Answers are grounded in this paper and cite their sources.
PSL (Pluis Structure Language) is built on one absolute invariant: every meaning field is always null. The motivation is adversarial rather than aesthetic -- a system that carries interpretable meaning on the client can be exploited by reading it (a cheat tool reads "this address is health," a toxicity filter reads "this text is an insult"); a system with no meaning on the client has nothing there to read. Meaning is permitted to exist for exactly one place and one moment: inside a server-only Meaning Execution Engine, generated, used, and discarded in three steps. This report closes a design-level gap a self-audit found in that invariant: role assignment for input tokens (pattern_registry.py) previously relied on a hand-built dictionary mapping words to human-understood categories ("happy" -> EMOTION) -- meaning smuggled in at design time even though every runtime field remained correctly null. We replace it with Scheme A, a hash-angle mapping reused unchanged from this project's own Voynich parody paper, and verify (8 checks) that semantically related words now scatter across unrelated structural labels -- direct evidence the categories are no longer meaning-derived. In the course of that audit we also found and fixed a live reference-aliasing bug in api.py that silently emptied every JSON response's payload field, and closed two independent implementation gaps flagged in an earlier audit: a self-referential baseline-contamination bug in world_model.py (7 checks) and a 32-bit hash collision risk in sna.py, widened to 64 bits (7 checks). We report one further finding honestly rather than silently: a magic-square-inspired token-cost balance check for the Existence Validation Kernel was implemented and measured (10 checks) and found to have no practical detection power -- at every tested tolerance, its detection rate tracks its false-positive rate to within about 0.1, and the best detection rate achievable at zero false positives is under 3%. It is intentionally left unwired from the validation gate rather than shipped as a defense it cannot provide.
No takes yet. Share an insight, caveat, or question.
ARIMOTO et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: