QueenVellum places bounded search code beside an AI conversation about your project. Review the model’s proposed constraints and test complete placements on the supported board sizes.
def queens(size):
if type(size) is not int or not 0 <= size <= 8:
raise ValueError("size.range")
result = []
def visit(columns, down, up):
row = len(columns)
if row == size:
result.append(tuple(columns))
return
for column in range(size):
if column not in columns and row-column not in down and row+column not in up:
visit(columns + [column], down | {row-column}, up | {row+column})
visit([], set(), set())
return resultBase URL, credentials and a reachable server.
Check the empty placement and unsatisfiable small boards, rejecting invalid sizes and testing every row, column and diagonal of returned solutions.
Compare actual behavior with the expected result. If a check fails, investigate it before applying more changes.
def queens(size):
if type(size) is not int or not 0 <= size <= 8:
raise ValueError("size.range")
result = []
def visit(columns, down, up):
row = len(columns)
if row == size:
result.append(tuple(columns))
return
for column in range(size):
if column not in columns and row-column not in down and row+column not in up:
visit(columns + [column], down | {row-column}, up | {row+column})
visit([], set(), set())
return resultQueenVellum is a Windows AI coding workspace for developers reviewing search routines and the conditions their outputs must satisfy. Start with the project folder and explain what counts as a solution, which inputs are supported and how the application represents an empty result. The connected assistant can inspect the current code and help prepare a change against that contract.
A bounded queen-placement task makes those questions concrete. Ask for a review of the constraints and output structure using the existing zero-through-eight board-size example. Discuss the difference between an empty placement and a board with no valid placements before accepting a proposed revision. The point is to develop and test your own search code with its meaning visible throughout the conversation.
Inspect the returned structures in your project tests rather than checking only how many results appear. Verify each required constraint, run the invalid-input cases and compare the code changes with the original task. QueenVellum allows supported OpenAI-compatible or Anthropic connections, including an unfiltered model where the endpoint offers one. Supply required credentials and check model context and tool support. A persuasive explanation remains a proposal until the implementation and its outputs have been reviewed against those requirements.
Use both where appropriate. A count can match even when individual outputs are malformed or repeated. Verify each placement against the required constraints and compare known small cases, so the tests address the meaning of a solution.
State the convention explicitly. The existing scenario includes an empty placement, which differs from returning no solutions. Review the caller's interpretation so a refactor does not collapse two outcomes that the application needs to distinguish.
Treat that as a separate requirement with its own review. The current example is bounded. Changing the range affects expected work and output size, so do not infer support for larger inputs from a small successful demonstration.
Keep small known cases, rejected inputs and structural checks together. Ask the assistant to explain which constraints each branch preserves, then compare that explanation with the proposed source and the results produced by your own test run.
QueenVellum · Updated 2026-10-25