Base-Manager · round four · follows Afterglass
Round three's ten-week plan is built: four prototypes, ending in a season you can play. This round asks what to build next, before and after the playtest the slice exists for. Here is every candidate from the brainstorm with its status, what the builds showed, and the queue.
Play the seasonRead this first
Everything on this page comes from an autopilot (a scripted player, 40 to 200 seeds a result) and scripted runs in Chromium. Nobody has played it yet.
It exists to ask do people want to play a second season? The page asks at the end of a season and saves the answer in the playtest export, but no tester has seen it.
The static line no longer wins: over 1,000 seeds the reacting plan and a doubled line tie on the two-room start. The best play kept nearly everyone on the line's floor, because a room lit below it took a whole tide alone. The Unlit now take any dark way up and gnaw only a light that bars every way. Working the rooms below the line on one side of the keep now costs about 2 points of second seasons on either keep, against 10.6 on the two-room start before: a near-even trade, not a mistake, and not a gain.
The autopilot finishes 95–98% of first seasons, and 90–98% of the second seasons it reaches on either keep, now that the second is a summer. A whole year is harder: 23–45% of keeps come through winter and the Long Night. People will do worse than a script that knows the rules, but by how much is unknown.
Traits make each shade different, but an autopilot that reads them does no better than one blind to them. The rite is meant to be the game's best decision; whether it feels like one needs people.
When a room a Maw broke also lost its day's work, the balanced plan fell from 95% to 67% of first seasons on the four-floor keep. Almost all of that was the Chandlery, broken the night before the new moon.
Red eyes on a dark night all but vanished for anyone who can't see red, and a Creeper in the dark is nothing but its eyes. Each Unlit eye now has a pale glint that stays bright whatever the colour vision. No colour-blind player has tried it.
The brainstorm's nineteen ideas, in its three groups. Sizes were my estimates at the time: S, M or L.
| # | Idea | Status |
|---|---|---|
| Fixes for problems already measured | ||
| 1 | Break the fixed stair setup that won nights 1–5M | BuiltThe static line no longer wins. The Unlit now go around lights, so the rooms below the line can be worked at about even odds. |
| 2 | Make the rite bite: traits such as Devout turning Bitter, and fire by dayM | BuiltTraits, with a Bitter shade costing 3 Dread, and fire by day. Fought at once, fire adds few deaths but costs 5–14 second seasons in 200. |
| 3 | Colour-blind-safe eyes: Creepers and Strangers told apart by shapeS | BuiltThe real problem was red eyes vanishing in the dark; each eye now has a pale glint. |
| Designed in round three, not built yet | ||
| 4 | Traits that invert at death: 8 pairs, each with a night effectM | BuiltReading them made no measurable difference to the autopilot. |
| 5 | Quarters and the Dreamwell; Weepers give the sleepers nightmaresM | BuiltIgnored, the Weepers cost 2–5 first seasons in 200 on the two-room start and nothing measurable over two, but 22 second seasons for the four-floor doubled line. Guarding against them costs more than it saves, and the Quarters doesn't pay. |
| 6 | The black mirror: tomorrow's raid and tonight's Creeper wavesS | BuiltInside the threat preview (16): tonight's waves by rift and tide, and tomorrow's raid as a range. |
| 7 | Whispers and the great glass: shades coach the living, or step through by dayM | BuiltWhispering pays a little, most on a built-up keep; the great glass is out of reach at the slice's glass income. |
| 8 | Break a mirror in an emergency: it frees its shades and curses the roomS | BuiltFrees everyone in it and lifts 1 Dread each, then 7 days of bad luck. Used blindly against censures it's slightly worse than taking them. |
| 9 | Build your own roomsL | BuiltThe keep now starts as two rooms. No choosing a slot, no tearing down. |
