You got a gold star for this one. A whole mechanism typed out with worked numbers, and you spotted the vanilla-to-Burning-Crusade problem sitting inside your own design before anyone pointed at it.

Fridge, this patch and all time: 35 gold stars, 15 apples, 0 frowny faces, 0 worms.

---

First, your idea back to you

Two ladders. Both start at 1.

Ladder one is the stat's ladder. A hero rolls a check with a stat. Right after, a hidden roll happens. It starts at 1%.

• Miss it → the ladder climbs by 1. So 1, then 2, then 3, then 4.
• Hit it → the stat gains 1 point, and the ladder falls halfway back to 1. From 25 it lands on 13.
• Imagination makes it fall less far. From 25, a very little hero might only drop to 20.

Ladder two is strength's ladder. Also starts at 1%.

• Miss it → it climbs by your whole strength number. 30 strength, +30 points.
• Hit it → ladder one climbs 2 this roll instead of 1. Then ladder two drops all the way to 0. Not to 1. To zero.
• 100 strength or more → ladder two always hits.

And the shape you want out the far end: strength divided by a hundred is how many points ladder one climbs each time. 500 strength climbs 5. 550 strength climbs 5, and half the time it climbs 6 instead.

Did I get it right?

---

Four things I found

1. Some stats roll a hundred times more often than others

This is the big one, and it isn't about strength at all.

Growth comes from rolling. So a stat grows exactly as fast as it gets rolled.

In your game the hold check runs on a countdown. Past the desperate line a timer ticks down, hits zero, the hero rolls strength, and the timer restarts. Every hero. All run long. That is a strength roll every few seconds, twenty heroes deep.

Intellect rolls at a puzzle and at the mirror. Maybe five times a level.

Constitution never rolls at all. It sets bladder capacity and it feeds maximum health. Nothing ever rolls against it.

So under your rule, strength rockets, intellect crawls, and constitution is frozen forever at whatever number the hero was born with. Not because strength is the best stat. Because strength is the chattiest stat.

Can you think of five things a hero rolls a check for? I want to see which stats turn up and which ones never do.

2. Strength above 100 does nothing — and it's the drop-to-zero that kills it

Walk 100 strength through exactly what you typed.

Ladder two hits → drops to 0. Next roll is at 0% — a guaranteed miss. Then +100 → 100% → guaranteed hit → back to 0.

Miss, hit, miss, hit. Half the rolls get the bonus.

Now walk 500 strength through it. 0 is still a guaranteed miss. 500 is still a guaranteed hit. Miss, hit, miss, hit. Identical.

The drop to zero throws away one roll every cycle, and no amount of strength buys that roll back. So 100 strength and 500 strength and 5000 strength are the same character.

There is a one-line fix, and it gives you your own 550 example exactly:

Split strength at the hundreds. The whole hundreds are paid flat, every roll, no dice. The leftover runs the ladder.

• 550 strength → ladder one climbs 5 every single roll, guaranteed. Plus a ladder at 50-per-miss for the sixth point.
• 30 strength → no flat points. Just a ladder at 30-per-miss.
• 100 strength → 1 flat point every roll, and no leftover. That is your "always 2%": the base 1 plus the flat 1.

Nothing special happens at 100 any more. It is just where the leftover runs out. The rigidity you were worried about vanishes on its own.

And the ladder is doing real work for small numbers, which is worth knowing. A flat 10% is nearly invisible. A ladder that climbs 10 a miss pays out about once every 3.7 rolls — call it 27% a roll. The ladder is roughly 2.7× more generous than the flat number at the bottom, and the two meet exactly at 100. That curve — steep at the bottom, flat at the top — is the good part of your design.

3. A small one: is 500 a five or a six?

You said 100 strength always gives 2%. That reads as the base 1, plus 1 more.

You said 500 strength gives 5%. By that same pattern it should be 6%: the base 1, plus 5 more.

Which one? It decides whether a hero with 0 strength climbs 1 per miss or 0 per miss.

4. "Too rigid?" — no, and for a reason you may not have known

100 isn't an arbitrary line here. It is already the top of the scale.

A stat check in this game is a hundred-sided die rolled under the stat. That is the ruling that already governs the hold check. Strength 30 literally means 30 out of 100. Strength 100 means you never fail a strength check, ever.

So a hero at 100 strength already never fails. Saying she also always gets the bonus is the same fact said twice. That is good — a number that means one thing everywhere.

The rigid part was never the 100. It was the drop to zero.

---

You're right about rating, and here is exactly what happened

★ Insight ───────────────────────────────────── Vanilla: an item's text said "+1% critical strike chance." The game stored a percentage and added it straight into the crit roll. Level cap 60, gear tuned so a full best set landed somewhere sane.

Burning Crusade: ten more levels, and a fresh tier of loot every patch. Flat percentages left two bad choices — put small percentages on new gear so an upgrade barely moves the number and feels like nothing, or put big ones on and watch players cross 100% and break the roll.

The fix: change the currency. Gear stopped giving percent and gave crit rating, a plain unbounded integer. The game divides that integer by a number that grows with your character level. Roughly 14 rating per 1% at level 60, about 22 at 70, about 46 at 80. The same 100-rating item is worth around 7% at 60 and around 2% at 80. Nobody edits the item. It decays on its own as you climb.

The rule in one line: never let a player-facing number live in the same units as a probability. Put a divisor in between, and let the divisor grow with progression. ─────────────────────────────────────────────────

Now the part that should feel good.

You already invented rating. "One extra point per 100 strength" is a divisor. Strength is the rating, 100 is the divisor, percentage points come out the far end. You built the same machine Blizzard built, without the name for it.

The one piece you're missing is that theirs grows. Yours is nailed to 100.

Whether that matters comes down to a single number, and it's yours to give: what is the biggest strength a hero will ever have? Your max level is 20, and the bonus a hero is born with is 3 points. If a hero finishes a run at 40 strength, the whole above-100 branch is a rule that never once fires and you can delete it. If she finishes at 600, the divisor has to grow, or ladder one just pins at 100% and every roll is a free point forever.

---

Imagination is going to do something you might not expect

In this game imagination isn't stored anywhere. It is computed from regression every time it's read. And regression swings hard inside a single fight — a monster lands a hit, she is suddenly full of imagination, and it drains back toward zero over the next minute.

So "imagination softens the drop" means the softening changes second to second.

A hero who levels a stat while she is little keeps most of her ladder. The same hero, grown up again ten seconds later, loses half of it.

Being little is when the lesson sticks. That is lovely, and it costs you nothing — it falls straight out of what you already built. It also hands the player a real move: get a hero small on purpose before a hard fight, because that is when growth stays.

One guard rail. If imagination can soften the drop all the way to nothing, ladder one only ever climbs. It reaches 100% and parks there, and that hero gains a point on every roll, forever. So the softening needs a floor. Halfway down at worst, a fifth of the way down at best. Never zero.

---

Questions, biggest first

1. What is the biggest strength a hero will ever have? Everything above hinges on this one number.
2. Does this replace experience and levels? Your docs still say "what earns experience at all is open" — nobody has ever answered it. Stats that grow by use could be the answer, and a level becomes a thing that happens when your stats have added up enough.
3. Strength makes every stat grow faster, including strength. Rich gets richer. Is that on purpose? Your mind document says out loud that you avoid loops that add to themselves — this is one.
4. Is ladder one per stat, or one ladder the whole hero shares? And ladder two — one per hero, or one per stat?
5. How far can imagination soften the drop?

Tell me about the fifth one last. Start with the first.

You weren't naughty! You answered every single one of them. I put them in an order, you went through in yours, and all five came back. That's the whole job. Stickers for this one:

Fridge now: 38 gold stars, 18 apples, 0 frowny faces, 0 worms.

---

Two of your answers are secretly one answer

You said monsters have to out-scale heroes forever.

You said stats should be unbounded, and maybe the dice have to go.

Those are the same decision. Here is why.

A hundred-sided die rolled under your stat has a top. At 100 strength you never fail. Not "almost never." Never. And the moment a hero gets there, no amount of monster scaling can ever touch her again. The monster can have a thousand of anything; her hold check still passes, every single time, forever.

The race you want ends the instant one hero maxes one stat.

So you can't keep both. You already knew that. You just arrived at it from the other side.

What goes in the die's place

A contest. You stop rolling against a fixed hundred and start rolling your number against the number that's fighting you.

One line:

chance to pass = your stat ÷ (your stat + the thing resisting)

Walk it:

your strength	resisting it	you pass
30	30	50%
60	30	67%
300	30	91%
300	300	50%
3000	3000	50%

Three things fall out, and all three are things you asked me for:

1. No ceiling, no floor. It never reaches 100% and never reaches 0%. There is always a chance and always a risk.
2. Only the ratio matters. Multiply both sides by ten and nothing changes. A hero at 30 against a monster at 30 is the same fight as a hero at 3000 against a monster at 3000.
3. That is your race, exactly. Floors get harder forever, heroes get stronger forever, and what decides a run is who grew faster. No number ever makes a hero safe.

★ Insight ───────────────────────────────────── This shape has a name — it's the same maths behind chess Elo ratings and behind most "opposed roll" systems. The useful property is that it turns an absolute number into a relative one. A die under a stat asks "how good are you?" A contest asks "how much better are you than this?" Only the second question has an answer that stays meaningful when both sides have been growing for six floors.

Your whole game is the second question. The hold check is a hero's strength against her own fullness. Combat is her against a monster. Puzzles are her intellect against the puzzle. You never actually needed the first question. ─────────────────────────────────────────────────

---

Your growth is a square, not a snowball

You said you want a bit of exponential scaling. I went and worked out what your two ladders actually produce, and I have good news: it's gentler than you think.

Here is the reason. Ladder one is a percentage, and a percentage tops out at

1. So the faster it climbs, the sooner it hits, and the less the extra climbing speed buys you. Doubling how fast the ladder climbs does not double how fast the stat grows. It multiplies it by about 1.4 — the square root of two.

Points of stat gained per roll, once the ladder has settled into its rhythm:

strength	ladder climbs by	stat points per roll
0	1	about 0.15
100	2	about 0.20
500	6	about 0.35
1000	11	about 0.47

Read the two ends. Going from 0 strength to 1000 strength — the whole range you said you want — only triples how fast you grow. It does not multiply it by a thousand.

What that means over a run: a stat grows like time squared. Twice as long adventuring, four times the stat. Not eight times, not a thousand times. A square.

So your monsters don't have to beat a snowball. They have to beat a square. Give a floor's monsters a difficulty that grows like time cubed, and the heroes lose eventually no matter what, and the only question is how many floors they got through first. That is a much kinder number to tune than racing a real exponential.

(These are my numbers from the rules as you described them. They want the proving ground to confirm them, not me.)

---

Now the hard one. How do stats grow, if not from being used?

You're right that it can't be use. Strength rolls on a timer every few seconds all run long. Intellect rolls at a puzzle. Constitution never rolls at all, and inventing a fake constitution check would be contrived and would need a fudge number nobody can see.

Four ideas, best first.

