How we read a contract.
What we claim. What we don’t.
While we deploy six independent engines and a battery of reputation databases to do the heavy lifting, the final “GhostScore” isn’t just the output of a cold algorithm. We recognize that machines are excellent at counting, but rather hopeless at judging.
Every score is tempered by human ingenuity, the kind of “proof of work” that a paying client deserves. This page serves as our source of truth: detailing the mechanics of what we measure, while acknowledging the vital gap where automated reads end and human discernment begins.
After all, if you want to engineer trust, you can’t just rely on a motherboard; you need someone who understands the messy, brilliant psychology of the person on the other side of the screen.
Six engines. Partners. Authorities. AI consensus. Other proprietary checks.
A GhostLabs read isn’t a single tool running checks in sequence. It’s six independent analyses running concurrently, cross-referenced against partner intelligence and regulatory databases, then synthesized by an AI consensus engine that reads what individual tools cannot. One score. One verdict.
AI Consensus Engine
When the engines disagree, something has to referee. This layer reads all of their findings side by side, throws out the contradictions, and asks the question a single tool never does: taken together, does this contract behave like something built to keep your money, or to lose it? It runs the moment you do, at any hour.
External Audit Partners
On high-value engagements, we do not ask you to take our word for it. Findings are checked against independent third-party auditors working from their own methods, so a blind spot in ours never quietly becomes a blind spot in your decision.
Regulatory & Sanctions Checks
Every address is run against the lists that carry legal weight: OFAC, the EU, FATF, and local regulators in the jurisdictions that matter. Sanctioned wallets, mixer-tainted funds, and anyone under active enforcement get flagged before you go near them.
Source & Bytecode Analysis
The unglamorous part: we read the actual code. Control-flow and AST analysis, matched against thirty-plus known exploit signatures, from reentrancy and integer overflow to delegatecall hijacking, hidden mints, and the quiet blacklist function that only bites on the way out.
Large-Model Interpretation
A large model reads the contract start to finish and says, in plain English, what it is actually trying to do. It catches the odd behaviour and the clever tokenomics that pattern-matching walks straight past.
Behavioural History
What the contract is doing right now, on its own chain. Who the owner is and what they have been up to, how deep the liquidity runs and whether it is locked, where the tokens actually sit, and every quiet upgrade in its past.
Four Databases
Code does not have a reputation; the people behind it do. We check curated rug and exploit registries, sanctions lists, and our own deployer graph, the one that remembers which wallets shipped which contracts, and how those stories ended.
Tokenomics Health
Where the supply sits, how it unlocks, what the fees really do, the quiet max-transaction and max-wallet limits, and who holds the treasury keys. This is where the trap doors hide that clean-looking code will never show you.
Continuous Upgrade Detection
For monitored contracts. The safe contract that quietly turns unsafe is the one that ruins people, so we watch every cycle for proxy upgrades, implementation swaps, and owner changes, and re-score within fourteen seconds of anything moving.
The Anatomy of a Verdict: Why Trust Requires a Soul in the Machine.
A score that arrives via a “black box” is not a service; it’s an oracle, and oracles are notoriously prone to hubris. At GhostLabs, we treat the assembly of a GhostScore as an act of Institutional Alchemy.
We start with the cold, unblinking logic of six independent engines, because machines are better at counting than humans. But we don’t stop there. We layer on “Partner Intelligence” and “Regulatory Screening” to provide context that a motherboard simply cannot grasp.
Finally, an AI consensus layer applies Adversarial Reasoning at machine speed, testing assumptions and resolving contradictions across all six engines simultaneously. We don’t just calculate a weighted average; we engineer a consensus. If an AMM contract is being judged, we tilt the scales toward Tokenomic Health, because in marketing as in math, relevance is the ultimate multiplier.
The five action labels
- EVACUATE (0 to 30). Direct path to loss of user funds. Potential honeypots, active exploits, hidden mints, blacklist-and-trap mechanics, malicious upgrade authority. Covers Tiers 1 to 3: Total Asymmetry, Exit Liquidity, Literary Fiction.
- CAUTION (31 to 50). Significant risk requiring trust assumptions. Unrenounced ownership with mint capability, mutable fee structures with no upper bound, heavy off-chain marketing masking structural issues. Covers Tiers 4 to 5: Synthesized Hype, The Mediocracy.
- OBSERVE (51 to 60). Operational concerns or reduced defensive depth. Code works as advertised but lacks polish. No critical findings, a few medium-severity flags. Covers Tier 6: Accidental Competence.
