The Game Commons platform has two halves. The archive is what everybody shares: cards, playing fields, players, documented sessions and artworks. The workshop is where you make and play your own things. Right now everything you make is kept in your own browser, so nothing leaves your computer until accounts and a shared server exist. Back up from My gallery now and then.
Browse the archive
- Cards lists every prompt. Filter by type (challenge, intervention, wildcard), tag or edition. A card page shows its text, its source, the artwork, who wrote it, which sessions used it and the recorded performances of it. A card can have several artworks: click the left or right side of the card to swap between them, or click a small picture below it, and read who made the one you are looking at, its licence and its source beside the card. Upvote the one you like best, because the most upvoted is shown by default. See prompts and artwork. The star saves a card to your gallery.
- Playing fields shows the battlegrounds, grouped under their buzzwords, with the moves played on each.
- Players lists people who have played in a documented session, with their sessions and recorded moves. Only people who actually played appear. A profile with a photo and a short text exists only if the person agreed to it.
- Sessions are the documented games. Open one to watch the recording and read every move with its card, field, side, score and, where there is one, the transcript line at the right time.
Make your own
Open Make. Everything saved there lives in this browser.
- Card maker. Write a prompt, choose its type and add tags and a source. You can add an artwork. Choose who sees it: only you, friends (you download a share file and send it to them), or the commons (moderators review it first). If the card has artwork and goes to the commons you can ask for support in BREAD and say what it pays for. The goal is your rate. You also choose your share of any support that comes after the goal is reached, from 1% to 50%. While you write, it tells you whether something like your card is already in the archive, and if nothing is, that you have found a gap. A card can say what current event it answers.
- Deck builder. A deck holds one type of card. Choose from the commons and from your own. It needs enough cards for the board it will be played on, and the builder tells you how many are missing.
- Board builder. A board is a set of wedges. Each wedge has a buzzword and two playing fields. Boards of 2, 3, 4 and 6 wedges are supported, and every field in a board has to be different.
- Edition builder. An edition ties together two team names, a board and three decks. The checks show what is still missing. A complete edition is playable. You can say which version it is based on, see how much of it is your own (half makes it a branch), and consent to it going on the tree, which is where forks of an edition are seen.
Cards you make can be a variation of an existing card. The link back is kept, so a card shows what it was derived from.
Play a game
Open Play. The game table is for the facilitator: it deals the cards, shows the board, scores the audience votes, rolls the dice and spins the Wheel of Speculative Fortune. You pick an edition, a board size and a team size. It recommends a setup (six to eight fields with two or three players per side is the most balanced), shows how long the board will take, and offers the standard hands from the manual: two or three starting cards per player, one or two shared cards, and draws, so that a team has a few more cards than the board needs. You can change the numbers.
Options at setup:
- Random deal or players choose their starting cards. When players choose, each one picks from their team's deck and anything left empty is dealt at random.
- Open hands (teammates can see each other's cards) or closed hands.
- Start cards, shared cards and how many draws. Keep drawing lets a team draw a new card after each one it plays, while the deck lasts.
During the game you record who played which card on which field, enter the audience scores and roll the die. A timer runs while a move is performed. It never stops the move: it shows how long it took, up to two minutes as a reference, and the time is kept with the move. If a move fails, the table asks for the story of why, from the audience first, and records who told it. The table applies the rules: a failed move discards the card and leaves the field open, a successful one claims the field and forces a response, two successes lead to a lightning round, and every completed round ends with the wheel. When all fields are decided it scores the board. Games are saved in the browser, and a finished game can be exported as a draft session in the same form as a documented session.
Mirror a game and send it in
If a game is played somewhere else, at a table, in a room or on a call, you can mirror it. Set up the edition first, with all its custom cards and fields, because that is what lets you pick what was played. Then on Play tick I am mirroring a game played somewhere else. Nothing is dealt: for every move you pick the card that was played from the deck, the field, who played it, the score and the die, and the failure story if it failed. When the game is over, open Send this game in, add links to the audio or video, or paste or choose an edited transcript, add notes for the moderators and choose Submit for moderation. You get a file with the draft session, the setup and the recordings. In the shared version this goes to the moderators directly, and after moderation it is added to the archive as an edition, with the mirror and everyone else credited.
Time the session
On Play, the Facilitator box has Start session and Stop session. Pressing Start begins a clock, and from then on every card played, score, roll, lightning round and wheel spin is time-stamped, in the log and in the exported draft. You can tick Record audio on this device: the recording starts at the same instant as the clock and downloads as a file when you stop, so the times match it second for second. Ask everyone at the table first. The recording stays on your device until you send it. If you record somewhere else, tick I am recording somewhere else and say how many seconds before pressing Start it began. A game that ends stops the clock too. Without Start the draft has no times, and a moderator adds them.
Print opens from a saved deck or edition in the builders and makes a printable sheet of its cards, with the artwork, so you can play without a screen. Use "Print or save as PDF".
Review and document a session
On a session page, Review opens an editor with the recording alongside. Set each move's time, and correct the card, field, player and score. Your edits are saved as a draft in your browser. Download the result when you are done, so the correction can be added to the archive.
Local tools for documenting a session
Documenting a session by hand takes many hours, and tools can do the first pass. The license (section 9) sets the rules: process recordings on your own computer or on infrastructure the commons controls, mark the output machine-made, not yet reviewed until a person has listened and corrected it, name the tool and model in the edition record, and never use a recording to train anything.