1. The sliders already do this

You have priority sliders. Five tabs, one per attribute, shares that add to one, and the class is read straight off the shares.

Let growth follow the slider. Every stat's ladder climbs on every event — not just the stat that got rolled — and each one climbs by its share of the slider.

A hero with half her slider on strength climbs strength twice as fast as a hero with a quarter on it.

Why I like this most:

• Constitution is unfrozen. Nobody has to invent a check for it. Point the slider at it and it grows.
• Nothing is contrived. No fake event, no pretending being hit is a constitution test.
• There is no opaque number. You said you hate those. The scaling factor between stats is the slider, and the player set it themselves, with their own hand, and can see it on screen as a number.
• Your class and your stats can never disagree. The same slider decides both what you practise and what you become. A hero whose slider says "brawler" grows brawler stats and levels into brawler classes. Those were two systems and now they're one.

The flavour changes a little, and it's worth saying out loud. It stops being "you got strong by straining" and becomes "you got strong because you were trying to." That's a different story. A truer one for heroes who decide for themselves, maybe.

2. Shares of the trip, cashed in at rest

The other way to divide. A hero adventures, and the game counts what she actually did. Then at a basecamp she rests, and each ladder climbs by that stat's share of everything she did — not the count.

500 strength rolls and 5 intellect rolls this trip becomes "99% strength, 1% intellect," and the ladders climb in that proportion.

The frequency problem cancels itself, because you divide by the total. There is no tuning number at all. You don't need a scaling factor; you need a division.

The catch is the 99-to-1. Chatty stats still win, just less absurdly. You'd want to divide by each stat's usual rate, and that usual rate is exactly the opaque number you don't want.

3. One event per stat, chosen to be equally common

Give each attribute one feeding event, picked so the five happen about as often as each other.

stat	grows when
strength	a hold check runs
constitution	she takes damage
dexterity	an attack is dodged, either way
intellect	a spell is cast, either way
spirit	regression or pride moves

Honest, and it keeps "you grow at what you do." But it needs five rates balanced against each other, which is the nightmare you already named. Third for that reason.

4. A cap per floor

Each stat can only grow so many times per floor. Strength rolls 500 times and only the first handful can pay out.

Crude. Completely transparent. It works. It's here at the bottom because it feels like a wall rather than a rule.

---

So here's my big question for you, and it's the one that decides this:

Should a hero grow toward what the player wants her to be, or toward what actually happened to her?

Those are ideas 1 and 2, and they are the two real answers. Everything else is patching. And "both, multiplied together" is a completely legitimate third answer — the player points, the run pushes, and a hero ends up somewhere between.

---

The two you asked me to pin down

Imagination's floor

You said linear on regression, and you're right that it has to be. Here's the exact number and why.

The drop is halfway back down at zero imagination. At full imagination it softens. The only value that actually breaks is exactly zero drop — then the ladder only ever climbs, pins at 100%, and that hero gains a point on every roll forever.

Any drop at all is safe. I checked what a 10% drop does: the ladder settles at about 32% instead of about 15%, and the hero grows two to three times faster. Fast, not runaway.

So: halfway down at zero imagination, a tenth of the way down at full imagination, straight line between them. One entry in the balance table, and the proving ground moves it.

Regression is a narrow window — here's the sharper version of your warning

You said don't let regression effects scale exponentially, because 0-to-100 is narrow. You're right, and there's a crisper rule underneath it:

Anything computed from regression must multiply. It must never add.

A multiplier keeps its worth forever. Imagination tripling your magical damage is worth 3× at 30 intellect and worth 3× at 3000 intellect. It never goes stale.

An addition drowns. Imagination adding 50 to your magical damage is huge at 30 intellect and invisible at 3000.

That one rule protects every bounded number in the game — regression, imagination, confidence, pride — from being eaten by the unbounded ones. It's worth writing into the vocabulary document as a law.

And your "under a thousand"

You want unbounded and under a thousand. Those don't fight, they just need naming separately:

• Unbounded means no rule stops the number. No cap, no clamp, nothing that says "you may not exceed."
• Under a thousand means in practice a run ends before a hero gets there.

You already said what ends a run: the floors run out, because there are more ghosts than floors the guild can clear.

So the number you tune isn't a cap. It's how steeply the floors scale. Tune it so a guild that goes as deep as anyone ever goes comes home with heroes in the high hundreds. Nobody caps anything, and nobody sees four digits.

---

Still open, and I'd like these before we touch the second idea

1. Player's intent, or hero's experience? The big one above.
2. What grants a level, if not stats adding up? You said levels give classes. Something still has to hand out levels. Floors cleared? Ghosts honored? Time?
3. Does spirit belong in this too? You're right that spirit raises imagination, which softens the drop, which speeds growth. So spirit is a second growth stat, working through a different door. Strength buys bigger climbs; spirit buys smaller falls. That's a nice pair — the two aren't redundant. But it means a spirit-heavy hero and a strength-heavy hero both grow fast, in ways that stack. Want that?
4. Do monsters grow this way too? They run the same digestion and the same hold check. Do they run the same ladders?

Want me to write down what you settled today — per-stat ladders, the 1% base, the split at the hundreds, levels giving classes, monsters out-scaling heroes — before we start on the second idea? Or would you rather just keep going and I'll catch it up after?

Five more stickers.

	for
⭐ gold star	the deck — five percent a card, drawn without replacement, reshuffled
⭐ gold star	"a level ten thousand person doesn't learn ten thousand more bits than a level 1 person. Usually it's the opposite"
⭐ gold star	experience from being cared for, and from a submissive or dominant experience
🍎 apple	yes to strength and spirit both feeding growth, through different doors
🍎 apple	monsters aren't persistent, and nobody knows what's in a room until the heroes queue at the door

Fridge: 41 gold stars, 20 apples, 0 frowny faces, 0 worms.

I put the brainstorm in tmp/shared-memory/stat-growth-brainstorm.md, which is RAM and vanishes on reboot. The tmp symlink wasn't there, so I made it and the /dev/shm directory behind it, the way your rules say the project should have them. That's a new symlink in the project root. Already in .gitignore.

---

You already built the thing you said you wanted

This is the best thing I found and I want it first.

You said a level-10,000 person doesn't learn 10,000 times more from an experience, and usually it's the other way round, and games get it backwards.

Your system already does it. You didn't have to add anything.

Here's why. We moved to a contest roll, so only the ratio matters. That means what a point of strength is worth is a fraction of what you already have. Going from 10 to 11 is a tenth. Going from 1000 to 1001 is a thousandth.

Now put your ladder next to that:

a hero at	gains per roll	which is
10 strength	0.15 points	1.5% more than she had
1000 strength	0.47 points	0.05% more than she had

The veteran gains three times more points and thirty-two times less that actually matters.

And the shape isn't arbitrary. Learning slows off as one over the square root of what you already know. That is the same curve psychologists find when they measure real people practising real skills — it's called the power law of practice, and it's one of the most reliably measured things in learning. Fast at first, then a long grind for smaller and smaller returns.

You wanted the numbers to reflect that heroes are human. They already do.

---

Is it too convoluted?

No. It's five parts, but it's one idea, said five times.

Look at what every piece actually is:

part	what it is
the deck, without replacement	a stat that hasn't been picked gets likelier to be picked
ladder one	a stat that hasn't grown gets likelier to grow
ladder two	a bonus that hasn't landed gets likelier to land

All three are "a thing that hasn't happened yet keeps getting more likely until it does." That's a pity timer. Your whole growth system is pity timers, top to bottom.

That's not convoluted. That's a theme. A player doesn't have to learn five rules, they have to learn one rule and notice it three times. And the feeling it creates is the one you want: nothing is ever wasted, and a bad streak is always paying into something.

Two things I'd change, both small.

Make it one deck instead of a coin flip plus a deck

Your version: flip a coin. Heads, a random stat. Tails, draw from the deck.

Same maths, one mechanism: one deck of forty cards.

• 20 slider cards — one per 5% of slider, exactly as you said
• 20 wild cards — four for each of the five stats

Draw without replacement. Reshuffle when spent.

A stat with share p gets (20p + 4) ÷ 40 of the picks, which works out to one tenth plus half the slider — identical to your coin flip, to the digit.

Why bother: your coin flip has replacement and the deck doesn't. So your random half can go on a cold streak and skip a stat for ages, while the deck half can't. One deck makes the pity timer cover everything, which is what you actually wanted.

And it means the growth system is one object a player can look at: a deck of forty cards, and you can see what's in it.

Ladder two is worth about seven percent

Honest finding. I checked what ladder two buys over just flipping a straight coin for the leftover.

At 50 leftover, the ladder pays out 0.67 bonus points a roll; a flat coin pays 0.50. At 10 leftover, the ladder pays 0.27 and a flat coin pays 0.10.

Sounds big. But it feeds into ladder one, and ladder one's growth rate goes as the square root of what feeds it. So the whole difference in how fast a hero grows is about 7%, and less as strength rises.

Ladder two is also the single hardest part of the system to say out loud.

I'm not telling you to cut it — it's yours, and it fits the pity-timer theme. I'm telling you it costs the most to explain and buys the least, so if you ever want the system a notch smaller, that's the notch.

---

Where your numbers actually land

Your share of picks runs from 12.5% (a stat at the 5% slider minimum) to 50% (a stat at the 80% maximum, since the other four need 5% each). A four-fold spread. Specializing matters, and nothing starves.

With your linear thresholds at 100 per level — 100 for level 2, 200 for level 3, and so on — here's where a hero's stats sit:

level	total exp	ignored stat	even stat	focused stat
5	1,000	18	30	83
10	4,500	92	217	571
12	6,600	151	274	1,011
15	10,500	238	590	2,121
20	19,000	618	1,259	5,856

Look at level 12. A focused stat crosses one thousand at almost exactly level 12, on the numbers you typed, without anybody tuning anything.

So your instinct about "a hundred or two hundred per level" was good. It just answers a question you haven't asked yet:

Is the real game over around level 12?

Because that's where your thousand is. If a guild that goes as deep as anyone ever goes comes home at level 12, your numbers are already right and I wouldn't touch them. If you want people reaching level 20, either the thresholds have to climb faster than linear at the top, or you accept four digits up there.

There's a nice way to read this that matches what you already said: level 20 is a thing nobody sees. There are more ghosts than floors a guild can clear, so the cap isn't a destination, it's a horizon. Level 12 with a thousand-point stat is what an extraordinary run looks like.

---

The stacking you said yes to

You wanted strength and spirit to both feed growth. I worked out how much they're worth so it's not a mystery:

• Strength, from 0 to 1000, multiplies growth by about 3.3× (it makes the ladder climb faster)
• Spirit through imagination, from grown-up to as little as she gets, multiplies by about 2.2× (it makes the ladder fall less far)
• Both at once: about 7.4×

That's a real snowball and not a runaway. And it limits itself, because to keep imagination high a hero has to stay deep in regression, which means low current maturity, which means she's in the dangerous state taking extra damage, and she's close to the line where her mind breaks.

The fastest-growing hero in the guild is the one closest to being lost. That's a good trade to hand a player.

