Discrimination per ATP: The Information Budget of T-Cell Proofreading

A T cell must respond to a handful of foreign peptide-MHC complexes among roughly 1e5 self complexes. This paper prices kinetic proofreading in full - discrimination, energy, information, and time - using the standard McKeithan cascade with published inputs (dwell times from Aleksic et al. 2010, Govern et al. 2010 and Huppa et al. 2010; step rates and counts from Altan-Bonnet & Germain 2005 and Feinerman et al. 2008; cellular ATP free energy 20 k_B T from Milo & Phillips 2015; foreign fraction 1e-5 from Irvine et al. 2002 and Purbhoo et al. 2004). At headline parameters (foreign dwell 10 s, self dwell 1 s, step rate 0.5/s, N = 4 steps, 310 K) the discrimination factor is 39.06 against a single-step value of 2.50 and an ideal slow-step limit of 1e4, which is reachable only when the foreign completion probability collapses to 0.0123 and the decision time reaches 80 s. A completed cascade carries 5.29 bits of log-likelihood evidence and costs 6.44 ATP per completed foreign complex. At the physiological foreign fraction of 1e-5, however, a completed cascade leaves the complex 99.96% likely to be self, with 2560 false completions per true one: the cascade alone does not classify, and its information per encounter is 2.07e-5 bits at 6.13e-15 J per bit, 2.07e6 times the Landauer floor k_B T ln 2 at 310 K. In a balanced reference accounting the cascade delivers 0.2594 bits per encounter at 164.8 times the Landauer floor. Proofreading is a cheap evidence amplifier - a full scan of 1e5 self complexes costs about 1.48e5 ATP - whose binding constraints are sensitivity and time, not energy, and whose output requires downstream integration to become a decision. Model calculations are labelled SIMULATED, NOT MEASURED. The deposit includes the paper PDF, the computation script (compute.py) that produces every number in the paper, and its outputs (results.json, results.txt).