The tuning pass, and three numbers that were lying
Capping hero aggro let a wave reach the granary for the first time, and that one change pulled three long-standing numbers out into the light. Each of them had been quietly calibrated for a game that no longer existed.
The hero had 120 hit points. Six collectors put out seventy-five damage a second between them, so he died in under two seconds no matter how well he was played. Nobody had noticed, because dying used to be instant and free — he blinked back at full health and carried on. The moment death cost something, survivability became a real stat for the first time, and 120 was revealed as nothing at all. He now has 520, which is roughly what clearing a knot of six actually costs.
Healing only happened by dying. There was no other way to recover, which made an escalating death penalty a tax on the single healing mechanic in the game. The hero now recovers during prep: you walk out of a wave hurt, and the village patches you up before the next one. Healing during a wave is what the Steward is for.
The granary had 260 hit points. That was set when the hero soaked entire waves and almost nothing reached it. With the wave getting through, nine collectors deleted the whole thing in about two seconds — not a defeat anyone could see coming, let alone answer. At 2400 the same leak costs about a third of it and takes twenty seconds, which is long enough to notice, turn round, and do something. Repairs went up to match, priced so that a bad wave costs a real slice of a day’s milling to patch up, and so that no single builder can ever out-heal a wave that is standing on the granary. You repair between waves. You do not repair your way through one.
And a death spiral that had been hiding as a difficulty result. Villagers taking cover ran to the nearest house — but the village centre sits about seven hundred units from the granary, which is inside the panic radius. So a single collector chewing on the granary kept every villager cowering in a shelter that was itself in the danger zone. They never went back to work. The mill stopped, the planks ran out, the builder had nothing left to repair with, and the granary bled out over four minutes while a full forest stood untouched and one collector took its time. In the logs this looked exactly like the game being too hard: defeat on day two. Cover is now the house furthest from the threat, and if nowhere indoors is genuinely out of range they put ground between themselves and it instead. A safe spot inside the danger radius is not a shelter, it is a deadlock.
There was also a lesson about the test harness, which is now three for three on misleading us. Its autopilot charges the nearest collector until it dies. That was harmless when dying was free, and became actively deceptive the moment it wasn’t: the harness stopped measuring “how hard is this game” and started measuring “how hard is this game with no hero for half of every wave”. It now breaks off and falls back when badly hurt. Any harness behaviour that no human would exhibit invalidates every number it produces.
The curve that came out the other side, measured with the autopilot, as a percentage of the granary remaining at its worst moment each day:
| Day | 1 | 2–3 | 4–8 | 9–10 | 11–12 | 13 | 14 | 15 |
|---|---|---|---|---|---|---|---|---|
| Granary low | 99% | 61% | 93–100% | 70% | 75–86% | 50% | 37% | defeat |
A gentle first day, then an early bite on days two and three while you are still learning what the job board is for. A competent middle once you have. The boss on day ten leaves a mark. Chapter two breathes, and then the last three days are a decline you can see coming and cannot quite stop. Peaks and valleys, a boss that costs you something, and an ending — which is the shape the chapter structure was built to produce, and the first honestly measured curve this project has had.
ChapterHealthStep is the difficulty dial: 0.45 ends the autopilot on day 15, 0.55 on day 8. A human should do better than the autopilot at both.
Prep length is a balance knob, not a comfort setting. Going from 45 to 60 seconds hands the village a third more production every day, and put the whole curve straight back into surplus — days twelve to seventeen at full granary. It cost a complete re-tune. Any future change to prep needs the same treatment.