When algorithmic HRM signals backfire: contestability thresholds in gig worker sensemaking

Purpose Platforms are widely advised to make algorithmic HRM rules clearer and more consistent. This study asks why doing so sometimes drives workers away, challenging the assumption that greater signal coherence uniformly improves retention and theorising instead a contestability-dependent threshold beyond which coherence backfires. Design/methodology/approach Two studies share one platform × country × quarter panel: 12 platforms, 27 countries, 16 quarters (2,944 observations; 204 platform–country pairs), integrating 9,842 archival governance documents with human-validated NLP measures from 1,230,672 worker posts. Study 1 uses three staggered court and legislative rulings as natural experiments in contestability; Study 2 applies dynamic panel system GMM to locate the threshold. Findings Coherence has a U-shaped relationship with expressed withdrawal orientation. It initially reduces disengagement discourse, but past a turning point of 0.78 SD it increases withdrawal as workers begin reading consistency as control. Where workers can contest algorithmic decisions, that point moves to 1.12 SD (? = 0.34 SD, p = 0.015). Inferred platform intent carries roughly 35% of the effect, which survives controls for pay, demand and macroeconomic conditions an interpretive, not economic, mechanism. The outcome captures attitudinal precursors to withdrawal, not confirmed exit. Originality/value The study replaces a monotonic account of signal coherence with a threshold-based one, identifies inferred platform intent as the switch that reverses coherence's meaning and recasts contestability as interpretive infrastructure rather than a moderator. It implies that transparency mandates without contestability rights may accelerate the harm they intend to prevent.