BloomFjord offers a Windows workspace for discussing probabilistic data structures with AI. Inspect suggested bit operations and key validation while keeping the difference between a candidate match and confirmed membership explicit.
from hashlib import sha256
class Bloom:
def __init__(self, width):
if type(width) is not int or not 1 <= width <= 8192:
raise ValueError("width.range")
self.width, self.bits = width, 0
def positions(self, key):
if not isinstance(key, bytes) or len(key) > 1024:
raise ValueError("key.bytes")
digest = sha256(key).digest()
return [int.from_bytes(digest[i:i+4], "big") % self.width
for i in (0, 4, 8, 12)]
def add(self, key):
for position in self.positions(key):
self.bits |= 1 << position
def maybe_contains(self, key):
return all(self.bits & (1 << p) for p in self.positions(key))from hashlib import sha256
class Bloom:
def __init__(self, width):
if type(width) is not int or not 1 <= width <= 8192:
raise ValueError("width.range")
self.width, self.bits = width, 0
def positions(self, key):
if not isinstance(key, bytes) or len(key) > 1024:
raise ValueError("key.bytes")
digest = sha256(key).digest()
return [int.from_bytes(digest[i:i+4], "big") % self.width
for i in (0, 4, 8, 12)]
def add(self, key):
for position in self.positions(key):
self.bits |= 1 << position
def maybe_contains(self, key):
return all(self.bits & (1 << p) for p in self.positions(key))A fast preliminary check can become a bug when a caller treats its answer as final. BloomFjord is a Windows AI coding workspace for reviewing that boundary with a connected model. Bring the data structure and the consuming code into the project conversation, then state what evidence the application needs before it acts.
For a concrete task, ask the assistant to inspect a Bloom membership filter used ahead of an authoritative lookup. Describe what the caller should do after a possible match and after a definite absence. Request a proposal that preserves the distinction in names, branches and tests, rather than relying on a comment beside an ambiguous boolean.
Review the integration with controlled fixtures. Include inserted keys and absent keys, and check that a candidate match still follows the required confirmation path. Inspect validation and configuration assumptions before accepting changes to width or hashing. Run the project’s tests to establish the behavior you need instead of treating the filter as a guarantee of exact membership.
BloomFjord supports OpenAI-compatible or Anthropic endpoint connections. An abliterated AI model is a selectable option where your provider offers it, with context, tool support and restrictions determined by that setup. The workspace lets you discuss and inspect the proposal while the application contract guides your final decision.
Follow the application’s authoritative confirmation path when exact membership matters. Ask the assistant to show that branch in the caller. A probabilistic result should not silently become permission to skip verification or treat a record as confirmed.
No. It checks that fixture, not every possible key. Use tests to verify integration behavior and preserve the documented uncertainty. Avoid changing the API name or caller logic in ways that imply the filter provides exact membership.
Insertions and lookups must represent the same logical key consistently. Include the code that converts application values to bytes, and inspect any normalization or serialization differences before changing the membership implementation itself.
Keep configuration changes separate unless they are necessary to the reported problem. They can alter the filter’s behavior and resource use. Request an explanation and appropriate checks rather than accepting a new width as an incidental improvement.
BloomFjord · Updated 2026-10-25