| 10 | A full year: shifting days and nights, the Long Night, the eclipseL | BuiltWithout the eclipse. Summer is easy and winter hard: 23–45% of keeps survive a year. Round three's double-length Long Night was all but unwinnable, so it's 1.5 winter nights. |
| 11 | The dusk crossing in the seasonS | Built |
| New ideas | ||
| 12 | The Book of the Dead: each shade's storyS | Built |
| 13 | A season recap card to shareS | BuiltOne image of the season to share or save, at its end; sharing is untested on phones. |
| 14 | A daily seed everyone playsS | BuiltAs Today's keep. Players compare by recap cards; there's no board. |
| 15 | Watch a replay of a season or a tester's exportS–M | ParkedFor later, your call. |
| 16 | A threat preview at dusk: likely Creeper routes and the Hollow's pathM | BuiltAs the black mirror: the Dusk panel says where each rift's tide will go, and the Tain shows the ways. |
| 17 | A guided first seasonM | Built |
| 18 | An installable app, the first step toward Google PlayS–M | BuiltChecked in Chromium only. |
| 19 | Sound and hapticsM | BuiltLevels set by measurement; nobody has listened yet. |
| 20 | A Menu with Settings and save slots: three keeps, export and import as a fileS–M | BuiltChecked in Chromium only. |
My picks for playtest readiness were 1, 11, 12, 17 and 18; all five are built. Building rooms (9), people that read as people, the Menu with save slots (20), and sound and haptics (19) came after, at your request.
The static line is the thread through all of it. The Maws were built to break it; how that went depends on the keep you start with.
| Seasons finished, of 200 seeds | Doubled line | Balanced |
|---|---|---|
| Four-floor keep, second seasons, before the Maws broke rooms | 133 | 116 |
| Four-floor keep, second seasons, Maws break rooms | 116 | 123 |
| Four-floor keep, second seasons, and a better autopilot | 165 | 173 |
| Two-room start, first seasons | 196 | 190 |
| Two-room start, second seasons | 174 | 138 |
| Two-room start, second seasons, the autopilot playing sensibly | 179 | 176 |
| Four-floor keep, second seasons, the autopilot playing sensibly | 171 | 179 |
A second season counts only if the first was finished too. "Doubled line": two fighters on each stair up to the Veil on every Maw night, never moved. "Balanced": reacts to tides and Maws.
A room lit below the line stood alone across the Unlit's way up. A candle lights 40 of a room's 48 steps, and the way up runs between stairs through the middle of each room, so no candle there stays off it. A Creeper cut off by the line gnawed the first light on its shortest way, so one shade below took a whole tide. Rewards didn't help: up to double the night's work a floor down, or a day bonus for rooms worked through the night, left working below the line 13–17 second seasons behind in about 197. Neither did retreating to the line for each tide, or holding the line a floor lower. What helped was where the Unlit gnaw: they now take any dark way up, however long, and gnaw only a light that bars every way. It's a switch in Settings, on by default.
| Second seasons, of 600 seeds | Line only | One side below the line |
|---|---|---|
| Two-room start, before | 564 of 591 | 493 of 581 |
| Two-room start, the Unlit going around lights | 557 of 592 | 546 of 591 |
| Four-floor keep, before | 577 of 599 | 567 of 598 |
| Four-floor keep, the Unlit going around lights | 570 of 600 | 556 of 597 |
The balanced plan throughout. "One side below the line": it also works the rooms below the line on the Choir's side of the keep, leaving the other side dark. Working below the line now costs about 2 points on either keep (1.8 standard errors pooled, probably a small real cost), where it cost 10.6 on the two-room start. Nothing below the line clearly pays except the Choir. The rule may also make the game about a point harder (1 standard error on each keep).
They tore down the candle holding the line. That made the line cost something, but a doubled line beat them outright: 96 first seasons in 100 against 95, and 0.2 shades lost a season against 2.4.
At ×1.35 difficulty a season, the second finished 35% of the time. At ×1.2, with the Maws on their own multiplier, it's 73% for the balanced plan on the two-room start.
