# Sink values and input efficiency — method notes

Data 1.2.3.1 · export 2026-06-25 · calculated 2026-10-27.
This is an original analysis of the SCIM/game-data snapshot (satisfactory-calculator.com en-Stable.json, from the game's Docs.json), not firsthand gameplay testing.

## What the download contains

[Per-item summary results (JSON)](sink-1.2.3.1.json) contains exactly these top-level keys:

- `meta`: version, export/calculation dates, source identity and SHA-256, caption, Node and solver versions.
- `method`: normalization, objective definitions, recipe-mode flags, solid-resource definition, runtime versions and exclusions.
- `sources`: official-wiki URLs for the mechanics and editorial labels.
- `schema`: one-line definitions of the public keys and their optimization objectives.
- `rows`: every exported item’s `name`, `points`, availability `status`, `rawOptimal` and `solidOptimal`. Both objectives have `standard` and `alternate` metrics. Unmodeled rows also have a row-level `notes` string: “unmodeled: no complete chain from modeled raw inputs; not unsinkable”.
- `scenarios`: each `id`, `name`, `allowedRaw` resource identifiers, `standardEligibleCount`, `alternateEligibleCount`, `standardWinner` and `alternateWinner`. Each winner has `name`, `points`, and both recipe modes under `rawOptimal` and `solidOptimal`, computed within that scenario’s restrictions.
- `powerOptimal`: power-ranking entries with `name`, `points`, and independently power-optimal `standard` and `alternate` metrics.
- `earlyPicks`: curated entries with `name`, `points`, `unlock`, wiki `source` and standard raw-optimal metrics.
- `earlyThroughput`: one standard Cable machine’s `name`, `building`, `clock`, output `rate`, `points` per item and `pointsPerMinute`.
- `couponRules`: `firstGroupCost`, `groupSize`, `multiplier`, `offset` and `cappedGroup` for regular coupons.
- `couponExample`: standard Reinforced Iron Plate `name`, `rate`, `points`, `pointsPerMinute`, `coupons` (each coupon’s `n`, `points` and `minutes`), `capPoints` and `capStarts`.

Every plan contains only these scalar metrics: `rawPerItem`, `pointsPerRaw`, `solidPerItem`, `pointsPerSolid`, `powerMW`, `pointsPerMW`, `buildings`, `pointsPerBuilding`. Each metric measures the plan identified by its enclosing objective; a raw-optimal plan’s power efficiency is not a power optimum. Null denotes an unmodeled chain or an undefined ratio, as explained by the method and row notes.

The public files contain computed summary results and these method notes only. They do not distribute the game-data export, recipe lists or loads, raw-input vectors, byproduct vectors, or site code. The ingredient quantities in the examples below are short hand calculations, not a downloadable recipe dataset.

The recipe rates come from the game data; you can re-derive any row with a linear-programming solver by following the model below.

## Model

1. Filter out extraction/self recipes and any recipe using a forbidden raw input. Admit recipes only when all ingredients are reachable from the allowed raw set; repeat to closure. This excludes chains needing gifts, harvested biomass, remains or nuclear waste as external inputs.
2. For each target, request 1 item/min. Nonnegative continuous recipe machine counts must balance every non-raw ingredient and produce at least the target output. Optimize standard-only and alternate-enabled recipes independently. A scenario constrains inputs, not unlocks.
3. **Points per raw-input unit (water counted):** minimize gross raw consumption with weight 1 for every input in sat.raw. Sum solid item quantities and fluid/gas quantities in their exported units. This is an equal-weight index, not physical mass, scarcity, net extraction or a map-wide production ceiling. Divide points/item by that minimum.
4. **Points per mined solid resource:** rerun the optimization with Water, Nitrogen Gas and Crude Oil weighted 0, all other raw inputs weighted 1. Crude Oil is excluded because it is a liquid. These fluids remain in recipes and in the reported input rates. Divide points/item by the minimum gross solid consumption. A feasible zero-solid chain has a null ratio, not a finite efficiency or an unsinkable status. This column is a separate optimum, not the first plan with water subtracted.
5. No raw output is credited, including recycled water. Gross process water is not net make-up water. Non-raw coproducts can satisfy other recipes; leftover surplus has no sink-point credit. The public objective families are recorded separately: `rawOptimal` minimizes the full index, `solidOptimal` minimizes solids, and `powerOptimal` entries minimize production power. Each contains standard-only and alternate-enabled results.
6. Production MW = sum(recipe pwr × machine count) at 100% clock equivalents. Buildings = sum(machine counts), without rounding to whole machines. Points/min/MW and points/min/building divide target points/min by those totals. Exclude extraction, transport, generators, Sink power, amplification and overclocking. The power table does not optimize building count. Tied resource optima may have different power usage.
7. All point values and recipe rates come from the snapshot. Coupon constants and the small curated unlock/status layer come from linked satisfactory.wiki.gg entries, checked via indexed pages on 2026-10-27. No unlock coverage is inferred for the complete export. The early shortlist ranks only its selected items using standard recipes, not all items available by a given milestone.
8. Preserve full precision internally and in rankings. Public plan scalars retain 12 significant digits to keep the download below 150 KB; page values use their displayed precision. Exported item points remain exact integers. Small floating-point tolerances apply to recipe balances. An unmodeled chain has null metrics and an explicit note, while its exported item value is retained.

## Hand-checkable examples

**Circuit Board, standard ingredients.** One board needs 2 Copper Sheet and 4 Plastic. The sheets need 4 Copper Ingot, made from 4 Copper Ore; the Plastic needs 6 Crude Oil. Thus 2 Copper Sheet = 4 Copper Ore and 4 Plastic = 6 Crude Oil. At 696 points per board, gross raw efficiency is 696 / (4 + 6) = 696 / 10 = 69.6; mined-solid efficiency is 696 / 4 = 174. Surplus Heavy Oil Residue receives no sink-point credit. These match the Circuit Board’s standard metrics under `rawOptimal` and `solidOptimal`.

**Reanimated SAM, standard production power.** A 4 MW Constructor consumes 120 SAM/min and produces 30 Reanimated SAM/min, each worth 160 points. For 1 item/min, the machine equivalent is 1/30, so power is 4 × (1/30) = 0.1333… MW. Using the unrounded value, 160 / (4/30) = 1,200 points/min/MW, and 160 × 30 = 4,800 points/min/building. These match the standard `powerOptimal` entry. Extraction and the Sink’s own power are excluded.

## Verification

Site tests hand-check basic standard chains and Cable throughput, coupon-cost boundaries against the wiki formula, monotonic rankings, all 138 lookup rows, scenario and power table values against their public objective keys, FAQ/schema parity, the public key restrictions and that this download matches freshly generated results. Full exported HTML size is checked when the editor’s output exists; article rendering is checked without running a build.
