Served vs cited: measuring which rules your agents use

A retrieval tool can tell you it sent a rule. It cannot tell you the answer used it. Those are two counts, and this category treats them as one.

The academic name for the gap is the attribution crisis: agents cite a fraction of what they read. The Always-On Agents survey, covering 400 works, lists verifiable provenance and “which stored information influenced a specific output” among the field's open problems. Not a solved dashboard. An open problem.

ETH Zurich's split is the one to keep in your head. arXiv:2602.11988: a developer-written context file moved success 2.4 points, a gain the authors could not tell from noise, and cost about 20 percent more inference. The file was served on every run. Almost none of it can have been load-bearing, or the gain would not be that small. Nobody running that study had a count of which lines were which.

The figures on the home page's ledger are sample data and labelled as such. The claim is the instrument, not a percentage we measured on your company. We do not publish a benchmark of our own. Any percentage improvement attributed to Harbor did not come from Harbor.

If you want the graph itself, it is the same CLI as the first note.