Search papers, labs, and topics across Lattice.
This paper introduces HindsightBench, a novel black-box behavioral audit protocol designed to assess parametric hindsight in time-indexed decision tasks performed by large language models (LLMs). The method employs a combination of date-manipulation matrices and dual memory probes to evaluate model behavior without requiring access to training data or backtesting, allowing for a cost-effective and efficient audit process. Key findings reveal that model behavior is influenced by training generation rather than scale, with significant implications for the validity of calendar-window placebo designs and the stability of audit results across different serving configurations.
Models exhibit a date-trigger reflex that is more indicative of training generation than size, challenging assumptions about LLM behavior consistency across versions.
Large language models leak parametric knowledge of realized outcomes into historical financial decision tasks. Existence is settled; what users lack is a cheap way to audit a given model for it. We present HindsightBench, a black-box behavioral audit protocol that profiles parametric hindsight in any time-indexed LLM decision task at probe-level cost (no backtests, no logprobs, no corpus access). The protocol chains a four-arm date-manipulation matrix (revealed/date-only/masked/transplanted), dual memory probes (date recovery; outcome recall), and six per-model metrics -- trigger strength, transplant effect, post-cutoff placebo, recoverability, behaviorally effective knowledge cutoff, and a recall-accuracy dissociation coefficient -- with explicit gates where identifiability is data-dependent. Applying it to 15 models from seven vendors on a 258-node vintage-correct macro panel yields three headline patterns: (i) the date-trigger reflex tracks training generation, not scale -- absent across the 2024 open-weight generation from 1B to 70B, present in every tested 2026-generation model, and switching on within one vendor lineage (Qwen3 ->Qwen3.6) at fixed MoE architecture and 3B active parameters; (ii) effective cutoffs span 22 months across vendors and precede vendor-reported dates by up to eight months, invalidating calendar-window placebo designs; (iii) audit results are not invariant to serving -- BF16 serving of an FP8-referenced model breaks the trigger estimate's stability while AWQ-INT4 preserves it, and a provider-locked reasoning regime makes one probe non-convergent -- so the protocol ships with operational requirements (pin quantization and thinking regime; disclose parser and sampling policy). We release the panel, frozen preregistrations, per-model audit rows with measured dollar costs, transcripts, and one-command regeneration.
