November 3, 2026 | Data-driven report. With the Meta Global AI Developer Hackathon 2026 approaching its submission deadline, a third-party project has built a searchable directory of public GitHub repositories that appear to be associated with the contest. Watt Mind's Meta Hackathon Project Directory tracks code activity, demos, tests, category labels and Meta-model mentions. It offers a compelling snapshot of visible developer activity, but it is not an official Meta or Cerebral Valley ranking, nor a verified list of completed submissions.

Who operates the tracker and what exactly does it collect?

Watt Mind calls its directory an Unofficial Community Tracker and says it is updated every four hours. It discovers public repositories through GitHub topics such as meta-hackathon-2026 and meta-ai-hackathon, along with descriptive mentions. Visitors can filter by technology, subject and activity, inspect repositories, follow live demo links and download project data in CSV or JSON. The authors disclose that they are themselves building watt-mind/pii-guard for the event.

This inclusion method creates selection bias. A directory entry does not establish acceptance into the hackathon, timely submission through the official portal, eligibility, project uniqueness or actual judging. Private code, missing topics and unpublished entries can be omitted, while related repositories from the same project can be counted separately.

Snapshot: 1,120 repositories and nearly 100,000 commits

In the directory snapshot provided for this report, with the latest visible observation marked November 3 at 19:31 UTC, the tracker reports 1,120 public repositories and 98,158 commits in the analyzed development window. It lists 672 repositories with at least 20 commits (60.0%), 664 with detected automated test suites (59.3%), 390 with live deployment links (34.8%), 292 using pull requests (26.1%), 216 with video walkthroughs (19.3%) and 203 with Dockerfiles (18.1%).

These are tracker-reported signals, not independently verified software properties. A test directory does not prove that tests pass, a Dockerfile does not prove a deploy succeeds, and a live URL does not guarantee functionality. Commit counts are especially poor proxies for quality, since tooling, coding style and project structure can produce widely different histories.

Which project categories dominate?

Watt Mind categoryRepositoriesShare
Business, finance and operations24021.4%
Developer tooling, infrastructure and security20518.3%
Healthcare, civic and sustainability13612.1%
Multimodal vision and video1039.2%
Gaming and creative media1019.0%
Education and knowledge systems948.4%
Autonomous agents and workflows766.8%
Robotics, IoT and hardware756.7%
General and emerging applications554.9%
Voice, speech and audio AI353.1%

The totals add up to the 1,120 indexed repositories. Business-related applications and developer infrastructure account for 445 projects, nearly 40% of this visible sample. That does not mean they represent 40% of validated official submissions or have better odds of winning.

Two types of tracks are being conflated

Watt Mind marks 782 repositories (69.8%) as Build for Your World and 338 (30.2%) as Push the Frontier, its Track 1 and Track 2. On Meta's official event page, those are the project themes; the participant tracks are indie developers and startups. These are not interchangeable. The tracker should not be cited as evidence that 69.8% of submissions are indie developers or that 30.2% are startup teams.

Which Meta technologies appear in public repositories?

The directory flags Muse Spark in 875 repositories (78.1%), Meta Model API in 712 (63.6%), vision segmentation/SAM in 495 (44.2%), open-weights fine-tuning in 274 (24.5%), Muse Wonder in 251 (22.4%), Muse Code in 157 (14.0%) and Muse Glimmer in 133 (11.9%). These groups overlap. They may reflect lexical references, SDK declarations and heuristic detection rather than a proven working invocation. They cannot be added up as mutually exclusive market shares.

How the unofficial 0–10 leaderboard works

Watt Mind applies a four-pillar automated rubric: impact 30%, Meta AI tooling 25%, technical depth 20% and demo completeness 25%. The mechanism looks for public evidence such as architecture documentation, GitHub activity, test and CI directories, Docker packaging, model names and links to demos. These heuristics can be useful as a repository-completeness checklist but cannot reproduce judges' assessments of originality, impact, user experience or scientific validity.

The most consequential methodological disclosure is the authors' own: their entry, watt-mind/pii-guard, scores a full 10.0 because they built against their checklist. That is a clear potential conflict of interest and a risk of self-favoring rubric design, even without alleging manipulation. A small but excellent project can rank lower than a large repository packed with detectable artifacts.

Automated demo-video audits: useful, but still imperfect

Beyond source-code signals, the tracker describes a pipeline using yt-dlp, ffprobe, frame extraction, Meta vision tooling and Muse Spark to score linked videos. Its six criteria cover playability/duration, event provenance, live product footage versus slides, visible Meta model execution, numerical benchmarks, and audio/caption quality. It says 216 repositories had videos audited and 22 reached the maximum video subscore, including its own entry.

Video heuristics cannot reliably establish end-to-end product quality, usability or real-world correctness. A persuasive short demo may omit edge cases. The scoring methods are disclosed, which helps readers assess them, but the scores remain unaffiliated with official judging.

Could this leaderboard predict winners?

No credible prediction follows from it alone. Winners are chosen by the event's official judges through their own process. Static GitHub metadata is at most circumstantial evidence of engineering work. A high rank means that a repository fits Watt Mind's chosen characteristics, not that Meta judges consider it superior. The directory's own explanation states that its numerical scores are not product-quality or creativity measurements.

How to use the directory responsibly

  1. Search for a public team or project and inspect its GitHub repository directly.
  2. Filter categories to study patterns rather than to create a winners table.
  3. Verify README claims, commit history, public tests, licenses and demo links.
  4. Try public demonstrations only with non-sensitive data.
  5. Treat video metrics and model-usage flags as automated findings needing context.
  6. When citing a statistic, preserve the snapshot date and sampling caveat.
  7. Use the official Meta site for deadlines, judging requirements and final winners.

Deadline: November 4 at 9 p.m. Eastern Time

Meta lists the official submission deadline as Wednesday, November 4, 2026 at 9 p.m. U.S. Eastern Time, equivalent to 3 a.m. on Thursday, November 5 in mainland Spain. Official judging is scheduled for November 4–8, followed by the final ceremony and announcement of winners on November 9, subject to change. Meta advertises a $1 million prize pool. Neither the overall count of admitted participants nor the number of eligible final submissions can be inferred from public GitHub repositories.

What this says about AI-assisted software development

The tracker demonstrates how much of a modern hackathon's work is publicly observable: commits, architecture files, deployment links and demo videos can all be harvested into near-real-time dashboards. That visibility invites useful questions about openness and engineering practices, but it also creates incentives to game activity metrics. As AI coding tools make generating large amounts of code easier, human review of product behavior and evidence will become more important, not less.

Sources and reporting methodology

Editorial note: all directory counts are attributed to the third-party tracker, not independently audited GitHub data. The directory updates frequently and is run by a contest participant. Its scores are not official Meta scores or predictions of winners.