feat: verify S233 @ShawTim bounty — YES_WITH_CAVEATS, floor blend is identity function (#541)

[xliry]

Issue: #204
Submission: #204 (comment)
Author: @ShawTim

Problem (in our own words)

The 30% floor anti-gaming penalty in the batch scoring engine is architecturally dead code. build_investigation_batches creates exactly one batch per dimension, so score_raw_by_dim always contains a single score per key. min([x]) == x == weighted_mean, making floor_aware = 0.7*x + 0.3*x = x — a pure identity function. The README's claim that scoring "resists gaming" is unsupported by this mechanism.

Evidence

• desloppify/intelligence/review/prepare_batches.py:296-324 — build_investigation_batches docstring: "Build one batch per dimension"; _DIMENSION_FILE_MAPPING maps each dimension to exactly one collector
• desloppify/intelligence/review/prepare_holistic_flow.py:267-280 — batch_concerns merges into existing batch, preserving one-batch-per-dimension invariant
• desloppify/app/commands/review/batch/core.py:535-536 — _accumulate_batch_scores appends one score per dimension
• desloppify/app/commands/review/batch/scoring.py:143-165 — merge_scores: floor = min(score_raw_by_dim.get(key, [weighted_mean])) = min([x]) = x
• desloppify/app/commands/review/batch/scoring.py:113-117 — score_dimension: floor_aware = 0.7 * x + 0.3 * x = x

Fix

No fix needed — the finding documents dead code. A fix would require either (a) producing multiple batches per dimension so the floor has distinct values to blend, or (b) removing the floor mechanism and its associated constants.

Verdict

Question	Answer	Reasoning
Is this poor engineering?	YES	A mechanism explicitly designed to resist gaming evaluates to an identity function in all cases due to the one-batch-per-dimension architecture.
Is this at least somewhat significant?	YES	The floor blend is one of two anti-gaming mechanisms (alongside issue pressure); its complete inertness undermines the scoring engine's anti-gaming claims.

Final verdict: YES_WITH_CAVEATS

Scores

Criterion	Score
Significance	5/10
Originality	4/10
Core Impact	4/10
Overall	4/10

Summary

The floor anti-gaming mechanism (_FLOOR_BLEND_WEIGHT = 0.3) is dead code because the architecture always produces exactly one batch per dimension, making floor == weighted_mean in all cases. The blend formula 0.7*mean + 0.3*floor reduces to 0.7*x + 0.3*x = x. Originality is reduced because this is ShawTim's third submission targeting the same floor mechanism (S116, S184, S233), and the S184 verification already documented the identity-function property as the reason the file-merging exploit doesn't work.

Why Desloppify Missed This

• What should catch: A "dead code" or "unreachable logic" detector that identifies code paths whose output is always identical to their input.
• Why not caught: The floor mechanism is syntactically live code (it executes, computes values, and returns results). Only semantic/mathematical analysis reveals the output is always equal to the input given the architectural constraint of one batch per dimension.
• What could catch: A mathematical identity detector that tracks whether blending formulas can produce distinct results given actual input cardinalities, or a test that verifies the floor penalty actually changes scores in a realistic scenario.

Verdict Files

• Verdict JSON
• Verdict Report

Generated with Lota