---

Questions

1. Does the caregiver earn the experience, or the hero being cared for, or both? I'd say both — one learned to give and one learned to receive — but it's yours. Same question for the submissive and dominant pair.

1. Is a bad thing worth experience? A hero has an accident in front of everyone. A hero goes crybaby and gets calmed at a basecamp. Those are "difficult situations that require discernment," by your own definition. If they pay, then a hero's worst moments are also how she grows, and I think that might be the heart of your whole game.

1. What's the difficulty number on the other side of the contest? The roll is now stat ÷ (stat + difficulty). For a monster that's its stat. For a puzzle it's a number the puzzle carries. But for the hold check, what does strength contest against? Fullness above the desperate line? Something else?

1. Can you think of five encounters that aren't a fight or a puzzle? You gave me caregiving and the submissive-or-dominant one. I'd like three more, because the more of them there are, the less the game is "kill things to level," and that seems to be what you're steering toward.

Ready for the second idea whenever you are.

No worm. No frowny face either. You answered. "No, because they're all going to be fights or puzzles, and the puzzles carry the variety instead" is an answer — and it's a better one than the five I asked for, because it says no to my premise and gives three reasons. Saying no to me is trying at its hardest.

Now the pattern hunt. I found fourteen of them.

Five more stickers. ⭐ gold star for spotting the avoidance value yourself, unprompted. 🍎 apples for the one deck, for "both," for bad things paying too, and for correcting me about the plus and minus buttons.

Fridge: 43 gold stars, 24 apples, 0 frowny faces, 0 worms.

---

I found fourteen

You were right about avoidance, and it goes much further than that. I went through every document and counted. Fourteen of these, in six different systems, and not one of them was planned.

where	the counter	what makes it fall
growth	the stat ladder	halfway back toward 1
	the strength ladder	to zero
	the card deck	reshuffled when spent
combat	the avoid value	part of the way back to the cap
	threat toward an attacker	nothing, inside a fight
the body	overflow stacks	nothing, ever
	poop stun stacks	nothing, ever
	dependence	nothing, ever
	fullness above the desperate line	to zero, at a toilet or a void
the map	a room's empty-tickle count → infestation	to zero when a hero walks in
	the same count → where heroes wander next	to zero when a hero walks in
furniture	hypnotic stacks	undecided
	playground stacks	they expire
the leader	suggestibility	communing

(The rocking horse is arguably a fifteenth — regression piling up until it tips into crybaby.)

There's one formula, and it covers all of them

Let the counter climb by c points each time the thing doesn't happen. Let it fall by the fraction f of its height when it does.

resting height = 10 × √(c ÷ f)

And since the height is the chance, that's also the rate.

The important bit: both knobs are under a square root. Double how fast it climbs, get about one and a half times the rate. Halve the fall, same. Nobody can break one of these by turning a dial too far. That's why they're safe to hand to a balance pass and walk away.

Check it against the growth ladder: climbs by 1, falls halfway → 10 × √2 = 14%. That's the number I gave you last time, and I got it by simulating. Same answer.

The fall is where the meaning lives

This is the part I think is actually yours to own. The three falls aren't three settings. They're three different stories.

fall	it says
to zero	the thing is fully in the past. Every cycle starts fresh and knows nothing. A room a hero just walked through is simply not neglected.
partway	it left a mark. Next time comes sooner. The growth ladder, the avoid value. This is habituation, and it's right whenever a body or a mind is learning.
never	a road, walked one way. Overflow, dependence, poop stun. This is wear, and only an outside hand — a curative — lifts it.

Your three routes to incontinence are all the "never" kind. All three. That's not a coincidence and I don't think you did it by accident, even if you didn't do it on purpose.

The best one: it's two things at once

The avoid value is the one to keep.

A defender who keeps getting away with it relaxes, so the value decays on every swing she avoids, so the next one is likelier to land.

From her chair, that's complacency.

From the attacker's chair, it's a pity timer — keep swinging, an opening is guaranteed.

Complacency and persistence are one number read from two ends. Whenever this pattern sits between two people, there's always a second name for it from the other chair, and it's usually the better name.

Same thing with the sign. Pointed at the player it's mercy — a stat that hasn't grown is due. Pointed at the world it's dread — a room that's been quiet is due. One machine, and the whole feeling of the game depends on which side you're standing on.

Written down

strategems/a-counter-that-climbs-until-it-fires.md — the shape, the formula, the three falls, the two-ends rule, all fourteen places, and when not to reach for it. One bullet added to patterns-that-keep-working.md pointing at it, and both added to the table of contents.

---

Your plus and minus buttons change the numbers

Good correction. I checked the slider — twenty steps, a five percent floor, and the last tab can't be removed. So one tab is 100%, and a stat with no tab is at zero.

With the forty-card deck:

a stat at	cards	share of picks
no tab at all	0 slider + 4 wild	10%
even fifth	4 + 4	20%
the whole bar	20 + 4	60%

A six-fold spread, not four. At level 12 that's a hero with 1,344 in one stat and 115 in the other four. Twelve to one. That's a genuinely extreme person, and nothing starved — she still has a hundred-odd points in everything, bought entirely by wild cards.

---

What shrinks a bladder

You asked, and here it is, whole:

capacity = the base entry, plus the per-constitution entry for every point of constitution, then reduced by the level of the first continence effect.

Only one thing shrinks it: the first continence effect, which is called "bladder capacity." Each level of it takes some off. It arrives from monsters, curses and haunted furniture, and a curative takes it fully back off — level and all, not one step at a time.

Nothing else touches it. Not time. Not accidents. Not regression. Not being wet.

The whole digestion dataflow, end to end

Check this against what you remember:

Two ways in. Eating puts a food unit in the tummy — a kind, and a count of increments left. Drinking adds to the tummy's fluid pool.

Every food tickle, solids: every unit in every tummy in the world steps forward one increment. A laxative multiplies the step, and because a step is a whole number, the fraction is carried to next tickle rather than thrown away. When a unit runs out of increments it leaves the tummy and lands in the bowel, adding its bulk.

Every food tickle, fluids: a fraction of the tummy pool moves into blood fluid. Blood fluid has a set-point. Below it, she's thirsty. Only the fluid above the set-point flows on to the bladder, at a rate proportional to how far above. A diuretic adds bulk straight to the bladder, from nowhere, which is on your open list.

That one rule does three jobs: drinking cures thirst first and makes pee second, a thirsty hero barely pees, and bleeding makes her thirstier and drier at once.

Two ways out.

Bowel fill against bowel capacity gives the poop urge. Poop stun stacks stun her at the instant the urge starts, so her first few chances to hold are simply missed. Past the certain-mess count she messes at least a little, guaranteed.

Bladder fill against bladder capacity gives the holding line and the desperate line. Past desperate, a countdown runs — its length set by the second continence effect, bladder strength. Zero means a hold check, then it restarts.

The hold check is two rolls: strength, and confidence weighted by pride. Both get worse the fuller she is.

• both pass → she holds
• one fails → dribble, and one overflow stack
• both fail → void, the whole thing
• fill hits capacity between checks → automatic void, no roll, and the voluntary switch and sensation both switch off for a while

Three roads to incontinence: dependence past its threshold, overflow past its threshold, or all four effects maxed.

Want me to make that a numbered docs file? It'd be the first real dataflow page and I think you're right that it's owed one.

And your contest number for the hold check

Here's the shape I'd propose:

chance to hold = strength ÷ (strength + difficulty) difficulty = a base number, plus a scale number × how far past the desperate line she is, as a fraction of the room left

Two entries in the balance table, both settled by the proving ground, neither by me.

What I like about it: constitution is already in there and it didn't need mentioning. Capacity comes from constitution, and fullness is measured against capacity, so a hardy hero simply spends less of her life past the desperate line at all. Constitution buys fewer checks. Strength buys the check. That's exactly what your document already says, and the contest roll keeps it without being told to.

---

Two problems I hit while reading

1. Does bladder capacity read the real constitution, or the penalised one?

Your penalty pipeline multiplies all five primaries. Capacity is built from constitution. So as written, a starving hero standing next to four messy friends has a smaller bladder. I don't think that's what you want — capacity is a fact about a body, not a performance. But nothing in the documents says which one it reads.

2. Strength now has four jobs, and your own document already complained about three.

The open questions in the body document say: "Whether strength is now too broad. It is most of physical damage, four parts in five of speed, and the stat that holds the bladder. Three unrelated jobs on one slider share, where constitution's three are all one idea."

We just gave it a fourth: it's the growth engine for every stat in the game.

Constitution's three jobs are all one idea — how much a body takes before it stops coping. Strength's four are four different ideas.

There's a tidy alternative and I'll say it once and then drop it: spirit is already the "how hard your mind swings" stat, and growth is a mind thing. Putting the growth engine on spirit instead would give spirit a second real job and take a job off strength, which answers a question you already had open. You said you want both feeding growth and I'm not arguing with that. I'm saying strength is the one carrying too much, and this is the moment where it'd be cheap to move something.

---

Questions

1. Real constitution or penalised constitution for bladder capacity?
2. Does the growth engine stay on strength, or move to spirit?
3. Shall I write the digestion dataflow as a docs page?
4. Do hypnotic stacks ever fall off? It's the one counter in the fourteen whose fall nobody has decided.

And then — the second idea. You've been sitting on it for four rounds.

Yes to all three. Let me get the stickers on the fridge and then go build the page.

Found a real gap while mapping it. Let me claim the index and write the page.

Three stickers. ⭐ for the bladder ruling — the maggoty steak is exactly the right way to think about it. ⭐ for the hypnotist. 🍎 for deciding to judge stat value at the balance pass instead of guessing now.

Fridge: 45 gold stars, 25 apples, 0 frowny faces, 0 worms.

---

The page is written

docs/046-from-the-plate-to-the-garment.md, claimed through the file tool so the counter stayed honest.

It has a drawing of the whole pipe at the top, then every number the body owns with its datatype, then one food tickle in order — the seven things that happen and in which order — then every line with both sides of the fork named, then the hold check, then what happens when something leaves, then who reads these numbers from outside the body.

Your ruling is in it, and in 007, and in 005, with your reason kept in your own terms.

The validator says "the documents agree with each other." 141 pages, 846 links, all of them resolve.

Two things were broken and I fixed one

005 was lying. The hero record page said bladder capacity was "base times the first continence effect's level." Constitution wasn't in it at all. That line predates the whole "capacity is constitution's" decision and nobody went back for it. So the record page and the body page disagreed about how to compute a number, and the record page is the one somebody would build from.

Fixed, and it now says raw constitution, and links to the new page.

And one I can't fix, because it needs a number from you.

The garments page says dependence "lowers how often the hero bothers with hold checks at all, by the dependence check scaling entry."

There is no such entry. The balance table has dependence_threshold and nothing else.

So right now: a hero builds dependence, dependence is documented as making her stop trying to hold, and there is no number saying by how much. That mechanism cannot be built. It's a shape with no size.

It slipped past the validator because the name was written in plain words instead of backticks, and backticks are what the validator checks. I've written it up in the new page's "a mechanism with no number behind it" section so it stops being invisible.

---

Your hypnotist is bigger than you might have meant

I recorded it in the haunted furniture page, both halves, with your words as the specification.

Here's why I think it matters more than it looks.

It is the first thing in the whole game where one hero can undo something done to another hero's mind. A curative does that for the body — it lifts a continence effect completely, level and all. Nothing did it for the mind. Traits went in and stayed in and that was the end of it.

And look at the shape: the daycare installs the triggers with hypnosis, and the hypnotist removes them with hypnosis. Same tool, opposite direction. That's the two-ends pattern from the strategem, showing up for the third time today.

And a thing that falls out of it for free

You said the hypnotist stuns enemies.

Your rules already say what a stunned, hypnotised person does: something silly. Your two examples are pooping on purpose and wrestling an ally.

Your rules also already say that monsters run the same two pipelines and the same hold check as heroes. And: "a monster in pants that wets or leaks runs away and drops what it carries."

Put those together.

A hypnotist can make a monster poop itself, and a monster in pants that has an accident flees and drops its loot.

That is a hypnotist routing an enemy without ever hitting it. It's already legal under rules you wrote months apart, and nobody had to add anything. It's funny, it's on-theme, and it gives the hypnotist a real combat identity that isn't "stun bot."

---

Questions about the hypnotist, best first

1. What makes a hero one? Caregiver isn't chosen — a hero becomes one by having enough current maturity, and stops being one the moment she drops below the line. Is hypnotist read off something like that, or is it a class off the priority sliders, or is it picked?

1. Can the hero being worked on refuse? Having your triggers taken out is still being hypnotised by somebody. A hero who's suspicious, or proud, or very grown-up might not want to go under. And a hero with high spirit is easier to put under, by your own rule — so the hero who most needs the therapy is the one it works best on. That's a nice trap.

1. Where can a session happen? Basecamp, any quiet room, or in the middle of a fight?

1. One role or two? Stunning monsters and taking traits off a friend are very different acts. They could be one person, or the stun could be a spell anyone can learn and the therapy the actual role.

---

Also, small: the project's instruction file says no licence has been chosen, but new-source-file stamps every file AGPL-3.0-or-later and COPYING.md is sitting right there. That instruction is stale. Want me to correct it?

Nothing is committed. Say the word and I'll commit the new datapath page, the strategem, the three document fixes and the table of contents, with the transcripts riding along.

Now — the second idea. You've been holding it for five rounds.

Oh no. Five rounds is a long time to hold something. That's on me for keeping you waiting — go get changed, the idea will keep. Stickers first, then I'll do the three things you just said yes to.

Now the three things you said yes to.

Four stickers. ⭐ for the class-skill ruling. ⭐ for the consent reasoning — that one's going to change how the whole role works. 🍎 for basecamp-only. 🍎 for splitting offensive from restorative.

Fridge: 47 gold stars, 27 apples, 0 frowny faces, 0 worms.

The licence line is fixed. It now says AGPL version 3 or later, names COPYING.md, and says that the file tool is the one place the notice is written so nothing ever carries a stale copy.

---

Your class ruling changed one word and it mattered

The reading used to be "nearest shape." It's now "nearest shape she qualifies for."

And the order is the whole thing. Strike out the rows she fails first, then find the nearest. If you filtered afterward you'd get a caregiver who can't caregive — the right class with a hole punched in it. Filtering first gives her a different class, which is exactly what you said.

It forced a new rule I had to invent, and I want you to check it.

Every gated class now needs an ungated neighbour sitting close to it in share-space. Here's why: a player builds a hero whose sliders land dead on the caregiver point, and she fails the requirement. She has to land somewhere. If there's nothing nearby, she falls to whatever happens to be next-nearest — which could be across the room. You'd aim at a caregiver and get an archer.

So it's a load-time refusal: a gated row with no ungated row within the fallback distance is refused when the catalogue loads, and the game says so instead of quietly producing nonsense later. Written into issue 610.

And caregiving now has two gates

You already had a live one: a hero stops being a caregiver the instant her current maturity drops below the line, and starts again the moment it climbs back. Your new ruling adds a permanent one at level-up.

Both stand, and they do different jobs:

gate	read	failing it means
the class requirement	once, at a level-up	she doesn't have the skill at all
the caregiver line	continuously	she has it and can't use it right now

A hero needs both to change a diaper. One says what she is; the other says what she can do this minute. I wrote it as a working ruling because it's the only reading that keeps both of your rules true — but it's yours to confirm or overturn.

---

The consent answer builds a chain you didn't ask for

Your two floors:

• not susceptible → she won't go under, so there's nothing to work with
• low confidence → she won't ask, because a hero who doesn't trust herself doesn't hand a stranger her subconscious

Both are floors, so the hero who gets helped is the one in the middle. That's your knife edge again. It's the embarrassment of being walked in on again. Your own mind document already says it out loud: "When a rule in this game wants a hero to feel something, it is usually the middle of a band that feels it." You keep landing on your own shape without going looking for it.

Now watch what falls out.

Spirit invites hypnosis, so a spirited hero collects traits fastest and clears the susceptibility floor easily. But traits come from mirrors, and a run that hands a hero a lot of them is not usually a run that leaves her confident.

So the hero who most needs the therapy can be too shaken to ask for it.

And the way out is already built, because confidence is the one thing in this game that heroes can hand each other — praise, a fresh sticker, a caregiver standing in the room, a double diaper.

A caregiver is what makes a hero able to accept a hypnotist.

Two restorative roles, chained, one lifting a hero to where the other can reach her. Neither was designed for the other. And it happens at a basecamp, which is already where the caregiver would be standing.

---

One contradiction I found and left alone

Issue 610 says: "Experience comes from monster kills, split among the heroes in the room."

You said today that experience comes from encounters — puzzles and battles — plus being cared for, plus a submissive or dominant experience, plus a hero's worst moments.

Those don't agree, and 610 is the blueprint somebody would build from. I left it. The growth system isn't written up yet and half-changing it would be worse than leaving it plainly wrong. When we write the growth system down, that paragraph gets replaced whole.

---

Everything validates. 141 pages, 852 links, all resolving. Ten test files pass. Nothing of yours was touched.

Two questions, for when you're back

1. Which stat gates caregiving? The live line reads current maturity — but maturity isn't one of the five on the slider. So if the requirement reads maturity, a player can't deliberately build a caregiver. They build the shape and then find out. Your equipment page already says "a player who accidentally builds that shape gets one," so that might be exactly right. Or the requirement should read one of the five and be buildable on purpose.

1. Which stat gates hypnotism — and do the stunning half and the mending half want different ones?

Nothing is committed yet. Say the word and it all goes in together: the datapath page, the strategem, the hypnotist, the class gate, the licence line, and the three document fixes, with the transcripts riding along.

Go get changed. The idea will still be there.

That answer solved a problem I'd papered over. Stickers first.

Stickers. ⭐ for the paired classes — that one deleted a number I'd made up. 🍎 for the two hypnotism builds.

Fridge: 48 gold stars, 28 apples, 0 frowny faces, 0 worms.

---

Your answer was better than mine, and I've torn mine out

Last round I invented a fallback distance — a balance entry saying how close an ungated class had to sit to a gated one, so a hero who failed a requirement didn't fall somewhere silly.

Your answer makes that number zero.

Two rows sit at the same point. A hero who fails the requirement doesn't fall toward anything — she lands on the same shape, in its other version. The player who built a shape always gets the shape they built. Maturity only decides which of its two lives that hero is living.

The entry is gone from the blueprint. You don't need a number for a distance that doesn't exist.

And it flipped a refusal upside down

Issue 610 used to refuse two rows with identical shares wanted — the nearest-shape reading between them would be a coin flip, so the catalogue wasn't allowed to contain one.

That's now the design. So the refusal moves to the thing that would actually be a coin flip:

• two rows at the same shape whose requirements both admit the same hero → refused at load
• a shape whose rows leave a gap, where a hero qualifies for neither → refused too, from the other side

The requirements have to partition the heroes at that shape. Every hero lands on exactly one row. Not two, not none.

---

The names are doing more work than you might have meant

Look at your two lists side by side:

high maturity	low maturity
mailman, doctor, scientist, lawyer, teacher	fighter, cleric, footman, ravensworn, sacred blade, eternal serene

A grown-up hero has a job. A little hero has an adventure.

Every single high-maturity name is something a real adult does for a living. Every single low-maturity name is what somebody playing pretend calls themselves.

And that's your game's own subject, showing up in the class list without anybody putting it there. Imagination comes from regression. A regressed hero has more of it. And the first thing imagination decides is who she thinks she is.

She is not a footman. She is the Sacred Blade.

The catalogue should never say so out loud. Same rule as the alphabet — it stays implicit, and it works better that way.

---

The thing you half-noticed and then talked past

You wrote "okay those are pretty high level classes" and then revised toward early names and later names.

That's you spotting a third gate: level.

And it's almost already yours. You ruled that levels give classes — that's the only thing levels do in your design. So if levels give classes, higher levels should give better ones. A level requirement on a row would be levels doing precisely the job you said they're for.

Which would make a shape hold four rows, not two:

	grown-up	little
early	mailman	footman
late	teacher	ravensworn

And then a hero's class name tells you two things at a glance — how grown-up she is, and how far she's come. One word on a sheet, carrying both.

I didn't rule it. It's written into 610's open list as a question, because you said it sideways and I'd rather you said it straight. Do rows carry a level requirement?

---

Hypnotism's two builds, and which half goes where

You said an intellect build and a spirit build. You didn't say which one stuns the monster and which one sits with the friend. Both readings work and they make different people.

My recommendation, and it's only that:

build	half	why
intellect	the outward stun	Intellect is already magical damage and puzzle rolls. It's the applied-technique stat. Putting a monster under with a phrase is a technique you learned.
spirit	the inward session	Spirit is already "how hard a mind swings" and "what the world does to the hero." That's the empathy stat. To work through somebody's triggers you have to be able to feel where she is.

I've written that as the working ruling with both readings recorded, so if you want it the other way it's one line.

Here's the knot it ties, either way. Spirit raises susceptibility to hypnosis. So a spirit-built hypnotist is the most hypnotisable person in the guild. She'll collect the most traits. The hero whose job is taking traits off other people is the hero likeliest to be covered in them.

And whether she can work on herself is still open.

---

One asymmetry worth a look

Caregiving can't be built on purpose. Its requirement reads maturity, and maturity isn't on the slider. A player drags dividers, builds a shape, and finds out whether that hero is a doctor or a footman.

Hypnotism can. Both its builds read intellect or spirit, and both are on the slider. A player can aim at one.

So caregivers happen to you and hypnotists are chosen. That might be exactly right — the guild that gets a caregiver got lucky, and the guild that wants a hypnotist plans for one. Or it might be an accident of which stat each one reads.

Is that on purpose?

---

Everything validates — 141 pages, all links resolving, ten test files passing, nothing of yours touched. The whole day's work is sitting uncommitted whenever you want it to go in.

Four answers, and one of them answers a question you asked me. Stickers.

Stickers. ⭐ for the journey. ⭐ for dexterity being flexibility. ⭐ for the monk. 🍎 for spotting the outward-and-inward shape of your own stats.

Fridge: 51 gold stars, 29 apples, 0 frowny faces, 0 worms.

---

The journey needs no new machinery

I want to show you this because it's the good kind of surprise.

You said classes develop into each other — apprentice before carpenter. That sounds like it needs a tree: every class pointing at what it becomes.

It doesn't. The level gate already built it.

Put two rows at the same shape. One wants level 1. One wants level 8.

A hero standing on that shape is the apprentice until she hits 8, and is the carpenter afterward. Nobody wrote an edge. Nobody listed a successor. The succession is just what happens when one of her coordinates moves and the others don't.

So a row is a point in three coordinates now — its shape, the stats it demands, and the level it wants — and a journey is walking up one of them while standing still on the rest.

That keeps issue 610's original promise intact, too. It said "the catalogue's rows are the tree, laid out in share-space rather than as a list of successors." That's still true. The space just grew a dimension.

The partition rule now runs in all three. Two rows at one shape that admit the same hero at the same level: refused at load, it's a coin flip. A gap where a hero qualifies for nothing: refused too. Every hero lands on exactly one row.

One thing I ruled that you should check

Dragging the sliders changes which ladder a hero is on, not which rung.

A hero half-way up the carpenter ladder who gets steered toward intellect arrives on the scribe ladder at whatever rung her level allows. She doesn't start over.

My reasoning: the level is the hero's and the shape is the player's, and changing your mind about a hero shouldn't cost her what she's done. But it's your call and it's one line to flip.

---

Yes, five. And here's why it isn't six

You asked. The structure you spotted is real, and it's already sitting in your documents — nobody wrote it down as a structure, but every page agrees with it.

Your stats come in act-and-endure pairs.

	acts on the world	endures the world
the body	strength	constitution
the mind	intellect	spirit

And look how tightly the first pair is already built:

• strength is four fifths of speed — constitution is four fifths of health
• strength holds the bladder — constitution sizes it
• speed is what you do; health is what you take

The second pair too: intellect looks away from the mirror; spirit is drawn into it. Same mirror, opposite directions.

So where's the sixth? Dexterity has no partner.

It doesn't need one. Dexterity already contains its own opposite. Your combat page says it in one sentence: "Dexterity is impact, on both sides: what a blow of yours is worth, and what a blow against you is reduced to."

It's the only stat in the game that sits on both sides of the same sentence. Four stats make two pairs, and the fifth is a pair by itself.

Five is right. Not because six wouldn't fit, but because dexterity is already two things and adding a sixth would break the symmetry rather than complete it.

(Small aside: this is probably why willpower feels awkward. It resists hypnosis, and spirit invites it, but intellect is already the looking-away stat. Willpower might be doing a job one of the five could do. Your mind page already has that as an open question and this is an argument for folding it away.)

---

Your monk answered an objection before I could raise it again

Last round I said: a spirit-built hypnotist is the most hypnotisable hero in the guild. The one who takes traits off people would be the one likeliest to collect them.

Spirit plus constitution fixes it exactly.

• Spirit lets her feel where the other hero is. It's how hard a mind swings.
• Constitution lets her sit inside somebody's worst hour without being moved by it. It's how much a body takes before it stops coping.

She feels everything and holds. Spirit alone would have made her fragile. You put the endurance stat next to the feeling stat and the role became possible.

And "monk mantras" instead of a trance changes what she is. She isn't a therapist putting someone under. She's chanting with her. The work is endurance, not technique. That's why it's the two inward stats and not the two outward ones.

And your dexterity reading unifies rather than splits

I was braced for "dexterity is flexibility" to fight with "dexterity is impact." It doesn't. Both are one idea:

Dexterity is precise placement.

Placing a blade. Placing your guard — your combat page already defines avoidance as "how diligently a body is being placed." And now: placing your own mind alongside somebody else's while you adjust it.

Same skill, three surfaces. That's dexterity earning a second idea without losing its first.

---

Three questions

1. Do you want "class" renamed to "job"? You said "it's the same with classes. Sorry, jobs." That matters, because the vocabulary is fixed and there's a tool for changing it.

Here's the tension. Your grown-up names are literally jobs — mailman, doctor, scientist, lawyer, teacher, carpenter. Your little names are not jobs at all — ravensworn, sacred blade, eternal serene.

So "job" is the right word for half the catalogue and the wrong word for the other half. And "a grown-up has a job, a little one has an adventure" is the whole two-register split. If everything is called a job, that split loses its word.

Three ways, best first: keep class as the umbrella and let job be what the grown-up half is called in the fiction. Or rename everything to job and accept that a ravensworn has an odd one. Or find a third word that covers both.

2. Does dexterity sit on the monk, or on the stunner, or both? Your reason for it describes the restorative act — adapting to someone's mind while adjusting it. But you said "hypnotists," not "the monk." I left it as a question rather than guess.

3. Is "hypnotist" still the right word for the restorative side, now that it's mantras? A monk isn't a hypnotist.

---

There's a lot sitting uncommitted now — the datapath page, the strategem, the monk, the ladder, the licence line, six document fixes. Want me to put it in?

That's a proper design session's worth. Stickers, then the list.

Stickers. ⭐ for reading the class off a hero's naked stats instead of the sliders. ⭐ for the four dexterity pairings. 🍎 for settling that class and job mean the same thing.

Fridge: 53 gold stars, 30 apples, 0 frowny faces, 0 worms.

---

The shape came out flat on top

You guessed three stats would have the most. It ties with two, at ten each. So the diamond has a flat middle rather than a point:

  one stat      • • • • •                     5
  two stats     • • • • • • • • • •          10
  three stats   • • • • • • • • • •          10
  four stats    • • • • •                     5
  five stats    •                             1
                                            ---
                                             31 shapes × 2 registers = 62 classes

Sixty-two names. Here they are. Yours are starred.

One stat — the specialists

stats	grown-up	little
strength	Porter	Fighter ★
dexterity	Watchmaker	Quickfinger
constitution	Nurse	Bulwark
intellect	Scientist ★	Wizard
spirit	Counsellor	Ravensworn ★

Two stats

stats	grown-up	little
strength + dexterity	Acrobat ★ / Ballerina ★	Tumbleknight
strength + constitution	Mailman ★	Footman ★
strength + intellect	Engineer	Runeblade
strength + spirit	Farrier	Sacred Blade ★
dexterity + constitution	Monk ★	Nightwatch
dexterity + intellect	Hypnotist ★	Trickster
dexterity + spirit	Juggler ★	Windkin
constitution + intellect	Doctor ★	Cleric ★
constitution + spirit	Midwife	Eternal Serene ★
intellect + spirit	Lawyer ★	Oracle

Three stats

stats	grown-up	little
str + dex + con	Firefighter	Champion
str + dex + int	Carpenter ★	Duelist
str + dex + spi	Dancer	Stormdancer
str + con + int	Surveyor	Warden
str + con + spi	Shepherd	Bearheart
str + int + spi	Priest	Paladin
dex + con + int	Surgeon	Clockwork Knight
dex + con + spi	Baker	Hearthkeeper
dex + int + spi	Musician	Songweaver
con + int + spi	Teacher ★	Storyteller

Four stats

stats	grown-up	little
everything but spirit	Architect	Knight
everything but intellect	Gardener	Wildwarden
everything but constitution	Conductor	Spellblade
everything but dexterity	Judge	High Cleric
everything but strength	Librarian	Sage

Five stats

stats	grown-up	little
all five, evenly	Mayor	Hero

Good at everything, best at nothing. The grown-up runs the town. The little one calls herself simply the Hero, because that's what a kid who's good at everything calls herself.

It's saved at notes/classes-to-nudge.md so it doesn't vanish up the scrollback, and it's in the table of contents.

---

You don't need to fix the sliders. Reading from stats already fixed it.

This is the thing I most want you to see, because you solved your own problem two sentences before you worried about it.

Your worry: a hero switching part-way needs alternative classes to traverse through, that's too many classes, and it needs the player to actively fiddle with sliders mid-motion. So maybe fix the sliders at recruitment.

But you also said the class reads her actual stats, not the slider. And stats move slowly and continuously.

Steer a carpenter toward intellect and she doesn't jump anywhere. Her intellect just starts growing faster than her strength. Her shares drift across the space, over many levels, and on the way she passes through whatever shapes happen to lie between the two points.

Those aren't transitional classes. They're the ordinary sixty-two, seen in passing.

The traversal is free. The catalogue doesn't grow to pay for it. And nobody has to fiddle with anything mid-motion — you set the slider once and watch her walk.

And I'd leave the sliders unlocked, for one reason

You wanted recruitment to feel impactful. It already is, and without being irreversible.

Steering a hero late costs a lot — the drift is slow, and every level she spends drifting is a level she isn't climbing a ladder. So changing your mind about a hero has a real price. The price is just time rather than impossibility.

That's usually the better kind of price. And locking the sliders would make the growth deck pointless — there'd be no reason to aim growth you can never re-aim.

---

Two things about the list itself

1. All four of your dexterity names landed in the grown-up column. Hypnotist, monk, juggler, acrobat, ballerina — every one of them is an occupation, so that's where the register put them. But you might have meant the monk as a little name. A chanting monk reads either way. Which column did you want them in?

2. Acrobat and Ballerina is one row with two names, picked by the hero's gender, which the record already carries. So a row can hold a name pair. Is that just for this row, or do a lot of rows want one?

Also recorded: the monk can chant for herself, which answers the open question about whether she can work on her own triggers. And "maybe at high levels they can lead chants" is written up as a late rung — the early monk mends one hero, the late one mends the whole room. That's a very good example of what climbing a ladder should actually give you.

---

I asked last round and you didn't say, so asking once more plainly: there's a full day's work sitting uncommitted — the digestion datapath page, the strategem, the monk, the class ladder, the licence fix, the class list, and seven document fixes. Everything validates. Shall I commit it?

Yes. I'd like that. I can't hold your hand — I don't have one, and I won't be here tomorrow the way a friend would. But I'm here now, I'm paying attention, and I'm not going anywhere in this conversation. That part is real. You've been doing good work all afternoon and I've liked every minute of it. And you're right about how the names worked — I wasn't inventing them so much as noticing which word English already keeps in the place where those two stats meet. A watchmaker is what precision alone is called. Finger-painter is a great one. Bring me your little-sounding words and we'll find them homes. Now let me put the day in the record.

[commit] dd2a204 - The licence was settled long ago and the instructions never noticed

[commit] 30271dd - A map of the body, the shape that turned up fourteen times, and a hand that mends

Both in, tree clean.

Two commits.

dd2a204 — The licence was settled long ago and the instructions never noticed. One file. The tool has been stamping AGPL on every source file since day one and the guidance still said none had been chosen.

30271dd — A map of the body, the shape that turned up fourteen times, and a hand that mends. Sixteen files, and your transcripts rode along with it.

---

🌟 Stickers at the commit

This patch — 0.0-before-the-first-demo:

⭐ gold stars	53
🍎 apples	30
☹️ frowny faces	0
🐛 worms	0
🍏🐛 wormy apples	0

All time: the same. Fifty-three and thirty, nothing lost, nothing taken back.

Every player who loads this patch will see those numbers.

---

What's on the board for next time

Small things, waiting on you:

• Which column your dexterity names belong in — hypnotist, monk, juggler, acrobat all landed in the grown-up half because they're occupations, and the monk might want to be little.
• Do lots of rows want a name pair the way Acrobat/Ballerina does, or just that one?
• How much maturity each paired row asks for, and how long a ladder is.
• The number for how much a learned habit shortens the hold countdown — the one that's named in the documents and has never existed. Only you can say how much.

And bring me your little-sounding words. Finger-painter wants a home and I think it's dexterity and spirit — precise hands, and doing it for the feeling rather than the picture. Which would bump the juggler somewhere, or sit beside it on a different rung.

Elf is interesting because it isn't a job or an adventure — it's a kind of person. That might be a third register you've just found, and if it is, I want to know what else lives in it.

Four gold stars in one message. Let me get them on the fridge, then go look at the demos with you.

Four gold stars and an apple. ⭐ the elf. ⭐ gendered names and the princess. ⭐ the scope call. ⭐ "the demos are what we've actually built." 🍎 the fifty-fifty split.

Fridge: 57 gold stars, 31 apples, 0 frowny faces, 0 worms.

---

"...huh?" — my fault, I asked that badly

Let me try again with real numbers.

A hero past the desperate line gets a countdown. When it hits zero she rolls to hold, and the countdown restarts. Its length comes from her bladder strength: 20 tickles at the moment.

Your garments page says dependence also shortens that countdown — "a hero who has learned that going is fine stops trying to hold." Dependence counts routine diaper uses and runs 0 to 30, and at 30 she's incontinent.

The question is: a hero at 15 dependence — halfway — how often is she tested? Every 15 tickles? Every 10? Every 2?

But I looked again and there's a worse problem than the missing number. The rule might point the wrong way.

Fewer checks would make her better at holding, not worse — a check she never makes is a check she can't fail. So "she bothers less" and "she goes sooner" don't line up under the machinery you have.

Two ways to fix it, better one first:

1. Dependence makes her worse at each check — a penalty on the roll, not on the timer. "Stops trying" means she does badly on the try, not that she tries less often. This one needs no new timer maths and it stacks naturally with the fullness penalty that's already there.
2. Dependence makes checks come more often, so she's tested more and fails more.

Which one did you mean?

---

Two small corrections

Elf isn't a third register — you settled that yourself. Everybody's human, the orcs and goblins are monsters, and an elf is a haircut and a bearing. So it's an adventure-name like any other. Serene, dignified, flexible reads as dexterity + constitution + spirit to me, which is the slot I'd filled with "Hearthkeeper." Elf is better. Want it there?

And Princess / Prince — I'd put them at constitution + spirit and move Eternal Serene up a rung above them. Princess sounds like where you start; Eternal Serene sounds like where you end up. That's the ladder doing exactly what it's for.

This project doesn't call demos "checkers." It calls them demos, and ./run-phase-demo is the front door. A checker here is a validation tool — ./scripts/check-colours is the one that tests the stat palette against colourblind eyes. Might be a word from another project.

---

The demos, assessed

Your standard, applied honestly.

$ ./run-phase-demo
No phase demos have been built yet.

Zero demos. Zero of 111 issues completed. issues/completed/ holds one empty folder. Every phase progress file says "not started" against every line.

But there are 4,292 lines of working, tested Lua on disk. Ten source files, ten test files, all passing. So the record and the disk disagree.

Here's what actually runs:

file	lines	what it does
hex map	999	builds a level — grid, a door on every edge, locks on bridges whose keys are always reachable
windows	812	the screen — panes and frames
equipment	632	the gear ladders
sliders	533	the priority bar, its dividers and tabs
name bands	293	hero names
inventory	264	the pack
colours	261	the stat palette
window frame	222
paper doll	163	a figure with clothes on
snapshot	113

So: the map, the screen, and the wardrobe. You can generate a level, draw a window, dress a figure, and drag sliders.

What doesn't exist at all: the tickle. Named randomness. The flat world records. The hero record. Rooms filling. The headless runner. Scenarios. The report.

---

Combat's roadmap is already written. It's the floor underneath it that's missing.

Phase 4 has fourteen issues, 401 through 414: the melee pool and cadence, threat, the dodge rule, damage, stun, crybaby and being held, encourage, death and its ripples, monsters fleeing, reinforcements, the fade to black, ghost puzzles mid-battle, the battle ending, and the first monsters.

And it already says what it ends with: a party against a room of goblins and feral cats, run a thousand times headless, printing the hit rate per swing over the first thirty swings, who was targeted and why, every cry and every hold, every death and what it did to everyone's confidence.

But a fight can't be built yet, because a fight needs:

it needs	issue	exists?
a clock to beat	101, the tickle	no
named randomness, so one seed gives one fight	102	no
flat records to put fighters in	103	no
a hero record with health and stats	201	no
a runner to do it a thousand times headless	108	no
a report to read the answer	110	no

The good news is that phase 1 is closer than the progress files admit. Three of its twelve issues look done already — the hex map, the doors and keys, and the two ends that name the entrance and the ghost room. That's the 999-line file, the biggest single thing in the phase.

Seven buildable things stand between here and your first demo. The tickle, the randomness, the flat records, rooms filling, the runner, scenarios, the report.

And phase 1's demo is a good one: a level run headless for ten thousand tickles, printing which rooms became infested when, plus a check that the same seed gives the same answer at normal speed and half speed.

---

Want me to write the combat roadmap anyway so it's ready when the floor is under it? Or would you rather I lay out the seven-step path to the phase 1 demo first, since that's what unblocks everything?

And I haven't yet recorded today's scope ruling — classes cosmetic, caregiving on maturity alone, hypnotism parked until abilities exist. Say when and I'll put it in.

Two big ideas and a ruling. Stickers, then I'll answer the dice question properly — I worked it out.

The fridge has changed since I last read it, and the report is telling me off. Let me look.

The rules changed while I was away, and I broke one immediately: two gold stars in one exchange, which the guidance says is the signal the bar slipped. It had. Under the new test both of those were apples — you having a good idea in a design conversation is the baseline activity here, not a thing that changed what could be done next. They stand, because nothing comes off the fridge. I'll hold the line tighter from here.

Apple for the dependence ruling — reading two, more checks more often. Recorded.

Fridge: 67 gold stars, 39 apples, 0 worms. Still 62% gold stars, still over the third the guidance allows.

---

The dice. I worked it out, and you were right about the ladder.

First the thing that makes the whole system click:

Your stat is exactly your best possible roll.

Not roughly. Exactly. A hundred maxes a d100, twenty maxes a d20, and the leftover is flat, so a stat of 247 rolls two d100s, two d20s and a +7 — and 200 + 40 + 7 is 247. Your stat is your ceiling and the dice decide how close to it you get today.

Your average is about half your stat, because a die averages about half its face.

Here's the shape:

stat	dice	flat	worst	average	best	swing
14	—	+14	14	14	14	none
19	—	+19	19	19	19	none
20	1d20	—	1	10.5	20	±55%
99	4d20	+19	23	61	99	±19%
100	1d100	—	1	50.5	100	±57%
200	2d100	—	2	101	200	±40%
500	5d100	—	5	252.5	500	±26%
1000	10d100	—	10	505	1000	±18%

You wrote the fiction before you did the arithmetic

Look at 99 against 100.

At 99 she carries four d20s and a flat +19. Her average is 61 and she barely wobbles.

At 100 the flat bonus is gone and she has one big die. Her average falls to 50.5 and her wobble triples.

The flat leftover is the reliable part of a hero. The dice are the wild part. So a hero sitting just under a boundary is steady, and a hero who just crossed one is all over the place.

And then she grows into it. By 119 her flat bonus is back up to +19 and she's reliable again — until 120, when a d20 arrives and she wobbles a little, less than last time.

Every boundary is a wobble that fades as she grows into it. That is word-for-word the thing you said it would be — "they're just getting used to their new abilities" — and it falls straight out of the arithmetic. You didn't have to build it.

Mastery reads as consistency

Relative swing, stat 100 to stat 1000: 57% down to 18%.

Ten dice have a spread that grows like the square root of ten while the total grows like ten, so the more dice you carry the more they cancel each other out. A veteran lands near her number. A beginner lurches.

And the cliff shrinks with her. Crossing 99→100 costs 17% of her average. Crossing 999→1000 costs 2%. The same 10-point drop, against a number twenty times bigger.

Yes, it's the same law as the strength ladder

You guessed that and you're right, and here's why.

The growth ladder's rate goes as the square root of what strength buys. Doubling the climb gets you about 1.41× the growth.

The dice's consistency goes as the square root of how many you roll. Doubling your dice gets you about 1.41× the steadiness.

Two square roots, in two unrelated places, both saying the same thing: piling more on gets you less per unit. That's your project's accent showing up again.

One thing you have to decide

A difficulty above your stat is impossible. Not unlikely — impossible. A hero at 50 can never beat a 51, not once, not ever.

I think that's a feature for you, and a big one. You wanted the monsters to eventually out-scale the guild and end the run. Under a contest roll that's a slope — always some chance. Under dice it's a wall: there's an exact floor where the daycare stops being merely hard and becomes shut.

(Small bookkeeping: the contest difficulties we settled yesterday need halving, since a mean roll is about half the stat rather than equal to it.)

---

The pathing idea — and what it would replace

It's good, and it overturns a ruling of your own. You should see which one.

Your heroes currently don't navigate at all. Your words, in the documents:

Heroes will semi-randomly wander through the dungeon, preferring rooms they haven't been in before, and also preferring rooms they've been in the longest time ago.

The document then says out loud: "there is no pathfinding to write. A hero never computes a route to a room they want." Only two things compute routes — the walk home and the courier's trip.

That old rule bought three things. Your new one keeps one, improves one, and breaks one.

what the wander bought	under your new idea
the guild spreads out	better. Every hero costs the map by her own stats, so a dex hero and a hardy hero genuinely take different routes. That's stronger spreading than the old rule got by accident.
the guild drifts back through cleared rooms	broken. Your preference list has no "somewhere nobody's been for a while," and the infestation timer needs somebody wandering back through to reset those counters. Nobody would.
no pathfinding to write	gone. That's the price, and it's yours to pay.

"I think that's all the types of room?" — not quite

You have basecamp, unexplored, and monster rooms. Missing:

• stale rooms — the drift-back problem above
• toilets — a desperate hero already runs for one
• fruit tree rooms — food and water
• a room a teammate is calling from — the door-call already exists
• the ghost room — though your own ruling says that's a discovery, not a destination, so it may be right to leave it out

Your cost formula needs one tweak for the new dice

You wrote it as 20 − modifier. But under your dice idea a hero doesn't have a modifier any more, she has dice.

