Internet of Things (IoT) gateways, edge authentication servers, and legacy systems need protection against offline password guessing after credential-database disclosure. Memory-hard key derivation functions (KDFs) are the preferred defense, but their resource requirements can exceed memory or latency budgets. ARMOR addresses this gap with a schema-preserving verifier-side transform for database-only (DB-only) compromise. Secret per-account chains of injective operations reside in an isolated Chain Registry. Fixed terminal escape encoding preserves distinct pre-hash values under each chain and moves outputs outside printable ASCII. We evaluate NAIVE adversaries using unmodified cracking tools and AWARE adversaries that enumerate chains. For the evaluated pre-hardened three-operation chains, NAIVE recovery is at most 0.10% on non-digit targets. Escape encoding yields zero recovery of digit-only targets under evaluated stock rule-based and non-rule attacks, closing a bypass permitting up to 45.3% recovery. Across 14.05 million eligible RockYou password entries, retaining duplicates, the median conditional guessing number is approximately 52 output-class tests at depth three and 317 at depth five. These values quantify the chain-search subproblem given the correct plaintext under a unit-cost model. A separate historical 14-operation AWARE evaluation matches all transformed targets at full dictionary coverage. Deployment measurements show microsecond-scale bare-metal transforms on three IoT boards, with protected storage dominating the measured TrustZone-M prototype components. Although an isolated high-entropy pepper provides stronger cryptographic protection, ARMOR combines compact per-account chain storage with low transform cost. When resources permit and the KDF accepts the complete transformed output, ARMOR composes as KDF(ARMOR(password)).

翻译：暂无翻译