Session Structure

What it is. The containers play nests into — a moment, a level or encounter, a run or session, and the meta arc across them — and the length and shape of each. Session structure decides “how long is one sitting, what does it contain, and how do sittings add up?” A 3-minute Vampire Survivors push and a 40-turn Slay the Spire climb are the same genre-family with wildly different session shapes.

Player fantasy / why it’s fun. A satisfying unit of time. A good session has a clean arc — start, build, climax, resolution — that fits a real sitting and leaves a hook. The pull is closure-plus-appetite: “one more run” works because each run finishes something and promises something.

When to use / when NOT

This is a framing system — every game has a session shape whether or not you design it. Reach for it explicitly when:

Design it deliberately when	It’s near-trivial when
Runs/campaigns/roguelite structure is core	A single continuous level (endless arcade)
You’re deciding meta vs. per-run persistence	A one-sitting puzzle box
Session length is a selling point (coffee-break vs. epic)	—

The nesting

Tier	Span	Owns	Ends when
Moment	seconds	one verb → feedback	the input resolves (`process-core-loop`)
Encounter / level	1–5 min	one fight or room	cleared / failed
Run / session	5–60 min	a sequence of encounters + choices	win / death / objective
Meta / campaign	many runs	persistent progress, unlocks, story	the long goal (`system-meta-progression`)

The two big shape decisions:

Decision	Options	Consequence
Session length	coffee-break (3–10 min) · sitting (20–60) · epic (hours)	sets stakes per run and save granularity
Persistence	fully reset · roguelite carry-over · fully persistent	resets make death cheap; persistence makes it heavy

Tuning levers

Lever	Effect	Watch for
Run length	stakes & commitment per session	too long = quitting mid-run loses progress; own it in `system-save-and-checkpoint`
Encounters per run	the run’s internal pacing arc	too few = no arc; too many = a slog
Meta gain per run	how much a failed run still earns	zero = punishing; too much = trivial
Hook strength	the pull into the next session	none = the loop doesn’t restart

How it wires to Hayao

A run is a seed + a snapshot. Because state is pure JSON in world.state with world.rng, one run is fully described by (seed, tuning, inputLog) and restorable from a snapshot — this is exactly the Workshop session artifact (docs/WORKSHOP.md, FUN.md law 7). Session structure and the replay artifact are the same nesting.
Runs are generated. A roguelite run is a generateLevels/generateDungeon sequence over a per-run seed (system-procgen-design, src/content/generate.ts) — the campaign that ships is a list of seeds.
Campaigns compose. composeCampaign(spec) (grep docs/API.md, src/content/campaign.ts) assembles proven levels into an ordered arc — the meta tier as data.
Between-run persistence is meta-progression saved across sessions (system-meta-progression, system-save-and-checkpoint).

Fails when…

Length ≠ save granularity. A 45-minute run with no checkpoint punishes anyone who has to stop — respect the player’s time (system-save-and-checkpoint).
No arc per session. A run with no build-and-climax is texture, not an experience (pattern-pacing-and-tension).
Failed runs earn nothing. In roguelites, a death that advances nothing feels like theft — grant meta or knowledge (anchor-hades).
No hook. A run that finishes and dead-ends stops the “one more” loop.
Deserts in the meta. Long stretches with no unlock or new goal (FUN.md §14 unlock-desert lint).

Verify

The whole arc paces: balance-sim the run/campaign; assert pacing windows and no unlock deserts across the arc (FUN.md §14).
A run is reproducible: golden-hash a full scripted run; snapshot→restore→hash round-trips (FUN.md law 7; docs/WORKSHOP.md).
Every level in the campaign is winnable: solver proof per level via the generator (FUN.md §10, src/content/generate.ts).

Composes with

system-procgen-design — a run’s content is generated from its seed.
system-meta-progression — the tier that persists across sessions.
system-save-and-checkpoint — session length dictates save granularity.
pattern-pacing-and-tension — each session tier needs its own arc.
system-difficulty-and-dda — the curve spans a run and the campaign both.