v1 vs v2 at a glance

The scannable version. For the narrative, see What changed from 1.0 to 2.0; for the upgrade steps, see Migrating to v2.

The 2.0 major is a small, clean break (four removed legacy paths, output unchanged). The big delta is the 1.x capability accumulation in the second table.

What breaks in 2.0

Removed in 2.0	Replacement	Action needed
`GOLDENMATCH_IDENTITY_ID_SCHEME` + legacy `:hash:` lookup bridge	Canonical `:h1:` scheme (only scheme)	Run `goldenmatch identity migrate-ids` before upgrading if you persist an identity DB. Un-fingerprintable rows keep `:hash:`.
`GOLDENMATCH_CLUSTER_FRAMES_OUT`	Arrow frames-out path (the only path now)	Remove the env var; it is silently ignored. Output is unchanged.
`RunHistory.cheapest_healthy()`	`pick_committed()`	Rename the call.
`_scale_aware_backend` (internal shim)	`auto_configure_df` / `dedupe_df` (v3 planner)	Switch to the public API if you imported it directly.

Nothing else in the public API changed. dedupe(), match(), pprl_link(), evaluate() behave identically.

Capabilities: 1.0 vs 2.0

Area	GoldenMatch 1.0 (2026-03-23)	GoldenMatch 2.0 (2026-06-14)
Core API	`dedupe` / `match` / `pprl_link` / `evaluate`, 96 exports, 21 CLI commands, REST + MCP	Same surface, unchanged
Config	Zero-config + YAML schema	Zero-config + the v3 introspective planner (auto backend/plan, refuses low-confidence at scale)
Identity	Run-local cluster IDs only	Durable Identity Graph (v1.15): stable `entity_id`s, evidence edges, event log, 6 surfaces
Single-box scale	Polars in-memory	Planner-selected `bucket` path + signed-off native Rust kernels (`GOLDENMATCH_NATIVE=auto`)
Distributed	Ray backend (basic)	Phase-5 streaming pipeline: 100M rows end-to-end, ~0.3 GB driver RSS; validated on a real 5-node cluster
Probabilistic matching	Scorer-level	Splink-class Fellegi-Sunter that beats hand-rolled Splink on every shared-evaluator dataset (zero training)
Acceleration	Pure Python / Polars	Optional `goldenmatch[native]` Rust/Arrow kernels; bit-parity to the Python reference
SQL	—	In-database UDFs for DuckDB / Postgres / DataFusion
Other engines	Ray only	Opt-in Sail (Spark Connect) tier; opt-in WASM acceleration for the TypeScript port

Get started

Concepts

GoldenMatch

GoldenCheck

GoldenFlow

GoldenPipe

GoldenAnalysis

InferMap

SQL extensions

Reference

Research

What breaks in 2.0

Capabilities: 1.0 vs 2.0

See also

​What breaks in 2.0

​Capabilities: 1.0 vs 2.0

​See also

What breaks in 2.0

Capabilities: 1.0 vs 2.0

See also