What Conyr sees—
and where it stops.
Conyr is a real-time Solana actor-intelligence layer. It turns decoded market activity into position-level skill, funding paths, behavioral resemblance, and coordinated-wallet evidence. This page defines the coverage contract behind those answers.
25
registered programs20 hops
provenance depth136
curated origins30-dim
behavior vector10 classes
bundle structuresFirst-party parsing, precisely stated.
Conyr's production pipeline receives Solana transactions from its own Yellowstone gRPC infrastructure. The public scale metric counts successfully decoded swaps. The active parser registry currently contains 25 trading program IDs; it does not imply that every Solana program, instruction, or transaction becomes a normalized trade.
- Recognized swaps from registered trading programs
- Transaction identity, slot and event time needed for evidence
- Dedicated instruction decoders for core venues plus generic flow parsing for supported programs
- Program families including Pump.fun, Raydium, Meteora, Orca Whirlpool, Phoenix and additional Solana venues
- Every Solana transaction is a swap or a product record
- Every program has a dedicated instruction-level parser
- A registered program equals one unique protocol
- Failed, malformed or unsupported activity receives guessed semantics
Raw activity is the input, not the product.
Conyr's differentiator is the state reconstructed above a trade feed: positions instead of isolated exits, funding paths instead of one-hop labels, and relationship graphs instead of unconnected wallet scores.
Market events
Decoded swaps, trades, price and OHLCV derived from recognized Solana trading programs. Records retain event time and transaction identity so downstream results can point back to chain evidence.
Position-level wallet accounting
Buys and sells are reconstructed into positions with FIFO cost basis, realized outcomes, remaining inventory, hold time, and entry/exit evidence. This is the layer required to distinguish a position win from one profitable partial exit.
Funding provenance
Inbound funding is walked backward toward recognized infrastructure origins. The result can include a root, path and hop count—or remain unresolved without inventing an attribution.
Behavior and similarity
Eligible wallets receive a multi-axis behavior fingerprint spanning timing, exit discipline, performance shape, sizing, infrastructure and provenance. Similarity means behavior, not identity or coordination.
Coordination and entities
Bundle structures, repeat leader-to-follower relationships, time-bounded token campaigns, wallet entities, crowd state, entanglement and wash-trade evidence are retained as scored relationships rather than flattened labels.
Delivery
The same intelligence families are exposed through REST, event streams, the Conyr app and 46 hosted MCP tools. Availability and historical depth depend on product tier.
Each layer has its own clock.
A streaming input does not make every aggregate instantaneous. These are operating semantics, not latency SLAs; planned maintenance, replay, upstream chain conditions and dependency health can extend them.
| Data family | Normal update behavior | Why it can lag or be absent |
|---|---|---|
| Decoded swaps and trades | As transactions are processed | Chain arrival, parser support, replay or resynchronization |
| Positions and realized PnL | As qualifying trades update a position | Incomplete cost basis, transfer inventory, unsupported paths |
| Funding provenance | As qualifying funding transfers are processed | Path ambiguity, history availability, recognized-root coverage |
| Wallet labels | Recomputed as qualifying evidence changes | Activity gates, evidence thresholds, cache interval |
| Behavior vectors | Target refresh approximately every 60 seconds | Activity gate, normalization refresh, index availability |
| Follow graph and campaigns | Near-real-time scored updates | Minimum support, lag window, pruning and retention |
| Bundle lifecycle | Hot-path detection and status updates | Supported structures, confidence thresholds, late evidence |
| Published scale measurements | Updated when formally measured | Treat registry values as audited evidence, not a live dashboard |
Unknown is not zero.
Conyr preserves the difference between no evidence, no matching result, a measured zero and a partial response. Consumers should keep that distinction instead of coercing every missing value to zero or neutral.
| Representation | Meaning |
|---|---|
| null / unavailable | The system cannot support a value with the evidence currently available. It does not mean zero. |
| empty list | The request succeeded but no retained records met the endpoint's scope and thresholds. |
| zero | The measured quantity is known and its value is zero within the stated scope. |
| partial: true | A composite response returned useful sections while one or more dependencies were unavailable or inapplicable. |
| unknown provenance | No recognized origin was resolved within the available path and 20 hops maximum depth. It does not mean self-funded. |
| no behavior vector | The wallet did not pass the activity gate or the vector service could not provide a current vector. It is not a low score. |
Coverage by intelligence family.
