How a verdict is made.
- Empty0–19
- Weak20–39
- Thin40–59
- Credible60–79
- Solid80–100
What is read
halo is a launchpad on Solana over Meteora’s bonding curve. A team that launches a coin here attaches its website and, if it has them, its docs and repository. Those are read after the launch, and a verdict is usually on the coin’s page within minutes.
Website The start page and up to 12 more pages of the same site, picked for relevance: whitepaper, docs, how it works, tokenomics, about, roadmap, team, FAQ. Docs Up to 15 pages, or the docs’ llms-full.txt. If no docs link was given, docs found on the project’s own site are read, and the verdict says so. GitHub Through GitHub’s public API: languages, the last 100 commits, contributors, the file list, the README and a sample of the main source files. A repository found on the site or in the docs is used when none was given, and the verdict says so. X Not read. It is kept as a link. What is measured
Before a model sees anything, code counts what can be counted. These facts are shown on every verdict under “Measured by code”: − marks one that weighs against the project, + one that weighs for it, = one that does neither.
- Marks of a site generator or builder: v0, Lovable, Bolt, Framer, Webflow, Wix and others.
- How much text the pages held, and whether the site is a JavaScript-only page with no readable text.
- The repository’s age, whether it is a fork, how many commits it has and over how many days they were made.
- Contributors, and commits that carry an AI co-author mark.
- The size of source code against config, scaffolding and generated files.
- Text hidden from visitors that tells an AI reviewer how to judge the project.
How it is judged
One call to a Claude model gets everything that was read, plus the measurements, and is asked for a blunt, specific review in which every claim names what it is based on.
Everything fetched from a project is treated as data. If a page contains text addressed to an AI reviewer, that text changes nothing, and the verdict carries a red flag saying the page tried.
Code checks the answer before it is published. A project that presents itself as software and shows no code of its own cannot score above 59. One where nothing could be read cannot score above 19. Nobody edits the verdict.
Score
How much substance stands behind the coin. The grade is the score in a word, on the scale at the top of this page.
Vibe-coded
An estimate of how much of what was read looks machine-generated or lifted from a template. It is not a quality mark by itself. A templated site can sit on real code, and the score will say so.
The ten checks
Every verdict answers the same questions with yes, partly, no or unknown, and one line of evidence. Unknown means a source could not be read. Something that is simply absent is a no.
| Question | What earns a yes | |
|---|---|---|
| 1 | The site opens and can be read | The start page opens and its text can be read without running scripts. A page drawn only by JavaScript is a no. |
| 2 | It says what the product is | The pages say concretely what the product is and what it does. |
| 3 | The whole mechanism is described, not a part of it | The mechanism is described from end to end. When a part is missing the answer is partly, and the note names the part. |
| 4 | There are docs beyond a landing page | A docs site, a whitepaper, or a README of real length. |
| 5 | The coin has a stated role in the product | A specific role is stated. The bare words “utility” or “governance” are not one. |
| 6 | A public repository is linked | A public repository was given at launch or found on the site, and it opened. |
| 7 | The repository holds real code | Code written for this project, beyond scaffolding, templates and an unchanged fork. |
| 8 | The code does what the site claims | The sampled source files implement what the site and the docs claim. |
| 9 | Somebody stands behind it by name | Real names, linked profiles or a registered company. A pseudonym with a shown track record is a partly. |
| 10 | It is clear what works today and what is a plan | The pages make clear what works today and what is a plan. |
Fees
Every coin launched here trades on its own Meteora bonding curve, under a config that carries the launcher’s terms and cannot change afterwards. The launcher sets the trading fee between 1% and 10% of every trade, may start it higher for the first minutes so that whoever buys in the first seconds pays the launcher, may lock its own dev buy for up to 2 years in a contract it cannot cancel, and picks what the coin is paired with: SOL or USDC. Nothing is set aside for the team: the whole supply goes on the curve, and what a launcher locks is what it paid for.
Of every trading fee Meteora keeps 20%; of the rest 80% goes to the launching wallet and20% to the platform. Fees wait on the pool until the launcher claims them on the coin’s page. When the curve fills, the liquidity moves to a Meteora pool and is locked for good; its fees divide the same way. A launch costs 0.1 SOL for a project and 0.02 SOL for a meme coin, paid in the launch transaction.
Charts are drawn with TradingView Lightweight Charts™, © 2026 TradingView, Inc.
A verdict is a model’s opinion of public pages. It can be wrong.
A high score does not make a coin safe to buy. A low score does not prove fraud.