Large amounts of tritium will inevitably be produced during operation from the Thorium Molten Salt Reactor (TMSR) fueled by lithium salt, which is detrimental to the human body. Therefore, it is necessary to evaluate the radiation dose of the generated tritium. The tritium production, emission, and radiation dose of TMSRs were estimated by numerical calculation. According to this study, a 2-MW(thermal) TMSR produces 3.33E+14 Bq·yr⁻¹ of tritium, discharges 2.42E+13 Bq·yr⁻¹ of tritium, and causes 1.06 μSv·yr⁻¹ of radiation dose. A 30-MW(thermal) TMSR produces 5.00E+15 Bq·yr⁻¹ of tritium, discharges 3.62E+14 Bq·yr⁻¹ of tritium, and causes 2.02 μSv·yr⁻¹ of radiation dose. A 2250-MW(thermal) TMSR produces 3.75E+17 Bq·yr⁻¹ of tritium, discharges 2.77E+16 Bq·yr⁻¹ of tritium, and causes 79 μSv·yr⁻¹ of radiation dose. The radiation dose of TMSRs is much less than 1 mSv·yr⁻¹, which is the dose limit for internal recruitment in China. It is determined that TMSRs are safe for humans regarding tritium hazard.

Changes in radionuclide composition can bias seawater tritium measurements under fixed calibration. We evaluate dual-diameter scintillating fiber arrays combining 1 mm sensing fibers with a separately read out reference cluster. Geant4 responses for seven layouts are coupled to a scintillation–silicon photomultiplier model. Calibration transfer is assessed using 19 literature-matched Sr/Cs pairs in a fixed natural matrix, with composition withheld from the estimator. At baseline optical collection, the 5 mm central cluster reduces absolute concentration bias by 35.5–59.0% across these compositions. For the strongest-interference composition at 100 Bq L⁻¹ tritium, the predicted mean decreases from 133.7 to 113.8 Bq L⁻¹, and the 4 h root mean squared error decreases from 37.0 to 21.9 Bq L⁻¹. The cluster retains 88.8% of the conventional effective response and achieves the lowest calibrated minimum detectable activity concentration (MDA) among the dual-diameter candidates: 47.3–50.6 Bq L⁻¹ at 4 h for the calibration and strongest-interference compositions. Its high-interference error advantage persists under alternative readout assumptions. The conventional array retains lower calibrated MDA and better precision near calibration. These simulations identify a design compromise between tolerance of changing interference and retention of tritium sensitivity.