ImprovedJune 10, 2026

Question depth backfilled, coverage bands rebalanced

You can now read the coverage score against a back catalogue where question depth is actually scored. A system-funded pass graded depth on the large majority of past analyses, the score header says plainly when an analysis still has none, and the Shallow-to-Complete labels are re-anchored so they split pages into even quarters again.

Problem

Two things were out of step. Question-depth scoring, how well a page answers the eight diagnostic questions an LLM needs, was new, so most older analyses were scored on concept integration alone, and the score header hinted that depth would be filled in on its own. Separately, the word labels on the score, Shallow through Complete, are meant to place your page against everyone else's, about a quarter of pages per label. Those cut points were set when almost no analysis had a depth score, so as depth scoring spread the real spread of scores moved up and the labels drifted: pages crowded into Solid while almost nothing reached Complete.

Context

This was the known follow-up from the recent rescale work, which flagged the bands as provisional until depth scoring covered the majority of analyses, and flagged the "depth will be added automatically" wording as a promise the system could not always keep.

Why now

A system-funded backfill re-graded question depth across every analysis whose source could still be fetched. Depth-scored analyses went from under one in ten to more than eight in ten. With the real distribution finally settled, the labels could be re-anchored to it, and the header copy corrected for the pages that remain integration-only.

What changed

Three things, all visible without any action from you.

First, most past analyses now carry a question-depth score. A system-funded pass re-fetched and re-graded every analysis it could, taking depth coverage from roughly 9% of analyses to 86%.

Second, the bands were re-anchored to the new, settled distribution. The Complete label now begins a little lower, so the four labels divide pages into roughly even quarters again. The only pages that change label sit right at the Solid/Complete boundary, and their underlying score does not move.

Coverage band distribution before and after the rebalance. Before: Shallow 21.5%, Developing 29.1%, Solid 35.6%, Complete 13.8%. After: Shallow 21.5%, Developing 29.1%, Solid 27.6%, Complete 21.8%. Only the Solid and Complete labels move.

Third, the score header now tells the truth about the analyses that still have no depth score. Where a page's source could not be re-fetched, the depth step reads not scored and points you to re-analyze, instead of promising an upgrade that will not arrive on its own.

The coverage score header for an analysis whose question depth was not scored. It shows the retrieval role and boundary counts, expected concepts present, concept integration, a Question depth step reading not scored because the source could not be re-fetched, and a 67% Solid coverage score.

Nothing else in the header changed, and the scoring formula itself is unchanged: only the word label near the top boundary and the depth step's wording moved.