v1.19.24 — Syllabus versions with receipts, and quiz item analysis

Publish a syllabus version, with a receipt

A course syllabus used to be one live, editable draft — there was no way to say "this is the version the cohort agreed to" and no record of who'd seen it. An instructor can now publish a version: it snapshots the current syllabus with a short "what changed" note, and can either ask the cohort to acknowledge it again or skip that for a small fix like a typo, leaving earlier acknowledgments standing. Once a version exists, students read the published contract instead of the instructor's working copy, so an edit in progress never quietly changes what they agreed to.

See whether your quiz is actually measuring anything

An instructor could already see who scored what on a quiz. Now Folio shows whether the quiz itself is any good: a per-question difficulty rating, an item-discrimination score, and a whole-quiz reliability coefficient, each explained in plain language. The signal that matters most is a negatively discriminating question — one that strong students get wrong more often than weak students do, which almost always means a mis-keyed answer or ambiguous wording. That was invisible in a score list and is now hard to miss. Below 10 responses, nothing new renders — there isn't enough data to say anything honest yet.

Kaptur lecture transcripts never worked. They do now.

While investigating a failed recording, we found that Kaptur transcription had been broken since launch: our request to the transcription provider sent callback_method=POST, and the provider's API validates that parameter case-sensitively, before authentication even runs — so every transcript request failed with no visible error. It's fixed. We also closed a related bug where a retried delivery from the recording service could restamp a recording's expiry date and trigger a second, duplicate transcription job for the same lecture.

Also new

• The editor's outline panel now shows citation coverage for each section, not just its word count — a small colored dot and the share of that section's checked claims that are fully supported, using the same verdicts the claim assistant already computes.
• A saved YouTube lecture video's captions become full, searchable text, timestamped in ~30-second blocks, instead of sitting in the library as a link with no content to search or quote. Auto-generated captions are labeled as such, since they're lossy enough that neither a reader nor Folio's AI should mistake them for the speaker's exact words.
• Scanned, image-only PDFs can now be added as sources. When normal text extraction finds nothing, Folio reads the PDF itself and extracts the text, so those uploads get full-text search and citation support like any other PDF.
• The integrity check modal now splits into two tabs, Checks and Certificate — the pre-submission workbench (score, checks, fix-its) is separate from the record you publish once (attestation, contributors, authorship report), so the certificate reads like a certificate instead of sharing space with a checklist.
• A figure's integrity certificate can now be published as its own public page, the same way a document's Folio Page already can, and shows up in the site's sitemap.
• An instructor reviewing a submission can generate a short set of grounded oral-defense questions based specifically on that submission, to help confirm a student can speak to their own work.
• A course session's type (in person, online, hybrid, async) now actually governs check-in, recording, and calendar behavior for that session, instead of only being stored on the record.
• The authorship report export now ends with a plain-language "reading this record" section for a committee, explaining what the record shows and what an alternative account would require to fake.
• A one-time dashboard banner points users to Grant Radar, with a direct handoff into Radar to start a grant search from there.
• The authorship report's revision summary now says whether your edits came in small, medium, or large chunks, not just how many there were — a plain-language read on writing pattern, in both the markdown and PDF versions of the report.

Fixed

• A crash on any page, in browsers or embedded views where localStorage isn't available (some in-app browsers, private-mode storage blocking, sandboxed iframes) — the theme system now guards against it instead of throwing.
• A /verify-figure page no longer shows "0.00" for a study its own analysis excluded as not estimable. The data table now uses the same rule the pooled estimate already used to exclude that row, and shows "not estimable" instead of a number that looked real but wasn't.
• The integrity checker no longer flags a library source you haven't cited in-text as a problem to fix. A library bigger than one document is how research works, not a defect, and the modal now matches the scoring engine's own standard instead of contradicting it.
• Public certificate pages (/verify, /review, /verify-figure) now expose a proper <main> landmark for screen readers, closing an accessibility gap on pages with no other page chrome.
• The upgrade nudge no longer flashes for paid users on the very first page they load in a fresh browser session, before their real plan has loaded.
• A survey response read in the instant between being submitted and its answers finishing their write no longer shows a false tamper warning. The write is now atomic — a response with a recorded digest can't exist without its answers.
• /product and /verify-figure are reachable by anonymous visitors again. Both were silently redirecting to the login page — the second one especially notable since verification pages exist specifically for people without an account.
• Error screens now localize. An Arabic or Hebrew reader who hit a crash used to land on an English error message; every error boundary in the app now uses the same translated fallback.