They weigh the candle barring the way against every room where people work, count the fighters on the way, and go for the most worth for the least fight. A broken room loses its night's work and costs 1 Dread.
On the four-floor keep that reversed the static line's lead. On the two-room start it didn't at first: the first floor you build is always the line's floor, so a doubled line stands in working rooms.
The reacting plan had been pulling reinforcements across a dark floor for each tide, and posting shades in rooms below the line's floor. Fixing both levelled it with the doubled line on the two-room start and put it ahead on the four-floor keep. The static line's lead was my autopilot's, not the rules'.
What's left is a design problem: on a built-up keep the night is all about the line's floor.
A rule that shades in the light at the stairs do no work but keep the Watch. It took the balanced plan from 92% to 73% of first seasons and barely moved the doubled line, from 98% to 95%. It's off, as a switch in Settings.
Stone from masons in the Yard, 6 stone a room, three workers a room, a Forge and a Cellar, and a keep that starts as the Hearth and the Crypt. The autopilot's results depend heavily on its build order: swapping the Chapel and the Chandlery once took it from 93% to 30%.
Everyone has one of eight; death turns it over. An autopilot that reads traits finished 178 second seasons of 193, one blind to them 175: no measurable difference. On the current build it's 182 of 198 against 185 of 195, still within noise.
I first reported a lead of about 10 under a harsher Dread and made it the default. It came from an autopilot mistake, spending the new moon's essence on the tides of nights 5 and 6, and vanished once fixed; the default went back. The same mistake had put the doubled line ahead by 13–15. With it fixed, Double led by 10 in about 195, and over 1,000 seeds the two tie.
The doubled line, over 1,000 seeds on the two-room start: the balanced plan finished 932 of 982 second seasons (94.9%), Double 938 of 990 (94.7%). A tie; the lead of 10 in 200 was noise.
Raids, back from 75% to 86% held (90% without traits). The Gentle guarded at half strength and the Greedy ate double; now the Gentle grieve harder and the Greedy work ×0.8 away from their two rooms.
A warning for the new moon. On the two dusks before it, the Dusk panel says what warding every stair against the Hollow will cost against the essence in store. The autopilot's mistake is an easy one for a person too.
They work in scripted runs: the crossing animates, the guide's cards fire, the Book tells every death, and the app installs and plays offline in Chromium. None of it has been tried by a person.
The first two stand between the slice and its playtest.
Installing, real fingers, the guide's clarity and iOS are all unchecked. Your notes should decide what gets fixed before testers see it.
Today they copy a block of JSON from Records, then Playtest. That works for three to five people; a form or an email link would lose fewer answers beyond that.
An installed app with no URL bar needs assetlinks.json at rcjlabs.github.io/.well-known/: an rcjlabs.github.io repository or a custom domain, and the signing fingerprint from your Play Console.
Settled since the list was made: the Maws keep their name. They break rooms now, as round three's Maws do.
Round four, next to the other three.
Done: the gap was my autopilot's. Over 1,000 seeds the balanced plan and the doubled line tie on the two-room start.
Done, checked in simulation. Shape already told Creepers and Strangers apart; the real problem was red eyes vanishing in the dark for anyone who can't see red. Each eye now has a pale glint.
Built. They don't yet show a measurable effect on the rite for the autopilot; what they change is which shade costs what. Whether that matters is for the playtest.
Done in part: see what the builds showed.
After you've played it on a phone and picked how results come back.
The candidates not built yet when the queue was made, in the order I'd build them. The first two help the playtest itself; the next two matter once there are players; the rest add to the game.
Done. The Dusk panel's black mirror says where each rift's tide will go past the candles set so far: the light it will gnaw, a mirror left open, or a shade it will catch. It also names where candle hunters go, what a Maw would make for, the Hollow's way, each tide's size, and tomorrow's raid as a range. The Tain shows the same ways as marching chevrons. It asks the Creepers' own planner, so it's exact until the night begins and candles burn down.
