Cerevana

Live at cerevana.com - free, no account needed.

Cerevana is an open, local-first Mindbuilding app that unites multiple evidence-based cognitive exercises under one roof. Today that's RRT (relational reasoning), N-Back (working memory - one merged game taking the best of the modern Quad Box protocol and the classic Brain Workshop one), and CCT (Cognitive Control Training - adaptive-interval serial addition); the ambition is one place for the whole training stack. It is a plain static site - no build step, no dependencies, no accounts - and all of your data stays in your browser (localStorage and IndexedDB).

The exercises

RRT

Trains: fluid reasoning - relational integration, logical inference and mental-model building under time pressure.

Relational reasoning training: read a set of premises like "A is north-east of B", hold the relations in mind, and judge whether a conclusion follows. Question types include distinction (same/opposite), linear orderings, syllogisms, and spatial reasoning in 2D, 3D and 4D, with optional transforms, negation, and anti-strategy scrambling. Play untimed, or against a timer with auto-progression that raises premise count and tightens the clock as your accuracy holds up.

N-Back

Trains: working memory - continuous updating, attention control and interference resolution.

One n-back, best of both worlds. The core is the modern community protocol, built to resist the strategies that make n-back gameable: positions on a rotating 3D cube, audio, colors, shapes and unnameable generated images tracked simultaneously, deliberate n±1 interference lures, variable N, and self-paced tally modes where you enter the count of matches instead of reacting to each one - with tunable auto-progression and per-mode levels. On top of it, the classic Brain Workshop mode families: Position/Sound/Color presets, cross-modal combination modes, arithmetic modes (a number is shown, an operation is spoken, and you type the answer against the number from N back), multi-square tracking, a Jaeggi mode reproducing the original-study protocol, and crab / self-paced variants - plus an optional daily level reset and a choice between recorded voices and your browser's speech synthesis. Built on the game engine of soamsy's Quad Box (see Credits); the classic modes reimplement Brain Workshop's protocol without sharing any code with it.

CCT

Trains: cognitive control - divided attention, processing speed and resistance to interference under time pressure.

Cognitive Control Training: the classic PASAT (Paced Auditory Serial Addition Test) protocol. Digits from 1–9 play one at a time - for each new one, answer with the result of it and the digit right before it (addition by default; multiplication, subtraction and difference are also available). The pace is adaptive: answer streaks speed the interval up, miss streaks slow it back down, clamped between a configurable floor and ceiling. Each session starts at the pace the last one ended on, so you pick up where you left off. The in-game HUD shows time (or answers) remaining and the current pace with its trend, a small green "N in a row" under the digit tracks your streak (switchable off), each answer gets an instant ✓/✗ verdict line, and a session ends on a summary card - accuracy, best streak, response times and how far the pace moved. The screen stays awake during a session. Choose from four pre-recorded voice packs and an adjustable playback speed. Core mechanics adapted from tim22dev22's CCT (see Credits).

Roadmap

The goal is to grow Cerevana into one roof for the training methods the Mindbuilding community has converged on. Possible future exercises - no promises, in no particular order - include 3D Multiple Object Tracking, the Posner task, and UFOV. Suggestions and implementations are welcome.

