The checks that ran entirely in winter: the right question, asked in the wrong month
Five plant checks were red, and the word on all of them was drift: something in the world had moved out from under the numbers we had recorded. Nothing had moved. Every one of those scenarios started its clock on the same winter day and ran for about a day of world time, which is the season and the span in which a plant does nothing at all.
Where the clock starts
A year in The Long Watch runs a hundred and ninety-two days, and a test world takes the point it starts from out of its seed. For the seed these scenarios use, that point is day 184. Deep winter, with something like four thousand ticks of winter still ahead of it.
A season lasts roughly twenty-four thousand ecology ticks. The windows these checks watch through are sixty, three hundred and six hundred ticks long, and the widest of them covers a little over one in-game day. There is no arrangement of those three numbers that reaches spring.
Winter is when the seasonal gates shut. No plant releases seed, no seed sprouts, a canopy tree’s authored growth for the season is exactly zero, and cold ground suppresses the microbial activity that turns fallen litter back into soil. So each scenario opened in winter, ran for a day, closed in winter, and reported that the mechanic it was named for had not fired once. Every one of those reports was accurate.
What turned them red were guards we had put in the week before, which fail a plant check outright when nothing in its scenario is alive to be measured. The guards did their job. The word we wrote next to the result did not.
Three failures wearing one word
Red checks accumulate a label the way anything on a list does, and once a set of them is carried forward together as drift, the set stops being examined and the label starts doing the thinking. Pulling these five apart took an afternoon and found three distinct failures, closed on three distinct dates, filed under one word.
Most of them were the winter problem. One was not: its recorded fingerprint had simply never been refreshed after the growth repair nine days earlier, which had changed how fast every plant in the world matures. And one was a duplicate. The same recorded value existed in two places, one of them a helper that nothing ever runs, so when the live copy was updated the second sat there disagreeing with it for nine days with no check in a position to notice.
A number copied out of the place it came from is a cache with nothing able to invalidate it. Both copies were written by people who were right at the time, which is exactly what makes the pattern durable.
A season is identified by measuring it
Rather than argue about which season each scenario wanted, we built a throwaway instrument that replayed every one of them at each of the four seasons and printed what actually happened.
Its first job was to reproduce the dead winter result we already had. Until it could produce the failure we knew about, nothing else it printed was allowed to mean anything — the same habit as the checks we break on purpose. Once it could, the table it produced chose the targets without any help from our intuitions: decomposition and soil enrichment to spring, seed dispersal and growth to summer.
One scenario technically passed in summer, with its first effect landing two ticks before the run ended. We moved it to spring instead. A margin that thin is a red waiting for a slower day, and a check that only clears by two ticks will be argued with rather than believed. Each relocated scenario now starts twenty-four days into its season, near the middle, where a small shift in anything upstream cannot walk it back out of the season it was placed in.
The dispersal scenario had logged nothing whatsoever from its winter start. From summer it logged seven hundred and forty-seven seedlings.
The arithmetic that had already been written twice
None of this needed inventing. One of the five plant scenes already began at a tick deliberately aligned to a chosen season instead of wherever the seed happened to land. It had been there for a while, it worked, and it had never been carried to the other four scenes or to any of the checks.
Two hand-copies of that derivation already existed, and they already disagreed with each other. That settled the question of whether to type it out a third time. The arithmetic went into one shared piece, and the first thing to adopt it was a scenario that had already been relocated by hand, so we could prove the shared version landed with no change in behaviour before anything else leaned on it.
Only then did we refresh the recorded fingerprints, all of them in a single pass at the end. That order matters more than it looks. The new numbers are trustworthy because decay was producing dead plants again, enrichment was touching soil cells, dispersal was producing seedlings and growth was moving species by species, not because a value was rewritten until the check went quiet.
Green has to be earned by the mechanic. It can never be granted by the number we write down beside it.
No behaviour in the simulation changed across any of it. The world we measured at the end of the day was the same world we had measured at the start; the only thing that moved was which month we were standing in when we asked.
The one that stays red
The canopy seeding check did not go green in any season, and it never will. Oak, pine and hazel release their seed in autumn and will only sprout in spring, and autumn and spring are never adjacent, because winter sits between them. Through a window a little over an in-game day wide, no starting month can hold both of those gates open at once. What that costs the forest, and the seed bank being designed to fix it, we have written up separately.
We measured all four seasons for it anyway, and the live configuration besides. Asserting that no season works is a claim about a whole category, and a category is worth enumerating rather than reasoning about, particularly when the conclusion is that a check must be left failing indefinitely.
The enumeration paid for itself the same hour. A neighbouring plan had written off both of the dispersal checks on the strength of that same ninety-six-day gap between autumn and spring, and the claim was too broad. The ground cover disperses in summer and germinates in spring and summer, so a single summer-aligned start opens both of its gates cleanly. One of the two was recoverable, and the sweeping version of the argument would have left it red for a reason that was true of the other one.
Cold ground that can finally stop
The last thing to land that day came from the same thread. The random wobble we apply to decomposition was added on as a flat amount rather than scaled to the decay rate underneath it, and across twenty-six plant species that wobble runs about ten times the size of the rate it is supposed to perturb. Ground frozen solid still rotted, slowly, forever. Multiplying the wobble through the rate instead of adding it makes a decay rate of zero come out as exactly zero, which gives cold ground the same true halt that winter growth got a week earlier.
So the arc closes with four scenarios awake, one red on purpose, and a defect in the world found while chasing a defect in the way we look at it. Not one of the five checks had been red because the world changed. The world had been doing what it was told, at the bottom of a winter nobody had noticed we were standing in.