The tools below are suggestions, not requirements. They are chosen for what suits a permacomputing approach: fully open code, work offline, run on modest or older computers (a laptop without a graphics card is enough for most of them), use little energy and stay usable for years. This list was compiled from general knowledge and was not run against this project's own recordings. Licenses and model terms change, so check the current license of the exact tool and model before you adopt it, and test it on a minute of your own audio.
| Task | Good first choice | Other options and notes |
|---|---|---|
| Transcript (speech to text) | whisper.cpp (MIT) with a Whisper model, small or medium, quantised. Runs on a CPU, many languages. | faster-whisper (MIT) is quicker on the same hardware. Vosk (Apache 2.0) is much lighter and works on very old machines, with lower accuracy. Whisper's training data is not published; OLMoASR from Ai2 is a newer model built on open data and worth testing. |
| Timestamps | The word and segment times from whisper.cpp or faster-whisper, tidied with stable-ts (MIT). | WhisperX (BSD) re-aligns words. Montreal Forced Aligner (MIT) aligns a transcript a person has already corrected, which is the most exact. Silero VAD (MIT) finds the silences, which helps split moves. Then set each move's time in Review. |
| Who is speaking | Do it by hand in Review. Speaker names are an agreement with people. | sherpa-onnx (Apache 2.0) can separate speakers offline into "speaker 1, 2, 3", which a person then names. pyannote gives good results, but its models are downloaded under terms you must accept. Never voice-print people. |
| Subtitles and format | The archive uses WebVTT. whisper.cpp writes it directly. | Edit with any subtitle editor, such as Aegisub or Subtitle Edit. |
| Translation | Argos Translate (MIT), which also runs as LibreTranslate. Works offline. | Opus-MT models (CC BY). Avoid models released for non-commercial use only, such as NLLB, as they cannot sit in a commons that allows the work to be reused. A human translator is credited as the translator. |
| Drafting, summaries, tags and search | llama.cpp or Ollama (both MIT) running a small open-source model of 3 to 8 billion parameters, quantised. Good open choices are OLMo (Ai2) and Apertus (Swiss AI), whose data and code are public and under the Apache 2.0 license, then Qwen or Mistral models under Apache 2.0, which publish the weights only. | Models with custom licenses (such as Llama and Gemma) restrict some uses and are not recommended here. For search by meaning, use a small embedding model such as bge-m3 (MIT) with a local index. Quotes and sources a tool gives must be checked by a person. |
| Reading scanned cards or documents | Tesseract (Apache 2.0) | |
| Artwork (AI placeholders only) | stable-diffusion.cpp (MIT) or ComfyUI (GPL) with an open model such as FLUX.1 [schnell] (Apache 2.0). | Most image models were trained on pictures gathered without their makers' consent, which is exactly what the commons is against. Models trained on licensed or Creative Commons images exist but are weaker. Use them for placeholders, label them as AI-made, and prefer a human artist. |
A practical routine for one session. Record the audio locally. Run whisper.cpp for a first transcript with times. Open it in Review and listen through, correcting the words, speakers and move times. Change the mark to reviewed with your name as documenter. Fill in the edition record, including the tool and model you used. Send it in.
Choosing the smallest tool that does the job is the point. A small model run overnight on an old laptop is better than a large model on a rented server: it uses less energy, keeps the recording in the commons, and anyone can repeat it.
Support artists and take part in the community
- Patronage lists artworks waiting for support. Pick an amount (in BREAD or by card or bank) and see how it is split. Up to the goal, 80% goes to the artist, 10% keeps the platform running and 10% goes to the commons pool. Support can then go on for a further 20% of the rate, like a tip: the artist keeps the share they chose (a third by default) and the rest is split evenly between the platform and the pool. When the goal is reached the card is released to the community gallery on Cards, where anyone can put it in a deck.
- Commons pool is where the community can vote on example themes and pairings for later sessions (the theme of the opening session is set by the curators), see the player pool, nominate new players, and spend support credits on proposals.
- Offer an artwork is where you can put forward an existing artwork for a card, your own or someone else's (the artist is asked first). The artist chooses to be credited only, to claim a share of what is given to the card (from 1% to 50%, the rest going to the platform and the commons pool), or to keep the artwork free and offer the printed card or the original for sale or auction to fund a project.
- Shop has the printed editions and the founding run. Say you are interested and your interest is kept in your browser until the shared version opens and asks you to confirm. Production starts when 101 orders are in.
- Program shows the monthly series: the next session, who curates it, how to propose a theme, submit a card or vote.
At the moment the money, the artists on the patronage page, the figures in the pool and the votes are simulated. Votes stay in your browser. Nothing is paid and nothing is sent.
Your work is yours, so look after it
- My gallery shows the cards you saved and everything you made. Use Download backup to save a file with all of it, and Restore from file to load it again on another computer.
- Artworks are stored inside the browser with a limit of about 5 MB each. If the browser says its storage is full, download a backup and delete some cards with artwork.
- Clearing your browser's data erases the work, which is why the backup exists.
For people running a server or a team
The platform also ships with a command line tool, gc-session. It checks session files (validate), compares them with the old site (parity), detects scenes, and moves sessions in and out of spreadsheets (export-csv, import-csv). It is meant for the people who document sessions and curate the archive, not for players.