He counted 100. The phone counted 64. The deadline did not negotiate.

Wesley had been a paralegal at a mid-sized litigation firm in Center City for four years, which meant long stretches at a desk preparing exhibit binders and reviewing discovery, and a body that had quietly absorbed the cost. He was not out of shape in any acute sense. He was 29, his weight was within five pounds of where it had been at graduation, and he could still hold a plank for a respectable amount of time when he tried. What he had lost, without quite tracking it, was midline strength — the specific kind of trunk stability that made standing up from a low couch feel ordinary instead of effortful, and that had once been a baseline rather than a variable. He had tried abs routines off and on since college. The ones that asked for 30 minutes had failed within a week because 30 minutes was the wrong unit for a 6:50am alarm and a 14-hour day during depositions. The ones that asked for nothing had produced nothing. A hundred crunches a day was a number he had seen on a fitness subreddit and dismissed twice before deciding, on a Sunday in early April, to actually configure the contract rather than read about it again.

Days one through five ran cleanly. He did the reps in two sets — 50 in the morning before his shower, 50 after work, before dinner. The pose detector counted both sessions clean on each of those days. He felt the work in his abs by Wednesday in a way he had not felt anything from his core in two years, and the soreness was the kind he liked: ordinary, located, the result of asking a muscle group to do something specific. Day six was a Friday. He had a deposition prep session that ran until 7:30pm and he came home wired and hungry and uninterested in a slow workout. He ate, sat on the couch for twenty minutes, then dragged the phone and the books into the bedroom at 9:48pm with 72 minutes to the deadline. He propped the phone, hit start, and started cranking. Fast. The morning sets had been deliberate — chest up, shoulder blades clearing the floor, hold for a beat at the top, controlled descent. The 9:48pm sets were not those sets. He was bouncing his head and shoulders off the floor in a quick rhythmic chop, eyes on the ceiling, counting out loud under his breath because counting out loud felt like progress. He hit 50 in his head in something close to 90 seconds, took a 30-second break, and went back into it. Another 50, slightly slower because his neck was burning. He sat up at 10:01pm with his own count at 100 and reached for the phone to confirm the green check. The counter showed 64. He stared at it for a moment. Sixty-four. He scrolled the session log. The model had been counting — it had registered reps in the first set with reasonable consistency, then started skipping reps about midway through the second set, then stopped counting entirely in the last fifteen seconds. He understood immediately what had happened and did not want to. The fast bouncing version of a crunch he had been doing — the one where his head and shoulders were barely leaving the floor — had not satisfied the midpoint angle threshold. The model needed his shoulder-hip-knee angle to drop below 100 degrees, which required his shoulder blades to actually clear the floor and his torso to flex meaningfully toward his knees. A head bob did not do that. The geometry was different. He had been doing a movement that looked like a crunch from the outside and was, by the actual definition the contract was scoring against, a partial rep. He got back on the floor. He had 58 minutes to log 36 verified reps. He started slower this time, deliberate, shoulder blades clearing, holding the top for a beat. Rep one: counted. Rep two: counted. By rep eight his abs were already shaking — the fast 100 he had banged out had pre-fatigued the exact muscles he now needed to access cleanly. By rep fifteen the model was flagging him intermittently, the depth there but the angle inconsistent. He hit twenty verified reps at 10:34pm and his form was visibly degrading on the playback — he could see it in the live feed. He pushed to twenty-eight by 10:51pm, took a two-minute rest, and managed three more reps before his abs simply refused to bring his torso forward through the threshold. The counter sat at 31 added reps, total verified for the day at 95. He tried two more reps. Neither counted. The deadline ticked over at 11:00pm with the day's verified count at 95.

Wesley reset the contract Sunday morning with no change to the terms — same 100-rep target, same $5 stake, same 11pm deadline. The change was structural and lived outside the contract. The first 50 reps had to happen before 9am, full stop, regardless of how the day was shaping up to look. Morning Wesley was the version of him who could perform a clean rep without thinking about it, and morning was when the verified count needed its cushion. Evening reps were for the second 50, when fatigue was real and the model's strictness was either a help or a problem depending on whether he had banked depth earlier in the day. He also added one personal rule for any session: if the live counter was lagging behind his own count by more than two reps, he stopped immediately, watched the playback for ten seconds, and adjusted his form before continuing. The rule was not about willpower. It was about closing the feedback gap before it grew large enough to swallow the day. From day seven forward he did not forfeit again across the next four weeks. Some sessions he hit 100 with three or four reps to spare. Some sessions the model gave him exactly 100 and not one more, because his last reps were marginal and he knew it. The $5 from day six stayed forfeited. He did not try to reframe it. He thought of it as the price of finding out that the model was not grading him on effort, and that effort without geometry was a thing he had been doing for years without anyone keeping score.