Demo Weekend
What the trials found
A trial is a full pass at the live app: build real adventures, play them with real seats, then say plainly whether this is something people would spend a Saturday night on. Each one leaves a set of reports behind, and this page is the front of them.
Generated from the reports themselves, so it cannot drift from them. For what there is to test, see the walkthrough; for what is built, the roadmap.
Played a night yourself? Tell me what you found. Nothing on it is required, it saves as you type, and a person reads every word.
2026-10-12
17 reports · 37,520 words
Ready, with conditions
What this was: one agent playing eight personas and two observers against production (https://dnd-maxvercel.vercel.app) from 21:33 to 23:48 Phoenix time, in real browsers — three signed-in seats and one signed-out guest window — with the repository shut until…
58/100
Current Product
70/100
Builder
54/100
Player Experience
60/100
Multiplayer
61/100
Fun
72/100
Adventure Creation
55/100
Production Reliability
Observed at https://dnd-maxvercel.vercel.app/ between 21:33 and 23:48 Phoenix time, from three signed-in regular accounts (@Demo1, @Demo2, @Demo3) and one signed-out incognito window. Nobody was an admin.
One agent wearing ten hats, not ten humans: eight personas with a reason to be there and two observers (Uma, UX; Quinn, QA). Every seat that is called "real" below was a separate cookie jar read off the page; where a seat could not be driven it says NOT…
Everything below is still in production and was not deleted, so the owner can open it. Names are exactly as the shelf shows them (read off @Demo1's Adventures page at 23:49).
Four tables were run. Three held a host and at least two other real seats at once, which meets §5's "one host and two players" bar. Every seat below was a separate cookie jar and its identity was read off the page.
Every defect seen in production on 2026-10-12, from the browser only (Stages 1–2; the repository was shut). Stage 3 adds the likely technical area, the duplicate check against docs/issues/ and the issue file for each one; until then the Issue column is blank…
Things that are not bugs but cost somebody their evening, grouped by the §12.2 classes. Each row: the moment · who · expected · what happened · what it cost · the suggested improvement (PROPOSED). Uma's counts close the report.
Everyone gathered after the forms. Nobody revised an earlier observation; where a persona's view here differs from the journal, the journal stands.
Everything the run proposed, ranked by value. Effort and risk are scored because the owner wants to know them; they never moved an item down (§12.4, "Nothing is cut for being big"). P3 and MOONSHOT are sequencing, not burial.
Stage 3: the repository open, the Top 10 and Next 20 from 08 made executable. File and line references come from six read-only passes over the source on 2026-10-13 and are approximate to a few lines.
Kept short on purpose. Each one makes DMAX look broken in the first ten minutes of a real evening.
A weekend you can print and hand to six people. Written from the 2026-10-12 agent run; it assumes the four demo blockers in 10 are fixed first. Where a step says "observer", that is a person who watches and writes, never helps.
Everything here comes from what the personas wanted on the night (§11.1–11.3). Nothing is ranked down for being big. Each AI proposal states how it keeps the architecture rule: AI may narrate or request actions; it never mutates game state directly.
Scored 1–5 separately for beginner (Pat, Rosa), intermediate (Duke, Kai, Theo, Sam) and expert (Grizz, Nadia), from what each group actually did on 2026-10-12. "—" means the group did not reach it.
Eight responses to the real form at /feedback (version 3), one per persona, each saved signed out as "DEMO-WEEKEND <persona>". The capture method and its one deviation from the prompt (a cleared browser id instead of a fresh incognito window per persona) are…
Every bug and broken thing from the 2026-10-12 trial, in one place, ranked by severity. Nothing else is mixed in: features and ideas are in 08. Stage 4 works down this list; 16-FIX-AND-RETEST.md records each fix and its production retest.
Stage 4, 2026-10-13. Bugs only, worked down 15-FIX-THIS-LIST.md by severity. Features and design questions stayed where they were.
Built from docs/testing/demo-weekend-* on 2026-10-25. Add a trial directory and this page follows.