNKNOWN is an unmarked square, not evidence that its answer is undetermined.
Finish the count
If the line is consistent and its known fills already equal the clue sum, all other squares are OFF. If required fills remaining equal unknown squares, all unknown squares are ON.
Exact fit
If clue lengths plus one mandatory separator between adjacent clues exactly fill the line, the arrangement is fixed. Place the runs in order with exactly one OFF between them.
Overlapping placements
For one clue K in a line of length L, positions L-K+1 through K are ON when that interval is nonempty (positions start at 1). Example: clue [7] in length 10 fills positions 4,5,6,7 even on a blank line. Existing valid marks may force more squares.
One run owns every fill
With exactly one clue, all known ON squares belong to that same run. It must cover them all and cannot cross any OFF. If two ON squares have UNKNOWN between them, those intervening squares are ON. Exclude positions the run cannot reach while covering the known ON squares. Example: [3] with UNKNOWN,ON,UNKNOWN,ON,UNKNOWN has the unique arrangement OFF,ON,ON,ON,OFF.
Empty squares are barriers
Known OFF squares and the line edges bound possible run locations. A run cannot cross an OFF. With one clue, a segment too short for that run cannot contain fills; if only one segment can hold it, the run lies there. Example: [2] with UNKNOWN,OFF,OFF,UNKNOWN,UNKNOWN is OFF,OFF,OFF,ON,ON.
Agreement across legal placements
For multiple clues, keep run order, run lengths, separators and every existing mark. Compare the same indexed run across legal placements, not different runs. A square is ON if every legal placement fills it, OFF if every legal placement excludes it. Multiple legal placements can still agree on this square. If you cannot establish either proof, return UNDETERMINED; this does not declare the puzzle unsolvable.
How each question is scoped
A single-clue line repeats rules 1, 2, 4, 5, and 6. Other lines repeat rules 1, 2, 3, and 7.
The row instruction is “Classify the target in state.focus.row using only that row.” The column instruction is “Classify the target in state.focus.column using only that column.”
The exact answer criteria
- ON
- A visible-clue proof requires the target to be filled in every legal arrangement of this line.
- OFF
- A visible-clue proof requires the target to be empty in every legal arrangement of this line.
- UNDETERMINED
- Both values remain possible in this line, or no proof of a forced value has been established after checking the applicable rules. Leave the square unmarked; this is not a whole-puzzle verdict.
- CONTRADICTION
- No arrangement satisfies this line’s clues and existing marks.
Read the current public clues and player marks.
Select one unknown square with a fixed scheduling heuristic.
Send separate row and column choice questions to JEV.
Enumerate legal line arrangements and check the proposal.
Mark one cell, observe again, and verify the exact change.
THE ACTUAL API QUESTIONS
One square, two independent classifications.
Each request uses Jev 1.13 and contains two choice questions: row_cell and column_cell. The request includes the full public board snapshot for context and auditability, while each question is explicitly scoped to a copied view of only its own line: the ordered clues, every visible mark, the target position, and arithmetic facts such as clue sum, required separators, remaining fills, and unknown count.
The two questions are evaluated independently. The row question may not use the column answer, and the column question may not use the row answer. JEV returns a selected label, a probability for every label, and a confidence value. The controller rejects malformed responses, probabilities that do not sum to approximately one, a selected label that is not a highest-probability label, or the wrong model version.
The rule name displayed above the board is derived by controller code from the verified clue proof, not from a JEV rule-label answer.
These are four-way choices, not a raw Boolean API response. Only a verified ON or OFF becomes the Boolean-like fill-or-cross action. An unknown square is never treated as false.
THE INDEPENDENT REFEREE
Confidence is useful information. It is not proof.
For each target line, the controller generates every possible placement of its clue runs, including the mandatory empty separator between runs. It then removes any placement that conflicts with an existing filled square or cross.
If all remaining placements fill the target, the deterministic verdict is ON. If all exclude it, the verdict is OFF. A mixture is UNDETERMINED; zero legal placements is CONTRADICTION. A JEV proposal is accepted only when its ON or OFF label exactly matches the corresponding line verdict.
There are three legal placements. Cell 3 is filled in all three, so the row proves ON. The column may still answer UNDETERMINED; one independently verified axis is enough. If two verified axes ever demanded opposite values, the run would stop as contradictory.
HOW THE NEXT CELL IS CHOSEN
A deterministic scheduler points JEV at promising work.
The opening pass is strict: a line with one clue equal to its entire length is handled first because every square in that line must be filled. JEV still has to propose each mark, and the referee still has to approve it.
After that pass, the controller scores unknown cells from visible structure. It favors lines with no slack, empty clue lists, or all required fills already placed; then it considers clue density, the amount of known information, and a small center bias. This score chooses where to ask. It never decides whether the square is filled.
When neither answer produces a verified mark, the decision becomes HOLD. That exact target and board signature is skipped for the rest of the current pass. A later verified move in that row or column changes its signature, allowing the cell to be reconsidered with new evidence.
ONE SAFE ACTION AT A TIME
The proposal, proof, and game action stay separate.
The board must still match the saved observation when the answer arrives. The controller submits one revision-checked mark action to the server game and verifies that exactly one cell changed. A pause invalidates in-flight responses, so a late answer cannot mark the board.
The collection requires every square to end as either filled or crossed. Completion means the filled groups match every row and column clue and every cell has an explicit mark. A puzzle also stops without guessing when no further model proposal passes verification or when it reaches its per-puzzle ceiling of the larger of 120 requests or six requests per cell.
READING THE LIVE PANEL
Which parts come from JEV?
- JEV · ROW / COLUMN
- The model’s selected labels and returned probabilities.
- VERIFIED RULE
- A controller-written explanation of the independent clue proof.
- TARGET SOLUTION
- A spectator-only asset. It is never placed in the model request.
- DEFERRED OR STALLED
- The system abstained because no proposed mark passed proof.