Deployer Reputation: supplied-edge heuristic

Scores and groups caller-supplied launch records. No Solana RPC fetch, wallet lookup, verified ownership attribution, fraud finding, calibrated probability, or investment recommendation. Shared non-exchange funding creates an association, not proof of a common operator. Known exchange funders never join distinct deployers, even when only one record flags that funder. The same deployer's launches can still group together. Caller labels are trusted as input, not verified.

Input and result

All transports use the same validation and scorer. Submit an object containing edges:

{"edges":[{"deployer":"wallet-A","funder":"wallet-B","mint":"unique-mint-A","outcome":"rugged","funder_is_cex":false}]}

Identifiers are required nonempty opaque strings, max 128 characters, without whitespace/control characters or Unicode surrogates. They are not checked as Solana addresses. Supply one record per distinct mint; missing, duplicate and conflicting mints are rejected. Batches contain 1–1000 edges. Unknown fields are rejected. Optional fields:

• outcome: rugged (mapped to rug), rug, alive, graduated, unknown (default).
• funder_is_cex: strict boolean, default false. Mark exchange/infra funders accurately.
• lamports: nonnegative integer up to 18446744073709551615; default 0.
• block_time: nonnegative Unix integer up to 253402300799, or null.

Scoring returns clusters, heuristic scores 0–1, bands, components, and raw launch/label counts. Grouping returns associated deployers/funders/mints and launch counts. Structural components may raise scores without outcome labels. The inherited CEX component is a heuristic, not evidence of misconduct. Research AUC, lift and p-values from separate studies do not validate this scorer. No performance or profitability claim is made. Launch timestamps currently do not affect scoring; the component called cadence measures launch count, not elapsed time.

MCP stdio

Python standard library only; HTTP/Actor dependencies are unnecessary for stdio:

python deployer_reputation_mcp.py
python deployer_reputation_mcp.py --selftest

Tools: deployer_reputation and cluster_launches. Newline-delimited JSON-RPC objects; 512000-byte maximum line. No JSON-RPC batches. Malformed messages return errors and the process continues. Notifications have no response. Advertised protocol: 2024-11-05. mcp-client-config.json contains the client registration example; replace its absolute path.

HTTP deployment

pip install -r requirements.txt
# Set REPUTATION_API_KEY to a securely generated value of at least 32 characters.
uvicorn api:app --host 127.0.0.1 --port 8080 --workers 1 --limit-concurrency 16 --timeout-keep-alive 5

GET /health is public. POST /score and /cluster require X-API-Key. Missing/trivial server key prevents startup. Actual streamed body size is limited to 512000 bytes; edge count is independently limited. Invalid auth/JSON/input/body size return 401/400/422/413. Use TLS at a reverse proxy before public exposure; retain request-header/body timeouts and a rate limit there. Bind Docker's port to loopback behind that proxy, e.g. -p 127.0.0.1:8080:8080. Dockerfile runs as an unprivileged user, one worker, 16 concurrent connections. The app does not store submitted edges or log request bodies. Infrastructure operators can still observe requests; avoid secrets in edge metadata. No billing, customer account system, API-key issuance, or monetization is implemented. Deployment status is external to this package.

Apify Actor packaging

.actor/actor.json, Dockerfile.actor, and actor.py package the same scorer. Input adds optional operation: score (default) or cluster. The default dataset is the sole result destination: one object per delivered batch, regardless of its cluster count. No result is copied into a key-value record. Read the dataset to determine whether a result was delivered; a successful process exit alone does not prove delivery or payment. Invalid batches fail the run before any result is persisted. No chain requests are made. Apify stores submitted input/output: account/platform access and retention rules apply. Platform pricing is configured separately from this source. The live Actor build 1.1.5 configuration on 2026-09-21 charges $0.005 per analysis result through the synthetic apify-default-dataset-item event, with no start or custom event. Its minimum run charge cap is $0.00501. Confirm current pricing in Apify before running; use an explicit maxTotalChargeUsd cap (for example 0.01). Do not add a custom charge for the same result. Scorer 2.0.0 is deployed as build 1.1.6 (2026-09-21), superseding baseline 1.1.5. It passed independent QA (22 tests three times, old-code negative controls, 200-batch numeric parity) and cloud validation. Owner tests do not establish external customer payment or settled revenue. The output schema links only the dataset, with an overview view.

Validation

pip install httpx
python -m unittest discover -v

Regression tests cover rug mapping, mint accounting, exchange separation, type/range limits, stdio recovery, notification silence, API authentication/body limits, and Actor scorer parity. These tests establish software behavior, not predictive validity.

Offline demo and verification

python scripts/demo_reputation.py     # golden evaluation kit: serial rugger vs clean vs shared-CEX
python scripts/verify_all.py          # unit suites plus demo kit, exits non-zero on failure

examples/demo_edges.json bundles three scenarios with expected outputs; examples/apify_input/ holds paste-ready fictional Actor inputs. All fictional, offline, no RPC or spending.

Use as an integration (dataset input)

Supply edges by dataset reference instead of inline JSON: set datasetId, or trigger this Actor from another Actor's Integrations tab — the platform supplies payload.resource.defaultDatasetId automatically. Only Apify platform datasets are fetched (alphanumeric IDs, batched reads), never arbitrary URLs or caller credentials. Upstream launch-tracker/scraper Actors produce the deployer/funder/mint rows; this Actor clusters and scores them. Inline edges always win over a dataset reference. Example static input for the integration form:

{"datasetId": "{{resource.defaultDatasetId}}", "operation": "score"}

Cost math (published pricing)

Flat $0.005 per completed analysis batch (up to 1,000 edges), charged through the platform's synthetic apify-default-dataset-item event; invalid batches fail before any result and are not charged. That is $0.005 for one launch or for a thousand — $0.000005 per launch at full batch, against $0.0005–0.02 per single token for Store scanners that fetch chain data themselves. Minimum run charge cap is $0.00501. Prices are configured in the Apify Console and may change there; check the Store page for the current values.

Scorer 2.0.0 migration (deployed as build 1.1.6)

IDs now use CL2- plus full SHA-256 of the canonical JSON array of sorted unique role-prefixed deployer/funder identifiers (ensure_ascii=True, compact separators). This replaces ambiguous delimiter serialization and 32-bit CL- IDs. Recompute stored IDs from original membership; old IDs cannot be converted reliably. IDs describe supplied membership, not persistent operator identity: adding/removing a funder or deployer changes the ID; changing only mint/outcome data does not.

Numeric weights and score calculation remain unchanged. High additionally requires at least one positive supplied rug label; otherwise high numeric scores remain elevated. Bands are policy labels, not calibrated probabilities. Global CEX classification now applies consistently to evidence and components, which can change scores for conflicting CEX flags. Caller input remains unchanged. evidence.labelCoverage is labeled launches / launches; notes flag absent or incomplete outcomes. Low does not mean safe. Score results include scorerVersion; both operations include clusterIdVersion.