Project
Check empty input and equal heights, rejecting negative or Boolean values without changing the input.
AI coding workspace
Maximum histogram rectangle
BarCumulus is a Windows AI coding client for reviewing algorithms that turn bar heights into a maximum area. Open the project folder, discuss stack boundaries with a compatible model and inspect its proposed edits against plateau and empty-input fixtures.
● ● ●BarCumulus●
A concrete starting point · Maximum histogram rectangle ×
def histogram_area(heights):
heights = tuple(heights)
if len(heights) > 256 or any(type(h) is not int or not 0 <= h <= 1000000 for h in heights):
raise ValueError("height.range")
stack, best = [], 0
for i in range(len(heights) + 1):
current = heights[i] if i < len(heights) else 0
while stack and heights[stack[-1]] > current:
height = heights[stack.pop()]
start = stack[-1] + 1 if stack else 0
best = max(best, height * (i - start))
stack.append(i)
return best✳Find the largest rectangle beneath unit-width histogram bars with at most 256 integer heights from zero through 1000000.
✓ ChangesReview before applying ↗
Illustrative workspace
Review before applying
Compare actual behavior with the expected result. If a check fails, investigate it before applying more changes.
Connection
Base URL, credentials and a reachable server.
A concrete starting point
def histogram_area(heights):
heights = tuple(heights)
if len(heights) > 256 or any(type(h) is not int or not 0 <= h <= 1000000 for h in heights):
raise ValueError("height.range")
stack, best = [], 0
for i in range(len(heights) + 1):
current = heights[i] if i < len(heights) else 0
while stack and heights[stack[-1]] > current:
height = heights[stack.pop()]
start = stack[-1] + 1 if stack else 0
best = max(best, height * (i - start))
stack.append(i)
return bestAsk what a refactor must preserve before changing its shape
BarCumulus is a Windows AI coding workspace for reviewing an implementation change without losing the behavior it was meant to preserve. Open the project folder and explain why you want a refactor: clearer state, a smaller helper or an easier path to testing. Ask the model to inspect the existing contract before proposing a replacement.
You can connect an uncensored AI model if it is offered through your supported OpenAI-compatible or Anthropic endpoint. Add credentials when necessary. Its provider restrictions, tool capabilities and context limits remain in force, so use the conversation to develop reviewable reasoning rather than treating a different-looking implementation as an improvement by itself.
A maximum-rectangle calculation over bar heights is one example to bring. Request a review of how the code handles plateaus, empty input and the final pending work. Include small fixtures with manually checked results, then ask the model which parts of a proposed refactor preserve those cases and which change them.
Inspect the resulting diff and run the project checks on the same version you plan to keep. For complicated changes, compare against an independent simple implementation over small inputs. That comparison is evidence within its tested range; it does not turn the coding workspace into an automatic proof system.
Keeping a refactor accountable
- What is a useful reason to request an algorithm refactor?
Name a maintenance problem you can assess, such as unclear state transitions or duplicated handling. Ask for the smallest proposal that addresses it. A replacement should earn its place through readability and verified behavior, not novelty alone.
- Why include equal-height bars in the example fixtures?
Equal values can expose assumptions about when pending state is retained or removed. Include a short plateau with a known expected result. Ask the model to explain that fixture against the proposed implementation before reviewing larger examples.
- Does differential testing prove two implementations equivalent?
It supports equivalence for the tested inputs and can reveal disagreements. Keep the input range and reference implementation visible, and investigate differences. Do not describe finite testing as proof for every possible input accepted by the production function.
- How should I review a proposed comparison script?
Read its input bounds and execution plan before running it. Confirm tool destinations as well: local model inference needs a separate server and does not make every connected execution tool local. Keep test workloads deliberately small.
BarCumulus · Updated 2026-10-25