Devlog
Notes from building the thing. Mostly about balance, and mostly about being wrong about balance.
Entry one
Simulating the grind instead of guessing at it
The first working version had four jobs, a housing ladder and combat, and the numbers were invented out of nothing. Rather than play it for six hours to find out whether the curve was any good, the whole game was extracted into a headless script and run at thousands of times speed with a crude bot playing it — clicking three times a second, buying whatever it could afford, fighting whatever it could survive.
The first run finished the entire game in eight minutes. The second, after a correction, found that combat was unwinnable at the point it unlocked. Every balance decision since has come out of that harness: set a target ("this fight should end with about half your health left"), measure what the player actually produces at that point, and solve for the monster's numbers.
Entry two
Three deadlocks, all the same shape
The simulator kept getting stuck, and every time the cause was identical: something required a material that only dropped from the content that thing was needed to beat.
The tenth dwelling wanted hide, and hide came from combat, which the tenth dwelling unlocked. Tier-four gear wanted essence that only dropped in the area tier-four gear existed to survive. Top-tier gear wanted deed fragments from a zone that required top-tier gear. Each one would have been a hard wall for a real player hours into a save. The rule now is that every zone drops the materials for the gear that beats the next zone, never its own.
Entry three
The economy ate itself
House costs were a plain exponential and income was not. By the midgame, gold had stopped being a constraint entirely — thirty-five of the fifty dwellings were bought within a minute of each other, in bursts, the moment a material gate cleared.
The fix was to stop inventing costs. A dwelling now costs a number of seconds of income for a player who has kept pace at that tier, derived from a model of what such a player produces. Upgrade prices, seed costs, and monster gold all hang off the same function, which means kill rewards can never quietly become pocket change again — a kill is worth thirty seconds of labour at any point in the game.
Entry four
The button that did nothing
Delete Save stopped working, and so had Ascend, and so had swearing to a different god. All three called the browser's native confirmation dialog, and the page was running somewhere that silently ignores those. No error, no console warning — the code read "no dialog" as "the user said no" and quietly did nothing.
Worse, the test suite had been reporting this for weeks as a harmless environment quirk. Every confirmation is now an in-page dialog, and the tests run with the native one deliberately deleted.
Entry five
Clicking is the whole game, so clicking has to be perfect
The interface rebuilt its panel on every click, which meant fast clicking could swap the button out from under your finger and lose the input. Everything that ticks — bars, totals, affordability — is now updated in place, and the panel is only rebuilt when something structural actually changes.
Holding a key down was also worth a hundred clicks a second, which made the click upgrades pointless. Auto-repeat is now blocked three ways.
Entry six
Everything is a small game now
The recent passes added things that don't fit the spreadsheet: three dragon pens that are each a self-contained clicker with their own fuel, a gambler's hall with real blackjack and a nine-tier poker pay table, athletics betting with odds derived from form, and a little house at the end of the market row that opens once a real day for anyone who knows what to say to it.
None of it is required. All of it feeds back into the same loop: get a better house.