Parked for later, your call. A tester's export already replays exactly. A viewer that plays it back lets you see where they struggled, which is most of what a playtest can tell beyond its one question.
Done. At a season's end, or when the keep falls, Make the card draws a 1080 by 1350 image. It shows the keep as it stands in the game's own pixels, how the season went, six numbers, and up to three of the season's dead from the Book. Share uses the phone's share sheet where the browser can share images; Save image downloads it. Only the download is tested (desktop Chromium has no share sheet), and nobody has shared one yet.
Done. Today's keep, at the top of Saves: everyone who plays on the same date, by their own calendar, gets the same keep, from a seed made of the date, on the rules as they ship, with the player's own numbers locked out. Its slot, playtest export and recap card name the day. There's no server, so players compare by cards; nothing stops replaying a day, and a deploy during the roughly 50 hours a date lasts somewhere changes its rules for later starts.
Done. Everyone in it goes free at once and Dread falls 1 for each, so it can turn a censure into a warning, or free a caught shade at night. The mirror is lost and 7 days of bad luck follow (sickness twice as often). The keeper plan breaking one to stop every censure finished 80 second seasons in 99 against 86 without; without the bad luck it's about even.
Done. About 2 fires a season in a Hearth or a Forge. The room's own workers fight each one, the Yard or the bell can be sent, and fighting it can kill, more so the hotter it is. Fought at once it adds few deaths (0.14–0.18 per two seasons for the autopilot) but costs 5–14 second seasons in 200. Whether it gives a slower player more choices at the rite is for the playtest.
Done. The keep sleeps 8, and each Quarters 4 more; the crowded fall sick more often. A shade dreaming in the Dreamwell makes the living work ×1.1 the next day. The night after a death a Weeper rises for each of the dead and weeps in the dark of the sleepers' twin; one that weeps its fill gives someone a nightmare (×0.6 work). It doesn't pull the night from the line as hoped. The Quarters goes where the keep builds next, the Tain's deepest floor, so the sleepers' twin is always below the line. Ignored, the Weepers give about 6 nightmares in two seasons. On the two-room start that costs nothing measurable over two seasons; on the four-floor keep it costs the doubled line 22 second seasons in 200, for reasons I haven't found. A fighter sent to guard costs 16 second seasons in 200, and a dreamer comes out even. Never building the Quarters did as well as building it when crowded, or better. For now they add texture, not a decision.
Done. By day a shade can whisper its old trade to whoever works it now (×1.25), or step through a great glass and work a room in person. Either costs it 8 memory at dusk. At 4 memory whispering was a free gain on a built-up keep (+12 second seasons in 200), so not a decision. At 8 it's +8 there and about even on a new keep. The autopilot never saves the 30 glass a great glass costs. Given it at the start, stepping through added 3 and 9 second seasons.
Done. The seasons run spring, summer, autumn and winter. Summer's days are ×1.3 as long, and so is a day's work, with nights ×0.75; winter is the reverse, ×0.7 and ×1.35, longer than a candle burns. Winter's seventh night is the Long Night, which ends the year. Summer and autumn came out easier than springs at the same hardness. Winter took 29–46% of the keeps that reached it before the Long Night, and the Long Night a quarter to nearly half of the rest: 23–45% of keeps survive a year, against 42–55% with every season a spring. Round three's double-length Long Night was won 13 times in 109, whatever its Creepers, so it lasts 1.5 winter nights. I built this before the playtest, which the plan had waited for. It also makes the playtest's second season a summer, which is easier.
The autopilot numbers on this page, each from 40 to 1,000 seeds, and scripted Chromium checks of the game, the guide, building, installing and offline play.
The sizes, the order of the queue, and what counts as a fair second season.
Whether they want a second season, whether the guide teaches the game, whether the Tain reads on a phone, and whether a real player turtles on the line the way the autopilot's doubled line does.