How the instrument reads.

VIRGIL converts a launch's public, on-chain configuration into a single grade using a fixed rubric. There is no model, no sentiment, no discretion. The same launch always yields the same reading — and you can recompute it yourself from the same public data.

Three weighted signals

Every launch is measured on three categories, each scored 0–100, then combined by weight into a composite:

• Launch configuration — 40%. The choices a founder makes on-chain: the anti-sniper tax, the 60-day founder commitment, the capital-formation module, launch radar, allocation. Choices that cost a serious founder little and a fleeting one everything.
• Creator history — 35%. The deployer wallet's age, transaction record, and how many tokens it had launched before this one. A fresh wallet on its four-hundredth launch reads nothing like a real builder's.
• Presence — 25%. Whether a real, claimed identity stands behind the launch: a linked account, a substantive description, documentation.

composite = (config × 0.40) + (creator × 0.35) + (presence × 0.25)

The full weights table

Every point is published — this is the entire instrument. Each category scores 0–100 from the sub-signals below, then enters the composite at its weight. Nothing is withheld, so any grade can be recomputed from public data.

An A requires the 60-day founder commitment. At 40 points of launch configuration, no launch can reach the ≥ 85 composite for an A without it — the bar is structural, not a matter of taste.

A reading is dated

Two of these signals describe the launch as it stood at launch. Launch configuration is fixed on-chain the moment a token goes live. Creator history is read as of that launch— only the deployer's launches that preceded this one count toward it. A wallet's later launches never reach back and change a grade already issued; what a deployer does afterward is its own future record, not a revision of this one. This is also what keeps the count reproducible: a tally of what came before never shifts, so anyone recomputing it later gets the same number.

Fixed grade thresholds

An A is deliberately hard: it requires, among other things, the 60-day founder commitment — a founder locking their own allocation. The bar is not the rubric being harsh; it is the rubric refusing to be impressed by anything less than real commitment.

Worked example — reproduce it yourself

Take OrbisAPI ($ORBIS), a real graded launch. Its three readings:

Nothing here is a judgment call. Apply the same weights to the same public inputs and the grade reproduces exactly — which is the entire point: a reading anyone can check is a reading both sides of a transaction can trust.

What a grade is not

A grade is not a price prediction and not advice. A high reading means a founder enabled the honesty signals available at launch — not that the token will rise. Most launches that grade well still fail. VIRGIL measures one thing: the integrity shown at the gate.

• Hidden backer or insider-wallet relationships — VIRGIL sees on-chain config only.
• Code quality or contract-audit status.
• Team identity or real-world credentials.
• Price performance or graduation likelihood — no prediction of any kind.
• Community size or engagement quality.
• Product viability — VIRGIL does not evaluate what the project builds.

Missing-data policy

Any signal VIRGIL cannot verify scores 0 for that signal. Unknown is not average. A wallet with no on-chain history scores 0 on wallet-age and tx-count — not an assumed median — so a launch is never flattered by the absence of evidence.

On-chain data sources

Every input is read from public data, at no cost — so anyone can reproduce it.

When the method changes

The rubric is meant to be stable, but when it is revised — or when a calculation is corrected — the change is logged openly, with its rationale and the number of grades it moved. The full record lives in the grading changelog.