Roadmap¶
This repository now measures one thing only: the diversity a feed actually exposes. What follows is what is missing for that measurement to be complete, in the order in which it blocks.
1. A log carrying a verifiable rank and a label¶
This is the only real lock, and it is not methodological.
Four public logs have been examined. None allows the exposed index to be computed:
| Log | Served rank | Label | Verdict |
|---|---|---|---|
| MIND | no — the order says nothing (\(z = +0.12\)) | editorial categories | served set, not ordered feed |
| Baidu-ULTR | yes (\(z = -206\)) | none | exposure measurable there, the index not |
| Open Bandit Dataset | yes, with propensities | anonymised attributes | ground truth, not viewpoints |
| EB-NeRD | the column exists and does not contain the rank (\(z = +1.05\)) | declared sections | the blind index measurable, the exposed one bounded |
What would be needed: an impression log carrying, per row, the served rank — whose informativeness can be tested — and a viewpoint label from a declared catalogue. Without it, the exposed index will stay bounded and never measured.
2. External validation¶
Twenty-seven corrections, 256 tests, and nobody outside has read this work. The tests check that the code does what is claimed, not that what is claimed is true. This is the one lock the repository cannot open by itself, and it will not open by writing one more chapter. → call for review
3. Two caveats calling for a measurement, not an argument¶
A section is not a viewpoint. The two labelling axes of the only corpus measured agree at only \(\rho = 0.16\). Knowing what the index really measures would take a corpus annotated in viewpoints — by humans, over an explicit catalogue — not by topical section.
Attention severity does not transport from one platform to another. It was measured on a search engine; nothing guarantees its value elsewhere, and the page-effect chapter establishes portability only from one page to another.
4. What was removed, and will not return¶
Three halves have left this repository: the analogy with quantum decoherence, refuted transfer by transfer; the statistical-mechanics formalism it had prompted, whose only own prediction was measured four times with no effect; and the regulatory apparatus — algorithm, floor, access request — hollowed out by its own measurements.
The audit keeps the entire record of their corrections. There is no plan to reinstate them: this repository measures, it does not prescribe.