- SECURE (61 to 80). Solid implementation. Low and informational findings only. Liquidity locks visible, owner privileges minimised or renounced, healthy holder distribution. Covers Tiers 7 to 8: The Rational Choice, The Alpha Signal.
- INSTITUTIONAL (81 to 100). Best-practice implementation. Multi-sig governance, time-locks, proper upgrade patterns, transparent treasury controls. Zero critical findings. Covers Tiers 9 to 10: Legacy Potential, The Gold Standard.
Empirically calibrated weights
The GhostScore™ is five pillars and a hundred points, and the weights behind them are not a matter of taste. They come from going back through 340+ documented exploits between 2021 and 2026, roughly $4.3 billion of other people’s money, and asking one blunt question of each: which pillar, had we scored it, would have caught this before the funds walked out the door? The answer was not the flattering one. Bad governance emptied about as many wallets as bad code did. So Security and Team & Governance carry equal weight here. Tokenomics and Value sit a rung below; Health sits below them. And no clean pillar can quietly cover for a rotten one: fail enough of the checks that actually matter, and an amplifier drags the whole score down, steeply.
A non-linear amplifier penalises projects that fail multiple critical-threshold questions, ensuring that one clean pillar cannot mask systemic risk elsewhere. Hard overrides cap the score for potential honeypots, sanctioned addresses, and active exploits.
For the full research paper on our calibration methodology, empirical findings, and the amplifier design, read The GhostScore: A 100-Point Framework for Smart Contract Risk →
Weighting principles
- On-chain history weighs heavier than source-code analysis when source is unverified.
- Reputation database hits (deployer with rug history) override otherwise-clean code analysis.
- Tokenomics structure weight scales with the contract’s liquidity-extracting potential.
- Time decay: an older contract with no incidents scores incrementally higher than a freshly-deployed equivalent.
Every contract we read is published.
A scoring model earns trust by producing scores you can challenge. Every GhostScore verdict, free or paid, is written to our public ledger the moment it’s finalised. Scores, tiers, timestamps, on-chain addresses: all verifiable, all permanent. No private lists, no hidden results.
The ledger is not a curated highlight reel. It is the full, unedited record of every contract that has passed through the GhostLabs engine, from Tier 10 blue chips to Tier 1 honeypots. Browse it, challenge it, cite it. That’s the point.
For detailed analysis of how scores are derived, including the empirical calibration methodology and the non-linear amplifier design, read the full research paper: The GhostScore: A 100-Point Framework for Smart Contract Risk →
For a comparison of how the GhostScore differs from other audit tools on the market: Smart Contract Audit Tools Compared (2026) →
Why some reads are capped.
The five pillars measure what a contract is. A second layer measures whether there is anything real behind it. When a token shows the observable signature of a templated, short-lived launch, that signature bounds the score, regardless of how clean the bytecode looks in isolation.
These are caps, not accusations: they can only lower a score, never raise one, and each attaches a plain-language reason to the verdict. They exist because a contract can be technically unremarkable and still sit inside a pattern that, empirically, ends the same way. We name the signal; we do not name-call.
The signals we read
- No verifiable surface. No working site or documentation that actually references the token. A project that cannot be found anywhere but the launchpad is treated as unproven, not promising.
- Concentrated or bundled supply. A small cluster of related wallets controls a large share of the token from the first block, the structural pre-condition for a coordinated exit.
- Decoy presence. A website exists, but never mentions the token it claims to represent. A borrowed page is not a project.
- Volume without depth. Reported turnover far exceeds what the liquidity pool could organically support, the footprint of manufactured activity rather than real demand.
- Templated launch flow. The launch’s fingerprint matches a recurring cohort whose members have, as a group, a high historical failure rate. Membership is descriptive, not a charge.
The score is earned back
A capped read is not a life sentence. As a token accrues the things substance is made of (locked liquidity, listings across multiple venues, time on-chain without incident), the cap loosens and the score is allowed to rise. The model is structurally generous to projects that simply keep existing and behaving.
Disagree with a verdict? Challenge it.
A score you can contest is a score worth trusting. Every GhostScore is timestamped and written to the public ledger with the exact signals that produced it, so any verdict can be re-examined against the evidence that was actually on-chain at the moment it was read.
If you believe a read is wrong (a cap misapplied, a site we couldn’t reach, supply we mis-attributed), tell us. Disputes are checked against the logged signals, not re-litigated from scratch. There are three honest outcomes:
- The score stands. The signals hold up, and we show you exactly which ones.
- The score is corrected. A signal was stale or wrong; we re-read, and the ledger updates, permanently and visibly.
