chore(suspect flags): Include filtered flag in output

[aayush-se]

• Performs a filtering step prior to RRF
  • Normalizes KL and Entropy scores with Box Cox transform then takes scores which have a z-score >= threshold
    • Threshold is currently 1.5 but can be adjusted if over/under filtering on RRF
• Exposes if the flag has been filtered out in the JSON response as a boolean such that all flags are still visible on the frontend if required

TODO:

• Update frontend to use these filtered scores when sorting by Heuristic + RRF or RRF

[codecov]

Codecov Report

Attention: Patch coverage is 97.26027% with 4 lines in your changes missing coverage. Please review.

✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/sentry/seer/math.py	92.30%	4 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           master   #95007       +/-   ##
===========================================
+ Coverage   38.17%   87.82%   +49.65%     
===========================================
  Files        9858    10436      +578     
  Lines      556058   604299    +48241     
  Branches    23550    23550               
===========================================
+ Hits       212265   530718   +318453     
+ Misses     343426    73214   -270212     
  Partials      367      367               

[cursor]

Bug: Box-Cox Lambda Calculation Inconsistency

The boxcox_transform function calculates the optimal lambda parameter using the original values via _boxcox_normmax, even when it internally shifts non-positive values to shifted_values for the actual transformation. This leads to an inconsistency where the optimal lambda is determined for a different dataset than what is ultimately transformed, potentially yielding suboptimal results. Furthermore, _boxcox_normmax duplicates the shifting logic, which is inefficient and brittle.

src/sentry/seer/math.py#L109-L129

sentry/src/sentry/seer/math.py

Lines 109 to 129 in aa96455

	shifted_values = values
	if lambda_param is not None:
	if lambda_param == 0.0:
	transformed = [math.log(max(v, 1e-10)) for v in shifted_values]
	else:
	transformed = [
	(pow(max(v, 1e-10), lambda_param) - 1) / lambda_param for v in shifted_values
	]
	return transformed, lambda_param
	optimal_lambda = _boxcox_normmax(values)
	if optimal_lambda == 0.0:
	transformed = [math.log(max(v, 1e-10)) for v in shifted_values]
	else:
	transformed = [
	(pow(max(v, 1e-10), optimal_lambda) - 1) / optimal_lambda for v in shifted_values
	]
	return transformed, optimal_lambda

Fix in Cursor • Fix in Web

---

Was this report helpful? Give feedback by reacting with 👍 or 👎

[trillville]

should this be _boxcox_normmax(shifted_values)?

[ram-senth]

Minor one - looks like you can avoid code duplication by first initializing lambda_param as lambda_param = _boxcox_normmax(values) if lambda_param is not None else lambda_param.