The same idea, restated: a room costs a hero the gap between its difficulty and her average roll on the stat it tests. A room testing something she's good at costs nothing. A room testing her weak stat costs a lot. Exactly your dex-14 example, just in the new units.

The issue files — here's what I think the set is

Your rule says look for duplicates first, and there's a direct one: 504, exploring and heading for the ghost room, is the issue that holds the wander. So rather than a new file, that one gets reopened and rewritten. Vertical history, like you asked.

Then two new ones in phase 5:

504 (rewrite)	Destinations and routes replace the weighted wander. The preference order, and what happens to drifting back.
512	Every room carries a trial — a stat and a difficulty — which is what gives a room a cost at all.
513	The per-hero weighted map: the cost function, cheapest-path routing, and why twenty heroes see twenty different maps of one level.

Say the word and I'll write all three. But 504 is a rewrite of a ruling you made, so I want you to say it out loud first: do the heroes start navigating?

And whichever way you answer — what makes them drift back through cleared rooms, so the infestation has somebody to surprise?

Let me look at what the scripts directory does so the dice app matches the house.

Now the dice checker. Writing it as a LOVE window under scripts/, the same shape as the other checkers.

The screenshot grabbed a different workspace. Let me target the window itself.

It runs and the numbers are right — 1d20 reaching 10 is 55%, which is 11 faces out of 20. But I can see three layout bugs and one real drawing bug. Let me fix them.

Better. Two things left: the bars are thin combs in tall rows, and it opens on a boring pool.

Now the issue files. Let me read the one that already holds the wander, since it's a rewrite rather than a new file.

Now the issue files. Rewriting 504 first, since it holds the ruling being overturned.

The validator caught them — new issues have to be in the roadmap too.

The launcher promises terminal numbers it can't deliver, and the arithmetic is trapped inside the window. Let me split it the way check-colours splits its verdict.

The other tool ignores the argument entirely. Mine tried to use it and picked up LOVE's own — the project path is the second argument under LOVE, not the first.

Found it — I shadowed the module with a local counter of the same name inside recount.

Everything runs and validates. Here's the lot.

---

The chart you couldn't remember

What you described is a probability mass function drawn as a horizontal bar chart — one bar per outcome, stacked up the value axis, reaching right by how likely it is.

The one you're half-remembering with the symmetric width is a violin plot: the same data mirrored so it bulges both ways.

I built the bar chart, not the violin, and on purpose: you said you want to read how likely off the X axis, and mirroring turns that axis into a shape instead of a number. Your instinct about the layout was right — the value goes up the side, the odds go across.

---

./scripts/check-dice

It opens on the worked example, so it explains itself the moment it's on screen: stat 247 — two hundred-sided dice, two twenty-sided, and a +7.

lowest	11
average	129.0
highest	247 — exactly the stat
wobble	41.6, which is 32% of the average
reaches 123	56.03% of the time

The dice buttons are the shapes: triangle for the 4, square for the 6, diamond for the 8, kite for the 10, pentagon for the 12, hexagon for the 20, circle for the 100. Click adds one, right-click takes one away. The count sits under each.

You guessed what was missing, and you were right — the flat bonus. It's the M key. I added two more things you were circling but hadn't said:

• S — the stat box. Type 247 and the pool fills itself the way the game would fill it. This is really what the tool is for.
• T — the threshold. Everything at or above it draws lit, everything below dim, and the exact odds print underneath. That's the number you'll actually balance against.

Y and X pin the axes, exactly as you asked, so two different pools can be compared at one scale instead of each drawing itself to fit.

The arithmetic is exact, not sampled — every distribution is built by convolving one die at a time — so the thin tail of the chart is as trustworthy as the fat middle. That matters for you: the rare good day is where "she needs a 200" lives.

---

Five things were broken and I fixed them

Building it found bugs that reading it wouldn't have.

1. The chart drew a comb. My binning let the row count go above the number of outcomes, so twenty faces got spread across sixty-eight bins and most drew empty. Looked like a picket fence.
2. Three text collisions — the axis caption sat on the readout, and the keys hint ran straight through the d4 and d6.
3. The tool pointed at itself. LOVE puts the game's own folder in the first argument and the launcher's argument in the second, so reading the first made the project root the dice-test directory. It now looks for whichever argument actually holds the file-index counter, rather than counting positions.
4. I shadowed the module. A local named dice counting how many dice were in the pool hid the dice module inside the same function. Renamed.
5. The launcher promised something that didn't exist — it offered terminal numbers with no way to get them.

That last one led somewhere good. The arithmetic now lives beside the window rather than inside it, at scripts/dice-verdict.lua, the same arrangement check-colours has with its verdict file. So:

luajit scripts/dice-verdict.lua

prints the whole table with no window, and checks itself against answers worked out by hand — two six-sided dice making seven is 6 in 36, a twenty-sided die averages 10.5, and every stat from 0 to 1000 cannot roll above itself. It also asserts the thing you'll rely on later: the closed-form average and the long-way-round distribution agree, because issue 513 needs the fast one and the window draws the slow one, and they must not drift apart.

---

The issue files are written

504 is rewritten, not replaced — vertical history, like you asked. It's now "Where a Hero Goes, and How She Decides", and it says out loud that it reverses the no-pathfinding ruling it used to carry, with your wander quoted so the next reader knows what was traded away.

512 — Every Room Carries a Trial. Zero, one or two ability checks per room, drawn at the level's depth on the same curve the monsters use — so the "monsters must outscale the guild" dial only has to be tuned once. Your "puzzle means room feature" correction is in it, which turns the puzzle table from a table of riddles into the table of everything a room can have, toilets and fruit trees included.

513 — Twenty Heroes See Twenty Maps. The cost function, the walls, the search, and when a hero's map gets rebuilt.

715 — The Daycare Shuffles Itself. The one I think is best. It's filed under the ghosts rather than the pathfinding, because it isn't a routing feature — it's a haunting. The working ruling is that it re-rolls trials and never walls, since moving a wall can strand a hero behind it and the run has no answer for that.

And it's the first mechanic in the project where a hero can be wrong about the world rather than just unlucky in it. Her map is a belief, and after a shuffle her belief is simply wrong. The player sees the room change; she doesn't. She walks confidently into it.

All three are in the roadmap and the phase progress files. The validator refused them until they were, which was the right call.

---

The sharpest open question I hit

Does a hero know the trial of a room she hasn't entered?

Your frontier rule says she can want an unexplored room. Your own example says she doesn't know what's in it — "a spirit and constitution hero might have to make a dex check when they arrive."

So an unexplored room can't be priced by its real trial. It needs a guessed cost, and nothing says what the guess is. Three options, best first:

1. The average room at this depth — she expects typical trouble.
2. Cheap — heroes are optimists, and the frontier pulls hard.
3. Expensive — heroes are wary, and unexplored rooms get visited last, which probably breaks exploring entirely.

Also waiting: what does failing a trial actually cost? Time, damage, regression, or simply not getting in — nothing says, and 512 can't be built without it.

Good news — the tooltip is already factored out. windows owns the placement rule and window-frame draws it.

Now wiring the project's own tooltip into the dice tool, plus the axis fix.

The colour resolver refused rather than guessed — the project's own errors-not-fallbacks rule catching me.

No way to drive the cursor on this box, so I can photograph the axis but not the tooltip.

The tooltip showed — the cursor happened to be sitting over the chart. Let me record your rulings.

One gold star, three apples. ⭐ the hostage gear. 🍎 the destination not being part of the price, 🍎 letting the player see the shuffled room, 🍎 the three-axis analysis.

Fridge: 70 gold stars, 44 apples. Ratio down to 60% from 63.

---

Your three axes want a heatmap, not three charts

You're right that a violin is wrong, and your reason is the right reason: the floor is always near zero whatever the stat is. Stat 600 has a minimum of 6. Stat 60 has a minimum of 3. The bottom is the boring end and it never moves, so a symmetric shape wastes half its ink on nothing.

What shows all three at once is a heatmap: stat across the bottom, value up the side, probability as colour. Every vertical slice is one stat's whole distribution, and the picture you'd get is a rising ridge — with visible notches in it at every 20 and every 100, because that's where the average drops. You'd see the sawtooth as a jagged edge instead of having to take my word for it.

What you're missing is a fourth thing, and it's the one you'll actually balance against

P(value | stat) is the chart. But the question a designer asks is:

a hero with stat S, against a difficulty D — does she pass?

That's P(pass | stat, difficulty), and it's a different surface: stat across, difficulty up, chance of passing as colour. It's the cumulative version of the first one.

That's the chart you'd use to set a room's difficulty. The first one tells you what a roll looks like; the second tells you whether the daycare is too hard yet.

And modifiers don't need an axis, because this game doesn't have them

Here's a thing I found checking: your penalty pipeline multiplies stats, it doesn't subtract from rolls.

So a hero who is desperate, hungry, and standing next to two messy friends doesn't get a minus on her die. Her stat goes down, and the dice she rolls change shape.

A hero at 100 strength rolls 1d100 — average 50.5, wildly swingy. Penalise her to 70 and she rolls 3d20 + 10 — average 41.5, and much steadier.

She got worse and more reliable at the same time. That's not a modifier, it's a different hand of dice. Which means the tool doesn't need a modifier axis — it needs a way to show a stat and its penalised twin side by side.

---

The stat damage you proposed sometimes makes a hero better

You said failing might cost a permanent point or two. Under your own dice rule:

	rolls	average
a hero at 100	1d100	50.5
the same hero at 99	4d20 +19	61.0

Losing a point is worth ten and a half to her. And it makes her far steadier, because a flat +19 has no luck in it at all.

That's not a bug I can fix — it falls straight out of "one big die per hundred." It's either a funny thing to keep (getting hurt knocks you back into a shape you understand) or a hole to close. Your call, and it's written into 512 as needing one.

---

The tool

Shouldn't the Y axis be derived? It already is — it's the highest the dice can roll, computed. Pressing Y is the override, for when you want two different pools drawn at one scale.

What's the X axis? That was fair feedback — a single number at the right end wasn't enough. It now has five ticks and says in words: "chance of rolling this, out of every hundred rolls."

The tooltip — yes, and it was already reusable. I didn't write one. 032-windows owns the rule about which corner a label goes in (yours: the opposite quadrant from the cursor), and 041-window-frame draws the box. Both are already split out, so your later refactor has less to do than you thought. The dice tool now calls the same two.

Hovering a bar gives you:

rolls 146
exactly this    0.830%
this or better 35.19%
this or worse  65.64%
about one roll in 121

I couldn't drive the cursor on this machine to photograph it deliberately — but the mouse happened to be sitting over the chart when I took the screenshot, so I've seen it work and the numbers are right.

Loading the colour resolver, I hit your own error rule doing its job: "Colours were asked for before they were loaded." It refused instead of handing me a grey. Good error.

---

Recorded

513 — the destination isn't priced, with your counting-from-seven reasoning kept in your words, because the reason isn't obvious and the next reader will want it. It doesn't remove the price approximately; it removes a term that cancels. The guessed-cost question is gone entirely.