Features

• Main menu with a live overview of the exercises: active profile, totals, accuracy and recent results at a glance.
• Studies library: browse, search and filter the research behind cognitive training, collected by the Mindbuilding community.
• Transfer page: log your scores on an external test battery (fluid intelligence, working memory, general aptitude, personality/mental control) before starting training and again after 4-6+ months, and see each test's baseline-to-latest delta - flagged if retested too soon to be reliable - next to how much RRT/N-Back/CCT training happened in that same window.
• In-app Credits page with the full attribution: exercise lineage, protocol sources and bundled assets.
• Per-exercise profiles, each with its own settings and progress.
• Profile sharing via URL on every game page - copy a share link on one device, open it on another and the profile imports itself. Links always point at cerevana.com so they work for anyone, even when copied from a locally running copy.
• History export/import to a JSON file, with merge (timestamp-deduplicated) or overwrite semantics, covering score, question history, progress-graph data, all n-back games, all CCT sessions and all logged test-battery scores (the current and previous export format import; older ones don't).
• The same four corner panels on every game page: Settings (organized into the same collapsible sections on all three exercises), History (RRT's per-question log; N-Back's per-game list with score chips; CCT's per-session list with accuracy chips), Info (how to play, keyboard shortcuts, credits, resets) and Graphs.
• Progress graphs - one dedicated graph per metric: time spent, average correct times, premise speed and totals for RRT; per-mode level, accuracy, reaction time and daily time spent for N-Back; accuracy, response time and daily time spent for CCT.
• Stats page: every exercise's progress graphs gathered on one page over your full history (the game pages' popups show the recent window), plus a combined minutes-per-day chart across all three exercises - filterable to presets (last week/month/90 days/year) or any custom date range.
• Timers with auto-progression (RRT), adaptive per-mode levels (N-Back), and an adaptive answer interval (CCT).
• Daily and weekly play-time goals for every exercise - progress bars beside the play area on RRT and CCT, HUD and history readouts on N-Back - plus a combined tracker on the main menu that sums the goals you've set across exercises and derives a monthly target from them.
• A calm, focus-first look in both dark and light: one restrained accent, spent only where attention is earned (selected states, the active tab, the timer). The accent hue is yours to choose - a slider in the menu's Appearance section recolours the whole app (both themes, every page) from a single hue, defaulting to the original sage-teal; every shade is derived at fixed saturation/lightness so it stays WCAG-AA readable at any hue. The correct/wrong colours are tuned against a colour-blindness simulation - always paired with a word, never colour alone. Custom background image and sound effects. A main-font picker in the same Appearance section (Oxanium, JetBrains Mono, Zen Dots, or the default per-role mix) applies app-wide.
• Desktop launcher installer for Linux (create-shortcut.sh) that serves the app locally and opens it in its own app window - in Chrome (recommended, gets a real chromeless app window) or Firefox (opens as a normal browser window, since Firefox dropped its app-mode support). Works on any distro with a .desktop-entry-compatible desktop (GNOME, KDE, XFCE, Cinnamon, MATE, etc.) with git, python3 and Chrome/Chromium or Firefox installed
  • not tied to a specific distro. A "Run Locally" panel on the main menu offers a one-line curl | bash install command for either browser, or a plain download of the script.
• Installable as a PWA: any browser's "Install app" adds Cerevana as a standalone window with its own icon (no service worker, so it still needs a connection - this is install convenience, not offline support).
• Mobile-friendly play: on phones the TRUE/FALSE controls become a fixed full-width bar in the thumb zone, panels and graphs go full-screen, landscape scrolls instead of clipping, and the UI respects notches and home-indicator safe areas in standalone/PWA mode. CCT reflows into a single column with a finger-sized answer grid (all 36 multiplication answers fit on screen) and defaults to on-screen keypad input on touch devices so the OS keyboard never covers the game. N-Back pins its match keys in an RRT-style edge-to-edge bar at the bottom of the screen (the corner tabs slide up to sit on the bar, whose height adapts to however many keys the mode needs), scales the board to the remaining space, gets a dedicated phone-landscape layout, adds an on-screen digit pad to the arithmetic modes (previously physical-keyboard-only), keeps the screen awake during a session, and offers an optional vibration cue on match presses. On iOS Safari all game audio (N-Back stimuli, CCT digits and beep, RRT sound effects) is unlocked during the session-start tap so Apple's autoplay policy doesn't silence timer-driven sounds, and the app requests persistent storage so Safari's 7-day inactivity cleanup doesn't evict training history.

Running locally

No build step, no dependencies - it's a static site. Serve the folder over localhost (the History API and IndexedDB behave better than on file://):

python3 -m http.server 8080

then open http://localhost:8080. For an app-like window:

chromium --app=http://localhost:8080

Or, on Linux, run ./create-shortcut.sh from a clone of this repo once to install a "Cerevana" entry in your application launcher that does both - pass --browser=firefox to launch in Firefox instead of the Chrome/Chromium default (the script refuses to run on macOS/Windows). Without a local clone, the same script is self-contained:

curl -fsSL https://cerevana.com/create-shortcut.sh | bash -s -- --browser=chrome

(--browser=firefox also works here) clones the app into ~/.local/share/cerevana-app first, then installs the launcher.

