Batch stability

A one-off script (not kept in the repo), jev-1.13.0, seed 20260918. The 8 USSD files the benchmark scans; 20 (file, question) pairs: 10 drawn from pairs whose normal-batch p is 0.15-0.85, 10 from the rest (broad questions excluded). Each is asked (a) alone, (b) in the batch jevscan sends (same request, so the same cache entry as a scan), (c) with 10 random other questions from any level, at a random position, and (d) alone again, sent past the cache. Jev is not deterministic: the same request sent twice can differ, so (a) vs (d) is the floor the batch effects are compared with. (d) was sent fresh. Size: 81 cached or uncached requests plus 20 repeats, 990,496 input tokens through the cache ($0.0416 uncached).

Comparison	Mean abs delta (all 20)	Max (all 20)	Mean (10 uncertain)	Max (10 uncertain)
alone vs the same request repeated (repeat noise)	0.009	0.030	0.016	0.030
alone vs normal batch	0.007	0.030	0.013	0.030
random batch vs normal batch	0.008	0.030	0.015	0.030
alone vs random batch	0.011	0.060	0.020	0.060

Largest mean abs batch delta: 0.011. That is below the 0.03 bar set before the run, so batching was left as is: question order is already fixed (library order), and only the batch boundaries move with file size. The batch effects are about the size of the repeat noise, so fixing the batches would not make scores reproducible. Single moves of 0.05-0.08 do occur, so a p within about 0.05 of a threshold should be read as a tie.

File	Question	(a) alone	(b) normal batch	(c) random batch	(d) alone, repeated
USSDRebalancer.sol	subcategory:arbitrary_external_call	0.180	0.180	0.170	0.170
USSD.sol	subcategory:invariant_violation_solvency	0.670	0.660	0.690	0.680
oracles/UniswapV3StaticOracle.sol	detector:oracle	0.300	0.290	0.290	0.290
USSD.sol	detector:deploy_uninitialized_impl	0.740	0.750	0.760	0.760
oracles/UniswapV3StaticOracle.sol	detector:general_dos	0.300	0.330	0.300	0.310
oracles/StableOracleDAI.sol	subcategory:uninitialized_or_shadowed_state	0.300	0.300	0.330	0.320
USSDRebalancer.sol	subcategory:rounding_direction_error	0.390	0.380	0.380	0.370
oracles/UniswapV3StaticOracle.sol	subcategory:external_protocol_assumption	0.290	0.300	0.300	0.260
USSDRebalancer.sol	subcategory:fee_on_transfer_or_rebasing_token	0.420	0.400	0.390	0.400
USSD.sol	subcategory:manipulable_twap_or_custom_oracle	0.260	0.290	0.320	0.270
oracles/StableOracleWBGL.sol	subcategory:single_function_reentrancy	0.020	0.020	0.020	0.020
USSD.sol	detector:kinetiq_native_receive_handling	0.030	0.020	0.020	0.020
oracles/UniswapV3StaticOracle.sol	detector:kinetiq_instant_unstake_pool	0.010	0.010	0.010	0.010
oracles/UniswapV3StaticOracle.sol	subcategory:liquidation_logic_error	0.020	0.020	0.020	0.020
oracles/StableOracleWBTC.sol	detector:c_field_truncation	0.010	0.010	0.010	0.010
USSD.sol	detector:vyper_augassign_oob	0.010	0.010	0.010	0.010
USSD.sol	detector:diamond_selector_collision	0.030	0.030	0.020	0.020
oracles/StableOracleDAI.sol	detector:compound_patterns	0.010	0.010	0.010	0.010
USSDRebalancer.sol	detector:vyper_range_negative	0.010	0.010	0.010	0.010
Migrations.sol	detector:solana	0.010	0.010	0.010	0.010