A significant update has been deployed for Slay the Spire 2, introducing long-awaited Steam Workshop support for mods. Alongside this, players can now access a comprehensive monster bestiary within the Compendium. The patch also replaces the frequently criticized Doormaker boss with a new antagonist, Aeonglass. While these are substantial additions, the most compelling aspect of this update addresses a peculiar mathematical anomaly impacting the game’s random number generation (RNG).
Player’s Eight-Hour Investigation Sparks RNG Correction
A number of players have expressed frustration with Neow’s Bones, an in-game relic that grants two additional relics upon selection, accompanied by a random curse. Certain curses, particularly ‘Debt,’ which deducts 10 gold each turn a player holds the card, have been perceived as disproportionately frequent. One player, known as tckmn, dedicated eight hours to compiling a video to substantiate their claim, estimating that the chance of receiving the ‘Debt’ curse when picking up Neow’s Bones was as high as 54%.
Following this extensive player-driven analysis, the development team at Mega Crit has re-examined their RNG systems. Their investigation has confirmed the player’s findings.
Developer Explains RNG Flaw
Ed Lu, a programmer at Mega Crit, detailed the issue in the patch notes. “Most games, including Slay the Spire 2, utilize pseudo-random number generators (PRNGs),” Lu explained. “A PRNG is initialized with a seed value, and with the same seed, it will consistently produce the same sequence of random numbers.”
Lu elaborated on the game’s internal workings: “Each playthrough is assigned a primary ‘run seed,’ which is a sequence of twelve alphanumeric characters. However, a single run employs multiple PRNGs that independently govern elements such as deck draws, combat rewards, event outcomes, and other random occurrences. Each of these PRNGs derives its seed from the main run seed.”
The team’s initial approach, Lu stated, was based on the assumption that by feeding different seeds to each PRNG, their outputs would remain unrelated. “We were mistaken,” he admitted. “It has come to light that our strategy inadvertently allowed players to predict outcomes by understanding seemingly unrelated aspects of the game.”
Lu provided visual aids, including graphs, to illustrate the change. The previous implementation, characterized by parallel diagonal lines in one graph, has been replaced by a new version resembling white noise, which the developer described as “exhibiting no human-detectable correlation.” Beyond the immediate balance implications, Lu noted that “once knowledge of this exploit becomes widespread, players are naturally incentivized to leverage it for optimal play. We aim to avoid encouraging rote memorization of correlation tables, as this is both tedious and unenjoyable.”
The complete patch notes are available on Steam.