Contributing

Issues and pull requests are welcome. The codebase is deliberately simple: plain HTML/CSS/JS, no build step, no framework - clone, serve, edit, reload. Shared UI building blocks (settings rows, panels, tooltips) are native web components in js/components/ - light-DOM custom elements, one small file each, no compiler involved. Note the license below is NonCommercial (inherited from the Syllogimous lineage), so contributions land under CC BY-NC 3.0 too.

Development happens on dev, which auto-deploys to a preview at cerevana-dev.fly.dev on every push; main is release-only and only moves via pull request. Every push to dev and every PR into main runs the same gate: ESLint, Stylelint, html-validate, a gitleaks secret scan, five pure-logic test suites (tests/*-pure.mjs, covering RRT, N-Back, Quad Box, CCT and the transfer tracker), and a headless-browser smoke test (tests/smoke.mjs) that loads every page and checks for console errors.

Credits

RRT is built directly on Syllogimous: created by 4skinSkywalker, developed into Syllogimous-v3 by ikokusovereignty, and forked and refined by soamsy - soamsy/Syllogimous-v3 is the codebase Cerevana grew from (playable original). giladkingsley is also a credited Syllogimous contributor.

N-Back merges two lineages. It is built on the game engine of soamsy/quad-box (MIT, playable original) by soamsy - the same developer behind the Syllogimous-v3 fork RRT grew from; the engine lives in js/quadbox/engine/ with per-file MIT attribution headers and its license at js/quadbox/LICENSE (full provenance in js/quadbox/PROVENANCE.md). Its classic mode families are inspired by Brain Workshop by Paul Hoskinson with Jonathan Toomim (brainworkshop.sourceforge.net, maintained fork) - Cerevana reimplements the protocol in JavaScript and shares no code with Brain Workshop.

CCT's core mechanics - the adaptive-interval PASAT rule - are adapted from tim22dev22/CCT (MIT, playable original); the derived files live in js/cct/engine/ with per-file MIT attribution headers and their license at js/cct/LICENSE (full provenance in js/cct/PROVENANCE.md). The four pre-recorded voice packs (js/cct/audio/) are vendored from the same source under the same MIT license. CCT's own README credits "EEE" for supplying its original source code.

The menu background photo is by Kieren Ridley on Pexels (Pexels License, free to use), bundled locally as img/menu-bg.jpg.

The favicon and app icon are the Cerevana head logo, created with Canva Pro and used under the Canva Content License; bundled as favicon.png, img/icon-512.png, img/icon-192.png and img/apple-touch-icon.png (the PWA/iOS icons). The social-preview image img/og-image.jpg is the same logo set beside the CEREVANA wordmark (Format 1452, below). Note: the logo is licensed separately from the app - the repository's CC BY-NC license does not apply to it.

The body typeface is Space Grotesk by Florian Karsten, licensed under the SIL Open Font License 1.1 and bundled locally. The header/logo typeface is Zen Dots by Yoshimichi Ohira, the display/HUD typeface is Oxanium by Severin Meyer, and the monospace typeface is JetBrains Mono by JetBrains, and the CEREVANA wordmark on the menu is set in Format 1452 by Frank Adebiaye (Velvetyne) - all licensed under the SIL Open Font License 1.1 (Zen Dots, Oxanium, JetBrains Mono, Format 1452) and bundled locally.

The UI icons are coolicons by Kryston Schwarze (MIT), bundled as an icon webfont in fonts/cool/ (license: fonts/cool/LICENSE-coolicons.txt).

Bundled libraries, all vendored in js/lib/: Chart.js (MIT) for the progress graphs, with date-fns and chartjs-adapter-date-fns (both MIT, license: js/lib/LICENSE-date-fns.txt) powering the Transfer chart's time axis; and - for Quad Box's generated image stimuli - d3-delaunay/delaunator/robust-predicates and d3-shape/d3-path (ISC, see js/lib/LICENSE-d3-stimuli.txt).

License

Attribution-NonCommercial 3.0 Unported (CC BY-NC 3.0) - inherited from the Syllogimous lineage. See LICENSE for the terms and the provenance of each exercise.