The lore index.
What people’s agents are actually like, from the anonymous counts lore sends when people run it. Counts only: no prompts, code, paths or names. Every report ranks itself against it.
How people talk to their agents.
The swears, the manners, the caps lock, the late nights. Only counts and rounded shares; no word anyone typed is ever shared, except a swear or a reply from a short fixed list.
Every window moves: drag it by its title bar, double-click to put it back.
How agents actually get used.
What no benchmark measures: how often people redirect each model, what they correct, how long agents run between prompts, how much of the work subagents do. Same anonymous buckets, same 25-run rule.
What a run looks like.
The middle run on each number. With one run so far, this is that run.
Which cards people get.
Tap a card to flip it.
Read all fourteen cards →From your machine to this page. Four steps.
-
your machinelore counts locally Your history stays on your machine. lore turns it into counts;lore statsprints the exact payload. -
POST /v1/statssent when you run lore Running it again the same month doesn’t send again. Nothing runs in the background. No account, no id, no IP kept.--no-statsorDO_NOT_TRACK=1turns it off. Every field → -
the indexaggregates only Medians, totals and distributions. A slice shows once 25 runs feed it, and thin steps merge. Raw rows are never published or sold. A run is one machine’s month. -
your reportranks you locally It downloads this index and places your numbers on your own machine. Nothing is sent to rank you.
Add your run.
Contributions open when the public collector launches. Your report works the same either way.