Report what a study cites instead of scoring it, and fix the phone

[Varnasr]

The badge on all 204 cards said the same word

CLAUDE.md already records that evidenceTier() returns strong for every study, because it reaches refs >= 4 before consulting anything else and every study carries four or five references. A constant is survivable in a data file.

It was on the page. Printed on all 204 cards, and drawn on study.html as a filled meter reading "Strong Evidence Base" at 80 or 90 per cent. A bar at 80 per cent implies a measurement against a maximum. There is no maximum and nothing was measured.

The two signals it short-circuits past are not constants

evidenceTier() computes hasRCT and hasMeta and then returns before either is used. Measured over the library:

signal	studies
cites experimental or quasi-experimental work	61 of 204
cites a systematic review or meta-analysis	7 of 204

They are computed into master-list.json by a new evidenceSignals() and reported rather than scored.

• Card footer: the reference count (4 or 5, so it says something) plus a marker for the stronger signal — Experimental · 5 sources · 5 min read in place of strong · 5 min read.
• Study page: the meter is gone. It states the cited-source count and, where true, that the study cites experimental work or a review.
• New sidebar facet — Evidence. Filters on both signals, with counts. Both selected means both, not either: a reader asking for experimental work and a review wants the studies carrying both. Deep-linkable like the other facets (?evidence=experimental returns 61 of 204, verified).

evidenceTier() is untouched and still returns strong for all 204. What counts as strong evidence here is an editorial judgement about the library rather than a bug to patch — CLAUDE.md says so and lists the three honest options. This changes what the interface shows, not the field. When you settle the question, that is the one function to change.

Three phone defects, measured at 390×844

The site had no name. .site-logo span:last-child { display: none } below 480px left a blue square in the corner and nothing else. It shortens to "Dev" now, which the markup carries in its own span.

The welcome panel filled the whole first screen. Over 600px of explanation before a single study, so a visitor landed on a help panel rather than a library. Its feature grid is hidden and its description clamps to three lines below 700px. Two study cards are above the fold now, and nothing was deleted — the panel is still dismissible and the text is still there.

The card footer put "Read study" and "5 sources" each across two lines, because a link and three metadata items were competing for one row. The link takes its own line below 460px.

Checks

npm run validate — 204 studies clean. Headless Chromium at 1280×900 and 390×844 on the index, a study page and the mobile index: no page errors, scrollWidth == clientWidth, the facet returns 61 of 204 and writes its URL parameter, and the word "strong" no longer appears on a study page.

Not done, and why

New case studies. You asked about fresh content and I have not added any. A study here is ~950 words with five real academic citations; writing five of them properly means verifying twenty-five sources, and I would rather scope that as its own piece of work than produce plausible-looking references at speed in a library whose entire value is that its citations are real. Say the word and I will do it deliberately, in a batch, with every citation checked.

A full visual redesign. Looking at it, the index is dense rather than severe, and the study page is genuinely well built — breadcrumb, serif title, evidence summary, key-data rail, "How to Think About This". The changes here quieten the metadata that was competing with the content and fix the phone. If you want a different look rather than a tidier one, that is a design direction to agree first rather than a change to push.

🤖 Generated with Claude Code

https://claude.ai/code/session_01MrvR2NXsJFVRCeJZFCPuNL

---

Generated by Claude Code