Trust page, not marketing copy
Methodology
Every scoring rule below is published, versioned, and implemented as a deterministic formula no learned model, no LLM in the loop, no non-reproducible step. Given the same inputs and the same algorithm_version, the Scorer always returns the same output.
No paid placement, ever
The scoring tables carry no relationship to any billing or customer record. A launchpad paying for a report about itself cannot touch its own final score enforced at the schema level, not by an internal policy someone could quietly waive.
Independent, not self-reported
Scoring only ever reads third-party market data and on-chain facts. Anything a launchpad submits about itself is a distinct data type that cannot satisfy the interface the Scorer reads from it cannot leak into a score by accident, only by someone deliberately changing the type.
Composite Weighting
A composite score is the weighted sum of five dimensions, clamped to 0–100, then mapped to stars via fixed thresholds.
graduation & anti-rug
audited factory, LP lock
depth, anti-wash
realistic ROI
temporal stability
- Missing data is never a number. A dimension needs data from at least 5 launches. Until then it shows n/a, is left out of the composite (the remaining weights are rescaled), and the score is marked provisional.
- Young tokens are not judged. Quality, Value and Consistency only count tokens at least 72h old graduation and price outcomes need time to play out. A composite also needs at least 3 of the five dimensions measured; with fewer, the launchpad shows as not yet scored rather than a number built from a couple of partial signals.
- No market is a zero, not a gap. A token that never reached a DEX pool counts as zero liquidity in Market Health rather than being skipped.
- Gains, not just stability. Value is measured on a log scale from the launch price (a token that never rose scores 0, 10× scores 100), and Consistency is multiplied by how good the typical outcome is a launchpad whose tokens all flatline is not rewarded for being predictable.
- Partial dimensions are capped. Where a dimension has several components but only some are measured yet (Mechanism: contract verification is one of three), its score cannot exceed the share that is measured.
Note on dimension count: this document uses five dimensions rather than the four referenced in an earlier product brief. Whether to collapse two of the five is an open decision treat this table as current truth until it's resolved.
Star thresholds
The same three thresholds define the red/amber/green colour ramp used everywhere a dimension score is shown colour and stars never drift apart.
Cold-start honesty
A launchpad with a tracked sample below 20 launches is marked is provisional and its star rating is capped at 1 regardless of the raw composite score. This is enforced inside the scoring function itself, not as a frontend warning layered on top there is no configuration flag that disables it.
In practice this means most of the directory will sit at 0–1 stars for months after a new chain launches. That's the system working as intended, not a bug to fix by lowering the threshold.
Initial backfill sampling
When a launchpad is first onboarded, Assay decides how many historical launches to pull in with a single, one-time rule separate from the hourly ingestion rotation that runs for launchpads already tracked.
Floor at 100
Below this many total launches, sampling isn’t worth the complexity take everything. This is a different concern from the 20 launch confidence gate above; a launchpad can be cheap to backfill in full and still end up provisional.
20% of the upstream total, capped at 1,000
The sample scales with the launchpad up to the cap: a launchpad with 500 total launches backfills 100, one with 5,000 backfills 1,000, and one with 276,000 also backfills 1,000. The cap keeps a very large launchpad from exhausting the upstream data APIs at that size a thousand launches already pins a rate such as graduation to within about three percentage points.
Recent-first, not random
Consistent with the scoring engine’s own recency weighting, a fresh dossier opens with a launchpad’s most current behaviour, and needs no separate argument for why old and new launches would otherwise be interchangeable.
Runs once, at onboarding only
After backfill, a launchpad’s sampled launches enter the normal hourly rotation, and any launch published after onboarding is picked up through the existing new-launch fast path never through this rule again, unless the launchpad is manually re-onboarded.
Try it
At or above the 100-launch floor: take 20% of total, most-recent-first, capped at 1,000.
A launchpad's dossier always shows both its sampled and upstream-total counts side by side a partial sample never silently presents itself as the whole population.
Scores are informational only and do not constitute financial, investment, or legal advice. Assay is not a registered investment adviser. Scores describe historical, on-chain patterns not predictions about any specific future token. Conduct independent research before making any decision.