---
title: "Find duplicate clusters in a list of strings"
method: POST
path: "/v1/matching/deduplicate"
tags: ["matching"]
---

# Find duplicate clusters in a list of strings

`POST /v1/matching/deduplicate`

## Request

- Base URL: `https://api.snipget.ai` (the document declares no server; this is the origin it was published from)
- URL: `https://api.snipget.ai/v1/matching/deduplicate`
- Auth: none declared

Cluster a list of strings by fuzzy similarity. Returns groups of likely duplicates with scores. Useful for data cleaning: 'which of these entity names are probably the same thing?' Tip: for organization or facility names, run /v1/names/org-name/clean on the inputs first so casing and punctuation don't depress scores. This is raw fuzzy string matching, not semantic — heavy abbreviations ('Univ' vs 'University', 'SLU') may still not match; lower the threshold or pre-expand abbreviations for those. Billing: 1 cost unit per item in the list, like every batch endpoint (e.g. 1,000 items = 1,000 units). The response's meta.cost_units shows the exact charge.

## Query parameters

- `include_receipt` boolean

## Headers

- `authorization` string, nullable
- `X-API-Key` string, nullable

## Request body

- DeduplicateRequest
  - `items` string[], required: List of strings to deduplicate.
  - `strategy` string: Similarity strategy: ratio (character-level), partial_ratio (best substring), token_sort_ratio (ignores word order), token_set_ratio (ignores order + duplicates, default).
  - `threshold` number: Similarity threshold for clustering (0.0–1.0).
  - `include_trace` boolean

## Response `200`

Successful Response

- MatchingDeduplicateResponse: Success envelope for `/v1/matching/deduplicate` — `result` carries this utility's precise shape.
  - `status` string
  - `confidence` number, required
  - `result` object, required: Result of :func:`deduplicate`. Key-complete on every ok path since bug #165 (items=[] used to omit ``unique_count``); the invalid-strategy / out-of-range-threshold rejects are ok=False request errors with no result payload at all.
    - `clusters` MatchingDeduplicateCluster[], required
      - `canonical` string, required
      - `index` integer, required
      - `duplicates` MatchingDeduplicateScoredCandidate[], required
        - `value` string, required
        - `score` number, required
        - `index` integer, required
    - `cluster_count` integer, required
    - `duplicate_count` integer, required
    - `strategy` string, required
    - `unique_count` integer, required
  - `meta` ResponseMeta: Metadata attached to every successful response.
    - `version` string
    - `build` string
    - `elapsed_ms` integer
    - `cost_units` integer: Cost units charged for this call: one per item processed (a batch of N items = N; live-registry lookups bill 2 per item). A call is one request; cost units are the billing measure.
    - `request_id` string, nullable
    - `trace` string[], nullable
    - `rate_limit_remaining` integer, nullable
    - `rate_limit_reset` integer, nullable
    - `quota_remaining` integer, nullable: Cost units left in the plan's monthly included bucket before this call's charge lands (None on unlimited tiers).
    - `quota_reset` integer, nullable: Unix timestamp when the monthly included bucket resets.
    - `credit_remaining_usd` number, nullable: Prepaid overage-allowance balance in USD, populated once the included cost units are used up and this call is drawing on it.
    - `receipt` ReceiptMeta: Opt-in execution receipt (ADR #68 B1) — caller-verifiable hashes. Requested with the ``include_receipt=true`` query parameter (free, like ``include_trace``). Reproduce either hash with any RFC 8785 (JCS) library: ``input_hash = sha256(jcs({"tool": <request path>, "args": <your JSON body minus include_trace/include_receipt>}))``; ``result_hash = sha256(jcs(result))``. For deterministic tools (everything except the generators) the same input on the same ``build`` reproduces ``result_hash`` — disputes are honored by replay.
      - `receipt_id` string, required
      - `build` string, required
      - `input_hash` string, nullable
      - `result_hash` string, nullable
      - `error` string, nullable
      - `message` string, nullable

## Other responses

- `400`: Input was syntactically valid but semantically unusable (`INVALID_INPUT`). Not retryable without changing the input.
- `401`: Missing or unrecognized API key. Send it as `Authorization: Bearer <key>` or `X-API-Key: <key>`. Keys are issued at https://snipget.ai.
- `403`: The key is valid but this caller may not use it: the request IP is outside the key's allowlist (`IP_NOT_ALLOWED`) or the account is deactivated (`ACCOUNT_DEACTIVATED`).
- `422`: The request body failed schema validation (`INVALID_REQUEST`). `details` lists the per-field errors. Not retryable without changing the request.
- `429`: Two distinct conditions share this status — branch on `error_code`. **`RATE_LIMITED`** is a per-second throughput throttle: retry after `retry_after_seconds` (the precise float; the `Retry-After` header is the same value rounded up). **`QUOTA_EXCEEDED`** is monthly capacity and is NOT retryable: recovery depends on `limit_type` — upgrade (`monthly_quota`), buy an overage allowance (`included_exhausted`), or top up (`overage_balance_exhausted`).
- `503`: A transient, retryable service condition — branch on `error_code`. **`MAINTENANCE_MODE`**: an admin-toggled maintenance window (auth, portal, and webhook surfaces stay up). **`UPSTREAM_UNAVAILABLE`**: a utility's external data source (PubChem, RxNorm, ClinicalTrials.gov, the FX feed, NPPES…) is down, blocking us, or timed out. **`UPSTREAM_RATE_LIMITED`**: that external source is throttling Snipget (it returned a 429) — the caller's own rate is fine, so this is NOT our `RATE_LIMITED` (429); it's a service-side throttle. All three are retryable; wait for `retry_after_seconds` / the `Retry-After` header when present.

---

[API](https://skmtc.dev/snipget/apis/snipget-api.md) · [All operations](https://skmtc.dev/snipget/apis/snipget-api/llms.txt) · [OpenAPI document](https://skmtc.dev/snipget/apis/snipget-api/revisions/ed1077fb1bcb?raw)