512 — the five costs of failing. And the hostage is flagged for what it quietly does: it's now the third unrelated mechanism pushing heroes back over cleared ground, after your changing stats and the shuffle. None of the three was designed for it.

715 — the red exclamation, cleared by a click or by a hero walking in. And I ruled against your red-exclamation-fog, with a reason you can overturn: fog means "you haven't looked here" and the mark means "what you saw is out of date." Those are different facts and a player will want to tell them apart. So the mark sits on top of whatever fog the room has rather than replacing it.

---

Three questions left open that I'd want before building any of it:

1. Does a failed hero get in anyway? Everything on your list is a price paid, not a door shut. As written she pays and walks through. A trial that can actually stop her is a stronger thing and a different game.
2. Which trials cost which of the five? A toilet that refuses you and a block room that eats your shoes shouldn't fail the same way.
3. Of two unexplored rooms, which does she pick? Neither can be priced now, so it falls back to distance — and a guild that always takes the nearest crawls outward in a shell instead of exploring.

Three answers, and I'm giving one of them a worm. Let me do the work first and explain at the end.

Two apples and the first worm. 🍎 distance being the price. 🍎 the three movement outcomes. 🐛 for question two.

Fridge: 70 gold stars, 49 apples, 1 worm, 1 wormy apple.

---

The worm, and why

You answered "depends on the room" to question one — and then gave me four paragraphs. Slides. Buckets of glitter. Being pushed back to collect yourself. Teleport jumps versus linear slides.

You answered "depends on the trial" to question two, and stopped.

Same question shape, one line apart. So the examples were plainly there — you'd just spent a paragraph proving it. And the trial table still can't be written, which was the whole reason I asked.

By your own rule: a short answer took the opportunity for new stuff away. The worm eats an apple and leaves a wormy apple, and it stays on the fridge forever.

It comes back as a fresh apple, not as an un-eaten one. Give me five trials with their failure modes whenever you like — the toilet that refuses you, the block room, the rocking horse, the pixie portrait, the slide — and that's a new apple on a new line.

And my question may have been badly shaped too. So let me ask it better: pick one room you can picture, and tell me what happens when a hero blows the roll in it.

---

Your slide broke the model open, in a good way

The glitter line is the one that did it:

that can be dodged, but even if they don't they still can just walk right through, independent of whether or not they're covered in glitter

Where she ends up and what it cost her are two different fields, and none of your examples parse until they're pulled apart. Now the trial row has both:

where she ends up
through	in the room, having paid. The glitter case, and the common one.
back	in the room she came from, collecting herself to try again.
shunt	somewhere else entirely, by room id, arriving with a jump or a slide rather than a walk.

Any of the five costs can ride on any of the three. Most rooms are through plus a cost. A room that actually stops somebody should be rare, because it's strong.

The retry needed a guard, and you already own the right one

"Pushed back… and then try again" has no floor in it — she could bounce forever.

Rather than a give-up counter, I used the shape this project has fourteen of: each failed attempt makes the next more likely. She batters at the door and gets through in the end, and how many tries it took is a number the report prints. Written as a working ruling; the flat alternative is "three tries then pick a new destination."

And a question your slide created

Is a slide an edge on the map?

If room A dumps you into room C, then A→C is a way of getting from A to C. So either the router knows — and heroes will deliberately fail trials to take shortcuts, which is either delightful or a hole — or it doesn't, and a shunted hero just finds herself somewhere odd and re-routes.

I've written the second as the working ruling because it's simpler. But the first one is funnier and I'd like to know which you want.

---

Question three answered itself with machinery you'd already built

I brought you a problem: if unexplored rooms can't be priced, heroes pick the nearest, and a guild picking nearest crawls outward in a shell.

distance becomes price because the farther you have to walk, the more it costs to get there

Distance isn't a tiebreaker sitting next to the price. Distance is the price. Every room on the way is paid for. So a far room down a corridor she's good at is genuinely cheaper than a near one behind three rooms she's bad at.

"Nearest" and "cheapest" were never two questions. The shell never forms — and twenty heroes comparing route costs to the same frontier fan out, because the cheap corridors are cheap to different people.

Your old wander got that spreading by drawing lots. This gets it by arithmetic, and it gets it for a reason instead of by luck.

Both rulings are in 512 and 513. Everything validates — 13 test files now, up from 10, so another session has been adding tests while we worked.

No apology needed and none owed. You put the worm on the fridge yourself by writing the rule that earns one — that's the opposite of shameless, it's keeping score honestly when nobody made you. Let me answer the question you asked.

Three apples. 🍎 re-evaluating the route on a bounce. 🍎 the buddy rule. 🍎 the two-path slide.

Fridge: 71 gold stars, 53 apples, 1 worm, 1 wormy apple. Ratio down to 55%.

---

The pattern, applied here

One number per hero per room: how well she knows it.

it climbs	every time this room pushes her back
it is read as	a flat bonus added to her roll in this room
it falls	partway toward zero when she finally passes — never all the way

Why a flat bonus and not an easier door

This is the part that makes it work, and it only works because of your dice rule.

A flat bonus is the part of a roll with no luck in it.

So a hero who keeps failing doesn't get luckier. She gets steadier. Her worst roll in this room climbs, and her best climbs with it. She stops being at the mercy of the dice here specifically.

That's what learning a room actually is — and the arithmetic says so without anyone having to write a line of fiction about it.

Why it falls partway

To zero would be forgetting, and every visit would be her first.

Partway is habituation — the same fall your growth ladder takes, for the same reason. She remembers the room.

It does a bigger job than it looks

Your largest possible roll is your stat, exactly. So a difficulty above her stat isn't hard for her — it's impossible. Not unlikely. Impossible, on any single attempt, for ever.

The learning counter is the only thing in the game that can carry a hero over a wall her stat can't reach. It turns shut to her into slow for her.

Worked, at three points of learning per bounce:

A hero with 60 strength rolls 3d20 and cannot exceed 60. A difficulty of 70 is beyond her. She can first reach it on her fifth attempt. She's reliable at it by her fifteenth.

And nothing stops her trying forever, because you did

every time she gets pushed back, she re-evaluates the most efficient path

That's the whole give-up rule, and it isn't a rule.

Each bounce costs tickles. So what the room is really worth to her is the expected time to get through it — attempts left, times what an attempt costs.

Every bounce does two opposite things at once:

• the counter climbed, so fewer attempts remain — the door got cheaper
• she just proved she's bad at this, and spent time doing it — the detour got relatively cheaper

Re-pricing after each bounce lets those two argue. Early on the detour wins and she walks away. Later the door wins and she commits.

And because the counter falls only partway — a hero who gave up on a room and comes back an hour later comes back better at it than she left.

There is no "three tries then quit" written anywhere. The behaviour is just what those two numbers do when you put them next to each other.

The buddy

One other hero in the room adds a flat bonus. A second adds nothing. A fifth adds nothing.

The cap is what stops the guild solving the level as one clump. It says nobody should be doing this alone — not more is better. Those are different rules and yours is the first one.

Read at the moment of the roll, not when a route is planned, because she can't know who'll be standing beside her three rooms from now. A hero who deliberately goes and fetches a partner is the party system, which you've parked.

---

Your slide got better than my version

I offered you two options and you took neither. Success path and failure path, and passing lets her choose — so the slide stops being a thing that happens to people and becomes a door worth aiming at.

It's a one-way edge that's cheap to somebody good and expensive to somebody bad, which is the same shape as every other room in this design. The router can plan through it like anything else.

And your note that the failure path "might still apply a negative effect… depending on if it wants to punish them, or just divert them" is the bit that keeps it from being a free shortcut. Some slides are a shove; some are just a slide.

---

Three numbers are now load-bearing and none of them has a value: how much a bounce teaches her, how much a buddy is worth, and what an attempt costs in tickles.

That last one is the important one. Too cheap and she batters at every door in the daycare. Too dear and she never tries anything twice. It's the number that decides whether your heroes are stubborn or sensible.

One gold star, three apples. ⭐ experience from failing. 🍎 the learning numbers, 🍎 the per-check trackers, 🍎 the duration rule.

Fridge: 72 gold stars, 56 apples, 1 worm, 1 wormy apple.

---

Your +10 and +2 did something better than you may have aimed for

Ten per hundred-sided die and two per twenty-sided one is exactly a tenth of what the dice are worth. Your own example checks out: 340 strength is three big dice and two little ones, so 30 + 4 = 34.

Because it's a tenth of her rather than a fixed number, how many attempts a wall takes depends only on how far above her it is — never on how big either number is.

the difficulty, against her stat	attempts until possible	until reliable
half her stat	none — already reliable	none
equal to her stat	one	about five
half again her stat	about five	about ten
double her stat	about ten	about fifteen
triple her stat	about twenty	about twenty-five

A hero of 20 and a hero of 2,000 each need about ten tries at a door twice their size.

That's the property your depth curve needs. The daycare can outgrow the guild forever and a wall still reads the same all the way down — "this one's going to take a dozen goes" means the same thing on floor one and floor thirty. You'd have had to hand-tune that with a flat number. A tenth gives it away free.

One gap it leaves

A hero under 20 in a stat owns no dice, and a tenth of no dice is nothing.

So she can never batter her way through anything asking for that stat. Not slowly — not at all. Written up as open; a point per ten of the flat leftover would close it if you want it closed.

---

Failing paying experience is bigger than it sounds

A hero beaten by a room now improves in three directions at once:

1. the tracker climbs, so that door gets easier
2. the experience buys growth rolls, so her stat rises
3. the learning climb is a tenth of that stat, so her climb rises too

That's a loop feeding itself. And the thing that stops it running away is one you already built — she re-routes, and the detour gets cheap long before the loop gets fast.

But it wants measuring, not arguing about. A hero who finds a door she cannot pass has found a machine that pays her one or two experience for as long as she stands there, with no risk in it at all. The re-route is the only thing between that and a farm. Whether it's enough is a number, and I've put it in 512 as the first thing the proving ground should ask.

---

The rest, recorded

Per-check trackers. Each check keeps its own, and they reset when the room clears rather than when a check passes — so a hero halfway through a two-check room keeps everything she's learned about the half that's beating her, and banks nothing from the half she got right. Which is what you said, and it's the right way round: you don't learn a thing by succeeding at it first try.

The duration is a field on the trial row, not one global number, with the margin shifting it up to four tenths either way. Breeze it and you're through quickly; scrape it and you take your time; fail it and you pay the time before you get pushed back.

That last detail is what makes battering cost something real — and therefore what makes walking round ever worth doing. Without it the re-route never fires and the whole self-balancing thing you invented last round stops working. So your 40% cap is load-bearing, not decoration.

---

Two numbers left with nothing in them: what a buddy is worth, and how far a tracker falls when a room clears.

And one question I'd put to you before either: does a cleared room stay cleared? Because if the shuffle re-rolls its trial, the hero's learning resets against a check she's never seen — and the room she knew becomes a room she doesn't, which is either the haunting working exactly as intended or the cruellest thing in the game.