Feedback Loops

What it is. A feedback loop feeds a system’s output back into its own input. Positive loops amplify (a lead makes the next lead easier — a snowball); negative loops dampen (falling behind makes catching up easier — a rubber band). Every game with state has both; the design question is which one dominates, and when.

Why it’s fun. A pure snowball decides the match at minute two and the rest is a formality. A pure rubber band erases skill — why play well if the game hands it back? The fun lives in the tension between them: a lead that matters (so building it is exciting) but is never safe (so defending it stays tense). The system-mastery-curve skill-delta only reads as fun when a good lead is real and a comeback is possible.

Why it works

A lead you can lose is a lead worth chasing. If the snowball is uncatchable, the loser stops trying and the winner stops sweating — both channels of tension die at once.
Loops are the engine of pattern-pacing-and-tension. Positive loops build the peak; negative loops manufacture the valley (the comeback beat) without the designer hand-scripting it.
Runaway is a bug you can prove. FUN.md law 2 (skill-delta) has a dark twin: if the winning strategy’s margin grows without bound across a session, a bot-vs-bot sim shows the game deciding itself. Assert the lead is bounded, or that a trailing bot’s win-rate stays > 0.

Levers

Lever	Turns the loop…	Example
Loop gain (how much output re-feeds input)	Steeper positive → faster snowball	Kill → gold → item → more kills (MOBA)
Diminishing returns	Caps a positive loop	Each territory worth less; XP curve inflates
Catch-up faucet	Adds a negative loop	Last place gets the blue shell; comeback mechanic in fighting games
Loss forgiveness	Softens the negative of falling behind	Bankruptcy protection; mercy income floor
Tempo of payout	When the lead converts	Bank-it-now vs compound-later (`pattern-risk-reward`)
Information	Whether players see the loop	An exposed “+N” lets players fight the snowball (`pattern-readability`)

Applied across genres

Genre	Positive loop (snowball)	Negative loop (comeback)
RTS (`genre-rts`)	Eco lead → more army → more map → more eco	Defender’s advantage; cheaper tech when behind
4X / city (`anchor-civilization`, `genre-city-builder`)	Cities fund cities; the classic runaway leader	War-weariness, unhappiness, upkeep scaling
Roguelite (`system-meta-progression`)	Build synergy compounds mid-run	Death resets the run; meta-unlocks aid the next attempt, not this lead
Deckbuilder (`genre-deckbuilder`)	An engine deck draws into itself	Deck bloat; scaling enemy damage punishes stalling
Horde survival (`genre-horde-survival`)	Build growth is multiplicative	Spawn pressure is superlinear (FUN.md §6) — the tide is the negative loop
Racing (`genre-racing`)	Clean line → speed → gap	Rubber-band AI; slipstream helps the trailer
Incremental (`genre-incremental`)	Production buys production	Flat payback ratio (~15–25s) caps runaway (FUN.md §14)

Overdone when…

The snowball is uncatchable — the leader wins on turn 3, the game plays out a decided result. Add diminishing returns or a comeback faucet.
The rubber band is visible and cheap — players see the game handing back their lead and stop caring. Comeback should widen the chance, not erase the gap (the trailer gets a shot, not a gift).
Catch-up punishes winning — if leading is strictly bad (everyone sandbags to avoid the target), the loop inverted. The lead must stay desirable.
Two positive loops with no damper — economy AND military both compound with nothing pulling back; the match resolves before skill can express.
A hidden loop — the snowball is real but the player can’t see it building, so they can’t fight it (pattern-readability). An exposed lead (the live “+N”, the score bar) turns a runaway from a fait accompli into a target.

Verify / feel-gate link

Feedback runaway is a skill-delta pathology (FUN.md law 2): run the intended line and a trailing/null line and assert the trailing side keeps a non-zero win-rate across a session — a lead that grows unboundedly is the bot proving the game decides itself.

Bounded-lead assert. Sim two bots; assert the leader’s margin stops growing (or the trailer’s win-rate stays > 0) — the negative loop is doing its job.
Economy pacing. For incremental/economy loops, the flat payback-ratio window (FUN.md §14, system-economy) is the runaway gate — rising paybacks are an unchecked positive loop strangling the late game.
Determinism rail. All loop randomness (catch-up rolls, blue-shell draws) through world.rng (FUN.md law 7) or the balance sim can’t reproduce the runaway.

Worked micro-example

“An RTS that never feels decided at minute two.” The core positive loop is eco → army → map → eco. Left ungated it’s a hard snowball. Add three dampers: (1) diminishing returns — each expansion costs more to defend than the last; (2) a defender’s-advantage negative loop — losing ground makes your remaining ground cheaper to hold; (3) an information lever — expose the score so the trailer can aim a comeback (pattern-readability). Now the lead is real (building it is the game) but never safe (defending it is the other game). Prove it: two bots, assert the leader’s margin stops compounding and the trailer keeps a non-zero win-rate.

Composes with

pattern-risk-reward — when to bank a lead vs compound it is a risk decision that rides on the loop’s gain.
pattern-pacing-and-tension — loops are the mechanism that makes peaks and valleys emerge instead of being scripted.
system-economy — faucets/sinks are the most common positive/negative loop pair.
system-difficulty-and-dda — dynamic difficulty is a deliberately-added negative loop; tune it so it corrects without erasing skill.