Arabic Full-Text Search with Query-Level Authorization in Institutional Document Retrieval: A PostgreSQL Benchmark Study

Abstract—Institutional correspondence systems require Ara
bic search that remains responsive at scale while enforcing
access policies within the query execution path. This paper
presents an applied benchmark study of ICAS1, a Laravel
based institutional correspondence and archiving system backed
by PostgreSQL. The evaluation uses a dedicated synthetic
benchmark database containing approximately 100,000 corre
spondences, 60,000 archives, and 120,000 recipient links. Post
greSQL Arabic full-text search (FTS) using to_tsvector and
plainto_tsquery with Generalized Inverted Index (GIN)
indexing is compared against an ILIKE baseline across three
primary institutional scenarios. Latencies are measured using
EXPLAIN (ANALYZE, BUFFERS) with seven repeated runs.
Results show FTS achieves median latencies of 21–55ms com
pared with 229–248ms for ILIKE, yielding speedups of 4.2×
to 11.7×. An extended evaluation across 10 Arabic institutional
keywords confirms a consistent warm-cache speedup of 4.15×
4.98× for the global administrator scenario and 10.3×–12.7× for
department-scoped roles. Session-level cold-cache measurements
(after DISCARD ALL and reconnect) show the same directional
advantage. Retrieval overlap analysis over 30 keyword–scenario
combinations yields Jaccard = 1.0 for the corpus vocabulary used.
Access-control validation using canary records confirms complete
agreement between expected and observed visibility, including
non-leakage of classified documents to restricted roles. The find
ings support combining Arabic FTS, query-level authorization,
and GIN indexing for efficient and policy-consistent retrieval in
institutional workloads, with the caveat that measurements reflect
single-node conditions on a fixed synthetic corpus, and generaliza
tion to full cold-start or distributed deployments requires further
investigation.