The following rules are the minimum context needed to interpret Conyr's strongest outputs correctly.
Addresses are not actors by default.
Conyr's latest published corpus measurement covered 3.4M+ active wallet addresses and 660K+ traded token mints. The live corpus continues to change. One actor can control many wallets; one wallet can be custodial or automated. A traded mint is covered activity, not an endorsement or a guarantee of complete metadata, liquidity or safety.
A sell is not automatically a won position.
Conyr reconstructs cost basis and position state before aggregating outcomes. Partial exits are allocated against inventory rather than counted as independent wins. Flash-swap and arbitrage handling reduce accounting artifacts. Transfer-acquired inventory, missing history or unsupported paths can still leave cost basis partial or unavailable; those cases must not be converted into fabricated profit.
Thin histories return no fingerprint.
The current gate requires at least 10 closed positions, at least one day of activity and at least 15 trades. Eligible vectors describe six groups: timing, exit discipline, performance shape, velocity and sizing, infrastructure, and provenance. Nearest neighbors are behaviorally similar; they are not necessarily connected or controlled by the same actor.
The root registry is broader than CEX wallets.
The maintained registry currently contains 136 exchange, custody, payment, service and other recognized infrastructure origins. The resolver searches backward up to 20 hops; it can terminate sooner or remain unknown. Infrastructure roots are useful provenance without implying every wallet funded by the same exchange belongs to one entity.
Structure, lifecycle and evidence stay separate.
The engine currently represents 10 classes, consolidated into five headline risk archetypes on some surfaces. Execution adjacency, shared funding and synthetic behavior are different evidence domains. Detection means the observed structure met retained thresholds; it does not prove malicious intent or common legal ownership.
A relationship is evidence, not an accusation.
Follow edges retain direction, repeated support, lag distribution and size resemblance. Campaigns are time-bounded token episodes; entities aggregate stronger shared evidence. Recognized infrastructure funders are suppressed where a shared service would otherwise create a false group. Results should be described as detected on-chain patterns, not declarations of intent.
Where confidence should stop.
These boundaries apply across the product even when an individual response is technically complete.
Coverage is parser-bounded
New or changed programs can appear before a parser supports their exact semantics. Generic flow analysis is not identical to a dedicated decoder.
History can be incomplete
A position or funding path that begins before retained or available history can lack the evidence needed for complete reconstruction.
Address identity is probabilistic
On-chain relationships can support an entity hypothesis but cannot by themselves prove beneficial ownership, identity or off-chain intent.
Labels can change
Behavioral and provenance classifications update as evidence accumulates. Downstream systems should retain the label timestamp and confidence.
Aggregate measurements
Scale claims retain explicit as-of dates. Use them as demonstrated production evidence rather than implying that a published value is a live dashboard counter.
Markets are adversarial
Operators adapt to known heuristics. Detection thresholds trade precision against recall and require continuous evaluation.
Short answers for the common category errors.
These answers are intentionally quotable and are also emitted as FAQ structured data.
Does Conyr cover every Solana transaction?
Conyr consumes a broad first-party Solana transaction stream, but its published 20M+-per-day claim counts successfully decoded swaps—not every transaction. Product coverage depends on recognized programs, parser support, record validity and the intelligence layer being requested.
Does an active-wallet count represent people?
No. A wallet is an on-chain address. One operator can control many addresses, and one address can be used by automation, custody infrastructure or multiple people.
Does a coordination edge prove two wallets share an owner?
No. It records scored on-chain evidence such as repeat timing and trade-size relationships. Common ownership or intent requires additional evidence and should not be inferred from an edge alone.
Why can a wallet have PnL but no behavior profile?
Position accounting can exist with less history. Behavior profiles are deliberately gated by closed positions, activity span and trade count so thin wallets do not receive misleading fingerprints.
Use the registry, not copied marketing text.
The public registry provides a citation, date, scope and limitations for every tracked headline value. Its JSON projection is safe for agents and does not expose private query shapes or database infrastructure.