Three fixes to how highlights work in conversations — small on the surface, but they rewire the memory of every match.

Your highlights now land exactly where you drew them. Before, the match overlay was showing the whole sentence you picked from — if you underlined three words inside a long line, the overlay would paint the entire line. Now the green (or red) band wraps the exact words you chose, on both sides of the conversation.

Notes are preserved. If you're on Unfair Advantage and you've ever typed a short comment alongside a highlight ("this is why I liked it"), the form was accepting your note and silently dropping it on submit. It was never reaching the other person. Every note you leave now arrives with your highlight, end-to-end.

And your memory survives edits. When you highlight a passage, we now anchor it to the exact version of the manuscript you read. Whenever the passage you underlined is still there, your highlight renders over those exact words. If the author later rewrote that very passage, or revised the manuscript so many times that your version has rolled out of our 10-snapshot history, your highlight doesn't disappear: it shows up in a new section below the current manuscript titled "This is what I highlighted when we first met", with the words you underlined and any note you left. A tender footnote rather than a broken overlay.

New tab icon on anketta.ru and manuscript.anketta.ru, and it now follows your system theme — a light variant when your OS is set to light mode, a dark variant when it's set to dark. Your browser picks automatically. Nothing to configure.

New favicon on both anketta.ru and manuscript.anketta.ru — a peeking variant that replaces the old tab icon. Purely cosmetic; no behavior changes.

The manuscript recommender is getting a serious upgrade. Instead of matching you to people based on broad "interests" like we did before, the app now learns what you actually highlight line-by-line and finds manuscripts with similar phrases. Tap "YES" on "drinking beer" in one manuscript and the app starts favoring other manuscripts that contain that exact semantic idea — not "food and drink", not "lifestyle". The highlights you tap on show up in a new "Your preferences" screen (Settings → Preferences) where you can see each quote, delete the ones that don't fit you anymore, or wipe the whole thing and start fresh. Your taste is entirely your own — the model runs on our servers and your highlights never leave our infra.

The feed also quietly mixes in 15% "fresh air" every refresh so you're not stuck in a bubble: sometimes the newest manuscripts, sometimes the most liked, sometimes brand-new writers with very few swipes. When you see something fresh, we now say so.

A big polish round on how text selection and mobile onboarding feel, pulled together from hands-on device testing. Text selection is now reliable end-to-end everywhere it can happen — swiping, the onboarding practice step, match manuscripts: the dwell-and-drag highlight no longer strays to the line above or below, it snaps to whole words, and it flows around parts of the text that are already marked up, so you can't pile a new amber band onto a blurred word or one that's already liked or struck through. Only one highlight lives on the manuscript at a time, and it never spans across paragraphs. The Yes / No action bubble reliably registers (the old tap-race is gone), and desktop users can drag with the mouse for a live word-snapped preview in the same amber color as mobile. Manuscripts you opened in an earlier session now restore with their highlights and strike-throughs on the correct characters — no more strokes sliding onto the wrong word while fonts load.

Mobile onboarding got quieter and more predictable. The "What should we call you?" step is fully static — tapping the name field doesn't shift the layout; you just type, see suggestions, and the moment you pick one the keyboard drops. On the city step, tapping the field slides the row up so suggestions fit comfortably below, and picking a city also drops the keyboard. The Continue button is now pinned to the real bottom of the page on every step (no more floating above the keyboard), and both the name and city dropdowns show up to five suggestions. The "How reactions work" tutorial finally shows real miniatures of the Like / Pass / Undo icons in its captions instead of emoji. And the practice-bio step got two quality-of-life tweaks: the ghost-demo captions ("Select text" / "Choose your reaction" / "Done") are bigger and higher-contrast, and the floating "Try selecting words you like" tooltip is gone.