Punishment must scale with severity and repeat offences

What it is

Punishment has two axes: behaviour severity (threats to persons > harassment > griefing > minor violations) and recidivism record (first offence versus repeat), with sanction strength climbing a ladder along both: a minor first offence may earn a warning or short mute, while severe or repeated behaviour escalates to long bans. Proportionate punishment is necessary from both directions: too light gives severe behaviour no deterrence (a three-day ban means nothing to a cheat seller), too heavy makes minor offences feel unjust (a permanent ban for a first-time argument triggers community backlash).

Why it happens

The proportionate structure's deterrence logic balances certainty and intensity: deterrence = detection probability × punishment strength. Detection probability never reaches 100% (relying on reports and detection), so the punishment ladder makes repeated behaviour face escalating intensity (first offence a reminder, second a cost, repeat exclusion), and the escalation itself signals "the system records your behaviour history"—a signal whose deterrence for rational actors rivals any single heavy punishment. Recidivism's technical base is cross-account behaviour history linking (the same person's multiple accounts, ban-evasion alt accounts); inaccurate linking breaks the ladder (new account resets everything). Proportionality's fairness-perception dimension matters equally: community punishment debates ("banned for saying one word") focus on disproportion, and a transparent sanction framework (which behaviour maps to which tier, published rules) makes outcomes predictable—predictable punishment draws far less controversy than black-box punishment.

Where it stops holding

The ladder's start and end need per-behaviour-type settings: threats to persons, anything involving minors, and profit-driven cheating start heavy (no "first-offence warning" tier), while verbal conflict starts light with room to correct. Recidivism's time window affects fairness: lifetime counts that still stack decades-old minor violations deny reformed players a return path; a sliding window (the past year's record) is sounder. Punishment's purpose is behaviour change, not revenge—the ladder should include a way out (return after sanction completion, demotion for good standing); a ladder with no exit turns punishment into exile, and the long-term community cost (hostility, retaliation) may exceed the sanctioned individual's own harm. An appeals mechanism is proportionality's error-correction component: wrongly punished accounts (behaviour during account theft, malicious-report-induced bans) need appealable, time-bound review—proportionality without appeal has no self-correction when errors happen.

Applying it

• Build a punishment matrix of behaviour tiers × recidivism ladder: behaviours graded by severity, each tier mapped to first/repeat/persistent sanction levels, with the matrix published in community rules.
• Link enforcement with appeals (sanction notices include an appeal entry), and suspend non-severe sanctions during appeal.
• Verification: track recidivism rates per sanction tier—high reoffence after a low tier means insufficient deterrence, while anomalously high appeal success at high tiers flags misjudgement and a standards review; monitoring recidivism and appeal rates together calibrates the whole matrix.

Related

• Same group: W9.05.1 Reporting must be simple enough to use mid-match · W9.05.2 Blocking takes effect client-side instantly, not after review · W9.05.3 Report outcomes need feedback the reporter can perceive
• Nearby: O2.02 Appeals and transparency mechanisms · W9.04 Disruptive behaviour · P1.04 Fairness perception
• Search terms: graduated sanctions · punishment system · ban policy · moderation fairness