Codex refills on three overlapping rhythms: the 5-hour rolling window for bursts, the weekly window for volume, and global resets — the famous all-accounts refills that average one every 6.9 days, with everything from back-to-day streaks to two-week silences in between.
The fastest read on "should I ration today" is the radar's pressure gauge: days since the last reset ÷ the average gap, over the full archive. Above 90% you're inside the historical high-probability window — that's the SEND-IT trigger line.
For depth, read by era: the archive divides reset history into nine periods (launch, winter compensation, the March incident cluster, growth, spring tuning, the July spree, the Sol era, milestones & banked credits, and the Astra era) — each with its own trigger logic: milestone-driven, incident-driven, or launch-driven.
FAQFAQ
Why can't resets be predicted on a fixed cycle?
Because they're event-driven: milestones, incidents and launches all trigger them, and the team never works from a calendar. The base rate (a 6.9-day average) tells you where you are, not what day it lands — which is why the radar publishes a pressure gauge, not fake dates.
Does 100% pressure guarantee a reset?
No. 100% only means the historical average gap has been reached; the archive contains stretches that ran well past it. It answers "where are we in the historical pattern", not "what will happen today".
How were the nine eras defined?
By clustering trigger logic and density: the July spree (10+ resets in three weeks) and the winter compensation era were driven by completely different causes. Every event page carries its era; the full breakdown lives on the About page.