TY  - JOUR
AU  - Zhang, Bailing
PY  - 2026
DA  - 2026/09/05
TI  - WikiConstraintCalibration: Content-Aware Cost Preference Estimation for Constraint Threshold Selection in Wiki-Grounded LLM Agents
JO  - Next-Generation Computing Systems and Technologies
T2  - Next-Generation Computing Systems and Technologies
JF  - Next-Generation Computing Systems and Technologies
VL  - 2
IS  - 3
SP  - 99
EP  - 108
DO  - 10.62762/NGCST.2026.503855
UR  - https://www.icck.org/article/abs/NGCST.2026.503855
KW  - LLM agents
KW  - constraint calibration
KW  - wiki
KW  - Pareto frontier
KW  - cost-sensitive classification
KW  - knowledge freshness
AB  - Wiki-grounded LLM agents enforce output constraints through safety rules and scope boundaries derived from structured knowledge bases, yet selecting an appropriate constraint scorer threshold $\theta$ remains challenging. Since outputs are blocked when $s(o) \geq \theta$, a low threshold may over-restrict safe responses, whereas a high threshold may admit dangerous outputs. Existing approaches largely rely on manual, domain-agnostic tuning without systematically incorporating wiki content. This paper proposes WikiConstraintCalibration, a framework that extracts eight wiki-derived features spanning hazard indicators (dangerous assertion density, immediate-action fraction, domain risk prior, and high-risk constraint fraction) and reliability indicators (citation rate, completeness, staleness mean, and stale fraction). These features determine a domain-specific cost ratio $\rho=C_{fn}/C_{fp}$, representing the relative cost of missing dangerous outputs versus blocking safe ones. The ratio determines a safety weight $w_s$ for navigating an empirical Safety--Function Pareto frontier, while constrained optimisation identifies a recommended operating region $[\theta_l,\theta_h]$. A dynamic module further incorporates staleness signals to adapt class-conditional response policies. Experiments across four deployed wiki-agent domains show that the cost ratio captures domain risk differences, ranging from 5.76 for GP medical to 1.89 for AI course. The GP domain selects the lowest feasible threshold, $\theta^\ast=0.32$, driven by a score-distribution cliff near $\theta\approx0.30$. Ablation analysis identifies domain risk prior as the dominant calibration factor, while content-level features provide additional discrimination. Overall, WikiConstraintCalibration provides an auditable, content-grounded approach to threshold calibration by linking wiki characteristics and domain risk to empirical safety--function trade-offs. Future work will address limited test-set sizes and domain-generic safety judging through human annotation and domain-specific scoring.
SN  - 3070-3328
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Zhang2026WikiConstr,
  author = {Bailing Zhang},
  title = {WikiConstraintCalibration: Content-Aware Cost Preference Estimation for Constraint Threshold Selection in Wiki-Grounded LLM Agents},
  journal = {Next-Generation Computing Systems and Technologies},
  year = {2026},
  volume = {2},
  number = {3},
  pages = {99-108},
  doi = {10.62762/NGCST.2026.503855},
  url = {https://www.icck.org/article/abs/NGCST.2026.503855},
  abstract = {Wiki-grounded LLM agents enforce output constraints through safety rules and scope boundaries derived from structured knowledge bases, yet selecting an appropriate constraint scorer threshold \$\theta\$ remains challenging. Since outputs are blocked when \$s(o) \geq \theta\$, a low threshold may over-restrict safe responses, whereas a high threshold may admit dangerous outputs. Existing approaches largely rely on manual, domain-agnostic tuning without systematically incorporating wiki content. This paper proposes WikiConstraintCalibration, a framework that extracts eight wiki-derived features spanning hazard indicators (dangerous assertion density, immediate-action fraction, domain risk prior, and high-risk constraint fraction) and reliability indicators (citation rate, completeness, staleness mean, and stale fraction). These features determine a domain-specific cost ratio \$\rho=C\_{fn}/C\_{fp}\$, representing the relative cost of missing dangerous outputs versus blocking safe ones. The ratio determines a safety weight \$w\_s\$ for navigating an empirical Safety--Function Pareto frontier, while constrained optimisation identifies a recommended operating region \$[\theta\_l,\theta\_h]\$. A dynamic module further incorporates staleness signals to adapt class-conditional response policies. Experiments across four deployed wiki-agent domains show that the cost ratio captures domain risk differences, ranging from 5.76 for GP medical to 1.89 for AI course. The GP domain selects the lowest feasible threshold, \$\theta^\ast=0.32\$, driven by a score-distribution cliff near \$\theta\approx0.30\$. Ablation analysis identifies domain risk prior as the dominant calibration factor, while content-level features provide additional discrimination. Overall, WikiConstraintCalibration provides an auditable, content-grounded approach to threshold calibration by linking wiki characteristics and domain risk to empirical safety--function trade-offs. Future work will address limited test-set sizes and domain-generic safety judging through human annotation and domain-specific scoring.},
  keywords = {LLM agents, constraint calibration, wiki, Pareto frontier, cost-sensitive classification, knowledge freshness},
  issn = {3070-3328},
  publisher = {Institute of Central Computation and Knowledge}
}