- The score is annotated. The read is defensible but context matters; we attach your evidence to the verdict so the next reader sees both sides.
What we need: the contract address, the chain, and the specific signal you’re challenging, with whatever evidence supports it (a live link that references the token, a locked-liquidity transaction, a governance disclosure).
Prefer email? Write to [email protected] with the address and your grounds. This page describes our review process; a dispute outcome is an analyst judgement, not a warranty.
From Total Asymmetry to The Gold Standard.
Every contract that hits GhostLabs lands on one of ten named tiers, each carrying a psycho-logic tagline and a directive Action label. The names are deliberate brand assets: quotable, screenshottable, used as shorthand by traders and treasuries alike. The Action label is the operational layer. It is what a compliance officer programs against, what a DEX aggregator filters by, what a trader uses as a binary signal. GhostScore™ and the ten verdict-tier names are trademarks of GhostLabs.
Every chain that matters.
GhostLabs reads across 35 chains spanning six virtual machine families: EVM, Solana (SVM), TRON (TVM), TON, Move (Sui & Aptos), and Clarity (Bitcoin L2). Coverage prioritises chains where capital actually moves, not chain count for its own sake.
Each chain family requires its own adapter layer: native RPC integration, chain-specific exploit-pattern detection, and reputation database mapping. EVM chains use Slither and Mythril for static analysis. Solana uses RugCheck and Birdeye. TRON, TON, Sui, Aptos, and Stacks each have dedicated intelligence probes calibrated to their unique attack surfaces. We add chains when meaningful TVL and active deployment justify the engineering cost, not based on launch announcements alone.
What each tier actually does.
Every tier uses the same scoring methodology. The difference is depth, deliverables, and the level of human involvement. Here is exactly what you get, and what you do not get, at each level.
Tier 1: Free Read ($0, unlimited)
- Identifies high-confidence security signals, common exploit patterns, on-chain history, and reputation flags across thirty-five chains. Catches the majority of straightforward rugs and obvious malicious patterns.
- Returns a GhostScore (0 to 100), tier classification, top findings, and a permanent public URL. No account required, no limit.
- A free read is a triage layer. It is not a substitute for a multi-week human audit. A high GhostScore is not a guarantee of safety; a low score is not a verdict. Read the underlying findings before acting.
Tier 2: Deep Audit ($98, one-time)
- Everything in the free read, plus line-level human-grade review, exploit-path narratives, severity-ranked findings, and remediation guidance.
- Deliverables: signed PDF report, findings spreadsheet (CSV), public verifiable URL, executive summary, and free remediation re-test for thirty days.
- Suitable for documentation, exchange listing diligence, and external trust signalling. Not a replacement for a bespoke engagement on nine-figure-TVL protocols.
Tier 3: Sentinel Monitoring ($129/mo per contract)
- Continuous upgrade detection, owner/admin/proxy change monitoring, and automatic re-read within seconds of any on-chain mutation.
- Alerting via Discord, Slack, Telegram, or email webhooks. Private monitoring dashboard.
- Monitoring catches what an audit cannot: the contract that was safe at deployment and changed after.
Tier 4: Enterprise (from $2,000/mo)
- Custom SLAs, embedded scoring APIs, dedicated review pipelines, and white-label integrations for wallets, exchanges, and aggregators.
- Scoped per engagement. The methodology is the same; the delivery model is built around your workflow.
Across all tiers
- New attack vectors emerge constantly. A contract that scored well today can score differently after a malicious upgrade. Scores are evidence at a point in time, not permanent guarantees.
- GhostLabs does not provide investment advice. We describe properties of code at a moment in time. Capital allocation is yours.
How a score updates over time.
A GhostScore is not a one-time snapshot. Three triggers can update an existing score:
- Re-read on demand. Anyone can trigger a fresh read on any contract at any time. Free. The score and timestamp update accordingly.
- Bytecode change. For contracts under Sentinel monitoring, detected changes (proxy upgrade, implementation swap, ownership transfer) trigger an automatic re-read.
- Reputation database update. If a contract’s deployer is later linked to an exploit, the affected contract’s score is re-weighted automatically.
Every contract has a permanent public page at /c/{chain}/{address} showing the current score, tier, top findings, and the timestamp of the most recent read. The public ledger indexes every read across all contracts.
Ready to read a specific contract?
Free, every chain, every time. The deep audit is the same methodology, line-by-line, with a sealed AI verdict. Enterprise pipelines embed the read inside your launch, listing, or treasury workflow.