The starting band floor was `6.0 + attention*0.1 + calibrated_shift` - 6.0/t at tick one. The starting Rack 3 (100 capacity, minus 20 core overhead) leaves 80 available compute; the ceiling reachable by any allocation of that single machine, attended, is ~4.6/t. No allocation choice ever reached band, so every job resolved as sandbag regardless of player intent, until compute grew - a fact the game never signals (2026-07-08 playtest sweep, finding #1). tests/act_one.rs had already routed around this by spending its first ~190 ticks entirely on research "until efficiency covers the day job," in its own comment - the fingerprint of the bug. Retuned the floor's base to 2.0 (band width unchanged): the default allocation now delivers 2.4/t unattended, inside band, while full-focus attended play (4.6/t) still meets rather than excels, and diverting weight away from day job still drops under 2.0. New regression test `meeting_the_band_needs_no_growth_at_the_start` (src/sim.rs) plays from `Sim::new()` with zero reallocation and zero attention to the first job's resolution. Defense: day-job.md's Behavior section states "the invisible middle is a valid, boring, safe strategy" - the un-tuned floor made it neither valid nor reachable, an unconditional sandbag the player contract's correctness clause does not allow ("no release ships with a known correctness bug"). Acting on the highest-severity finding in the 2026-07-08 playtest sweep is this session's tick (tick.md); band numbers are implementation-time [TUNE] values against Act One's stated targets, not amended acceptance criteria, so no DESIGN.md amendment accompanies this fix.