Sim — the simulation of FIRST LIGHT: RESOWN

game.rules.Sim is the whole run: the day cycle, the plot state machine, the zyme's well-up and spread, leavening, the beam's burn, the seed economy and the win/lose conditions. It is the only writer of FL.state. Every other slice reads that table and reacts to FL.events.

Nothing here touches an entity, a material or a screen. Rules.component (mounted on the game anchor) hands it the frame's elapsed time; the rest of the game calls the functions below.

local Sim = require("game.rules.Sim")
local FL  = require("game.FL")

Sim.start()                                    -- a fresh run, the first night
Sim.beamAt(x, z, dt, Sim.beamPowerFor(FL.state.phase, FL.state.phaseT))
local did = Sim.act(col, row)                  -- "sowed" | "harvested" | nil

The API

Two things a caller has to get right

Power is the aim factor, not the sun. beamAt multiplies the power you pass by the phase's own gate (PHASE_BEAM), so a beam at night burns nothing no matter what you pass, and a day beam is a quarter as fast as a first-light one. Pass Sim.beamPowerFor(phase, phaseT) — or your own 0..1 alignment factor on top of it. Do not pass FL.sunHeightFor directly: that function answers where the sun is (−1..1, and only ~0.3 at the top of first light), not how hard the plate burns, and handing it over makes the beam roughly three times too weak.

tick sub-steps. It advances in slices of 0.1 s, so Sim.tick(30) in one call produces the same run thirty seconds of frames would — a test can reach a later dawn without waiting for it, and a frame hitch cannot skip a well-up.

The day

night 40 → firstlight 30 → day 40 → dusk 10, from FL.PHASE_LEN. dayIndex rises as first light ends, which is what a dawn survived means; reaching FL.DAWNS_TO_WIN (5) there ends the run with won, at 550 s of a clean run.

The plot state machine

bare ──sow(1 seed)──▶ sown ──grow──▶ crop ──grow──▶ ripe ──harvest──▶ bare  (+2 seed, +4 leavened)
any ──bloom──▶ zyme ──beam──▶ scorched ──sow(1 seed)──▶ sown

• A plot turning to zyme loses whatever grew on it — that is the cost of letting a terrace go.
• scorched soil is where the bloom spreads fastest (SPREAD_SECONDS_SCORCHED, half the normal interval), so burning ground buys the window and sells the next night.
• Growth runs while the sun is up (first light, day, dusk) at GROW_SECONDS (55 s) from sowing to ripe: sow in a day and lift it in the next.
• Leavening marks sown, crop and ripe plots that sit beside a bloom through a night. The design's table names crop/ripe; sown is included here so a plot deliberately sown next to a bloom is a real gamble from the first dusk. The double yield only ever pays at ripe, so nothing else changes.

The shrine

FL.state.shrineThreat climbs by the share of row 1 carrying bloom, and a row 1 fully held for THREAT_SECONDS (40 s — one night) is 1.0 and a loss. Clearing row 1 walks it back over RELIEF_SECONDS (25 s). The shrineThreat event fires when the value has moved THREAT_STEP (0.02), and always at the ends, so a HUD meter can follow it without a per-frame event.

The opening hillside

M.OPENING dresses rows 1..6 at start(): crop, ripe, crop, sown, bare, bare, uphill first. The hill reads as a farm from the first frame, the ripe row teaches harvest before the first night ends, and the bare lower terraces are where the zyme arrives. Seeds start at 6 (FL.resetState).

Determinism

The well-up and the spread draw from a small LCG seeded to a fixed value in start(), so two runs of the same inputs produce the same hillside. A reproduction is a reproduction.

Measured

• A full-grown bloom under a full-power first-light beam: 2.02 s.
• One untouched night leaves 9 blooms on the lower terraces.
• A player who never acts loses at 224 s, in the day of the second dawn.
• A player who burns the nearest bloom each first light and works the soil each day wins at 550.4 s with 83 harvests.