> For the complete documentation index, see [llms.txt](https://jordan-7.gitbook.io/jordan-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://jordan-7.gitbook.io/jordan-docs/internal/briefs/2026-07-28.md). # MegaChat brief — 2026-07-28 ## Objectives overnight * Ship the fan-facing bounty build to production * Make verification REAL: read our proof-codes out of actual broadcast footage, no trust, no manual review * Get Kick to parity with Twitch * Fix the moderation test that had been failing for days ## The headline **The machine can now read its own proof off a mangled video stream.** The little code badge the overlay shows during clips was redesigned as a machine-readable mark — chunky dots inside a white frame, like a tiny barcode a human can still read. A purpose-built reader finds that frame anywhere in a video image, at any size, and reads the code back — no AI service, no per-call cost, fully testable offline forever. Then the whole loop was proven in one motion, over the real product surfaces: a fan pledges, a streamer claims, a clip "plays" with a real code issued, the real overlay renders it, the footage is deliberately crunched down to the quality Twitch actually delivers — and the verifier reads the code back out of those crunched pixels, marks the playback verified at high confidence, and computes the payout from it. Money from evidence nobody self-reported. ## The numbers (from a few hundred generated test frames) * Full quality and 720p: **100% detection**, high confidence * Heavy motion (worst case for compression): **100%** * 480p: 92%, right at the legibility floor — **720p is the documented minimum quality** for verification; 480p works but sends more cases to human review; below that we say "can't verify" instead of failing quietly * Badge shrunk by a cheating streamer: every read measured under the floor — **not paid** * Badge covered up: never read as present * No badge at all: **zero false positives** — nothing ever pays on a frame with no code in it * Replayed footage from a different broadcast: verifies **zero**, pays zero ## Also done * **The days-old failing moderation test was a ghost.** A test server from July 24 never died and sat on the port every later run needed — so the test had been quietly testing a three-day-old zombie with dead settings. Killed it, fixed two genuinely outdated assertions, 10/0. The moderation pipeline itself was never broken. * **Kick is real now**: claiming a Kick handle requires signing in with the Kick account that owns it (same proof standard as Twitch, and a Twitch login can't claim a Kick name — checked, not assumed). Viewer counts are captured from Kick at every clip playback, same as Twitch. One address needs registering in Kick's developer console: `https://megachat.fun/auth/kick/callback`. One platform finding: Kick's official API can't list past broadcasts, so Kick verification checks the live stream rather than the recording. * **Honest failure states**: a deleted, subscriber-only, or still-processing recording — or a missing extraction tool — goes to the human review queue as "we couldn't look", pays nothing while unresolved, and is never treated as "the streamer cheated". ## What's genuinely left * **One real broadcast.** Everything is proven on synthetic footage built from the real overlay; the only untested stages are grabbing from a real live stream and finding the right recording afterwards — untestable without a broadcast. It's one command when you're ready: `node _rehearsal-run-b.mjs --handle ` (with your stream key set, it runs the whole thing unattended, \~12 minutes; or go live yourself with the overlay and add `--skip-push`). * **The weakest joint in the whole design, stated plainly**: verification proves clips AIRED, not that anyone watched. A streamer could play everything at 4 a.m. to zero viewers and collect. We already capture the audience size at every playback for exactly this reason — making the payout scale with it is the next money change, and it should land before real settlement does. ## Outlook: bullish * The product thesis — trustless proof of broadcast — went from mocked to demonstrated tonight, with measured accuracy numbers instead of hopes. * The remaining unknowns are narrow and named: one real broadcast, one economics change (audience-weighted payout), and the long-standing no-real-settlement caveat. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://jordan-7.gitbook.io/jordan-docs/internal/briefs/2026-07-28.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.