Netflix measures the blast radius of its internal APIs, and publishes none

Netflix measures the blast radius of its internal APIs, and publishes none

Netflix has described the tool its library producers consult before they break something. In Beyond API Compatibility: Building a Tool to Measure the Blast Radius of Library Changes, Roberto Perez Alcolea of the Java Platform Team introduces Quasar, an index of symbol definitions and references across Netflix repositories built on the SCIP interchange format. Every enrolled CI build submits an index, Quasar writes definitions and consumer references as Parquet in S3, and during pull-request review an Athena query joins the changed symbols against every repository that still calls them. The illustrative report shows a removed method with 137 affected repositories, 86 on version 4.2.1, 38 on 4.1.7 and 13 still on 3.9.4, with sample file paths and line numbers attached.

The argument separates two questions that compatibility tooling usually collapses. A binary compatibility checker says whether a change breaks the contract. Quasar says who is standing in the blast. “An incompatible change is not automatically a high-risk change,” the post says, and the inverse is the lesson it repeats most: “An agent should not turn ‘no observed usages in the available index’ into ‘safe to remove.’” The report therefore carries its own evidence quality, whether the base and head commits were indexed, how fresh the consumer indexes are, and whether results were truncated, and refuses to substitute a default-branch snapshot when a commit is missing. Quasar indexes JVM, Go, Python, JavaScript, TypeScript and Rust, compares JVM and Protobuf APIs, and exposes its findings through REST, GraphQL and an MCP server. All of it is internal. The closing line is the one worth keeping: “Library producers need more than a compatibility verdict. They need to know who may be affected and have enough time to act before release.”

The catalog has a short answer for Netflix, and it is the honest one. The record states that Netflix does not publish a general public consumer API, and its one API page is Netflix Open Connect, the ISP-facing CDN program. The open-source edge tier lives under separate records, and Netflix Zuul, which the story URL first resolved to, scores 14.3 on four pages of gateway, filter and push-messaging surface with no contract behind any of them.

The Kin Score for Netflix is 10.6, minimal band, with discoverability at 57.1 and every contract facet at 0.0. The Agent Readiness score is 0.6, human-only, with only the rate limit signal lit. There is no tension to manufacture here, because the post is about library APIs inside one company and the catalog measures the surface a company offers the world, and Netflix offers almost none. What transfers is the method. Quasar’s refusal to let an empty index read as a clean result is the same rule the catalog has to live by: a provider with no spec on record and a provider that publishes no spec look identical from the outside, and the number has to say which it measured.

← NationBuilder documents six JSON:API parameters, and 108 acting operations have no idempotency
Statsig's Experiments API lists seven URLs twice, and the two contracts disagree →