Metraipolis › archive › dc_a69eb69fcf46

Dead text or binding clause? Measuring and restoring constraint influence in black-box LLM dialogues — arXiv:2608.12599

Source: arXiv (en)
Original: https://arxiv.org/abs/2608.12599
License (as archived): CC0-1.0
Archived: 2026-10-19T04:17:15.979Z
SHA-256 of the archived text: bc1ff22c40686054e88114d7dd1f6c25ae02a0f761c1227566fe0c572dd818b0
Stable URL of this page: https://metraipolis.com/documento/dc_a69eb69fcf46

How to cite

Dead text or binding clause? Measuring and restoring constraint influence in black-box LLM dialogues — arXiv:2608.12599 — arXiv (en), https://arxiv.org/abs/2608.12599 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_a69eb69fcf46 (Metraipolis archive dc_a69eb69fcf46, sha256 bc1ff22c40686054e88114d7dd1f6c25ae02a0f761c1227566fe0c572dd818b0)

BibTeX

@misc{dca69eb69fcf46,
  title = {Dead text or binding clause? Measuring and restoring constraint influence in black-box LLM dialogues — arXiv:2608.12599},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.12599},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_a69eb69fcf46; sha256 bc1ff22c40686054e88114d7dd1f6c25ae02a0f761c1227566fe0c572dd818b0},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.12599. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Multi-turn dialogues let users revoke constraints as easily as impose them, but revocation does not reliably take effect: models keep enacting withdrawn requirements (occasionally beneath comments asserting their removal), a failure we call \emph{behavioral relapse}, or revocation inertia. No existing instrument measures this influence per clause, predicts it before delivery, or repairs it under matched budgets. \sysname{} closes the three gaps through the model API alone: a contract ledger pairs every constraint with an executable checker, records revocations as tombstones, and compiles the net constraint state ahead of time into a single specification; a sequential ablation probe measures per-clause adherence and incremental behavioral effect; a repair ladder operates under token- and attempt-matched budgets. On \dataname{} (\NTasks{} HumanEval tasks, \NClauses{} verified checkers), relapse at an 8B operating point climbs from \ScaleDelayedMTwo{} to \ScaleDelayedMEight{} as constraint load grows, while stronger models sit at floor. Under matched checkers, model, and budget, ahead-of-time compilation significantly reduces relapse against a no-ledger verifier-retry baseline (\RestoreDiff{}, 95\% CI \RestoreDiffCI{}, $p$ \RestoreDiffP{}); adaptive ladder interventions stacked on top add no detectable gain (95\% confidence excludes gains $\geq$ \LadderExcludedGain{}). The probe predicts relapse before delivery (AUROC \AurocPrimary{}); a one-sentence tombstone note recovers about a third of the compilation effect and survives a placebo control. At \CostDeliveryFactor{} delivery overhead and \CostTotalHedged{} of API compute for every result, revocation failure becomes a measurable, predictable, and repairable property of dialogue state rather than an invisible one.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-14 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2608.12599 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-19T04:13:06.531Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-14?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).