The daily session is the product. Not the drill library, not the critique: the thing that decides what you do today, in what order, for how long, and why. So the honest way to audit it is not to read the code but to run it. We built a harness that plays a simulated writer through the real engine one day at a time, records every drill the way the app would, applies the same promotion rules, and moves the clock forward. Then we ran a perfect student, 85 percent accuracy, songs finished, belt tests taken, for ninety days.
The session engine prescribed a warm-up, a drill and a reflection every day. In ninety days it never once asked for a lyric. The writing step existed in the code; it was the first thing the time budget trimmed, and on the default fifteen-minute session the budget was always tight. A songwriting school that never asks for a line is a quiz app. Nobody had noticed because nobody had run it for ninety days.
It got worse. The rest-day cadence, one day off in seven, fired once and then never again, because it was inferred from a drill history that gets pruned after two hundred entries and the inference broke around day forty. A single draft the simulated writer left untouched for a week turned every following day into a "Song Day," a special session with two long freewrites, for the remaining eighty days, because nothing recorded that the Song Day had been taken. And the perfect student was still a white belt at day ninety: leaving white requires a graded structure attempt, the only structure drill at white ran ten minutes, and the skill slot on a short session was eight.
Every one of those is fixed, and the fix is the same shape each time: stop inferring a schedule from state the schedule itself never updates. The writing step is now non-trimmable; when a full ten-minute prompt does not fit beside the drill, the same prompt runs as a five-minute sprint. Rest days and Song Days each write a ledger when they complete, so a rest earned is a rest recorded and a stale draft earns one Song Day, then waits a week. A five-minute Section ID sprint exists at white so the gate is meetable on every session length. The default length is the full thirty-minute arc.
The part we are proudest of is not the fixes. It is that the ninety-day run is now a test in the repository. Every change to the engine has to keep a simulated writer writing on eighty percent of training days, out of white belt by day fourteen, resting at least every ten days, and never on a Song Day two days running. When we broke the engine again a week later (we did, tuning the trim order), the test said so before anyone used it.
The sample plan on our landing page is generated by the same engine on a fixed seed, thirty days for a green belt, and a test binds the calendar you see to the engine's output. It is not a marketing calendar. If the engine changes, the page changes, or the build fails.
The drill on this page is the one the simulated writer did most often: a title sprint, ninety seconds, as many titles as you can. It was in every week of the ninety days. It is a good warm-up. It was never meant to be the whole session.
