#Updates

#Breaking Changes

• Remove @open-spaced-repetition/binding-win32-ia32-msvc and the i686-pc-windows-msvc build target. Native 32-bit Windows builds are no longer supported; Windows x64 and ARM64 packages remain available. (#429) Thanks @ishiko732!
• Require Node.js 24 or later and use an ESM default loader. Threaded WASI initialization is exposed through ./dynamic; ./dynamic-wasi remains an alias. (#406, #431) Thanks @ishiko732!
• Resolve timezones in Rust for CSV conversion instead of accepting a JavaScript offset callback. Use convertCsvToFsrsItems(data, nextDayStartsAt, timezoneOrOffset, modelVersion?), with an IANA timezone name or a fixed UTC offset in minutes; IANA offsets are resolved per review timestamp, including daylight saving transitions. (#384, #550) Thanks @ishiko732!
• Default optimizer/model workflows to FSRS-7, with fast stability and fractional study-day intervals. Set modelVersion: 'FSRS-6' explicitly for FSRS-6 training/evaluation and pass 'FSRS-6' to CSV conversion to retain whole study-day intervals. Use the same model version throughout conversion, training, and evaluation. new FSRSBinding() (or an empty parameter array) also defaults to FSRS-7; pass an explicit FSRS-6 parameter vector to retain FSRS-6 in that constructor. (#530) Thanks @ishiko732!
• Require an explicit positive numeric decay for computeOptimalSteps: FSRS-4 uses 1, FSRS-4.5/5 use 0.5, and FSRS-6 uses weights[20]. This single-decay fitting API does not support the FSRS-7 mixture curve. (#538) Thanks @ishiko732!
• Memory-state JSON serialization now includes stability_fast. Update strict stored-data schemas and preserve this field for FSRS-7; do not treat an FSRS-6 memory state as a complete FSRS-7 state. (#530) Thanks @ishiko732!

#New Features

• FSRSBindingReview.deltaT preserves fractional elapsed days, and FSRSBinding.nextStates accepts fractional days for FSRS-7 while retaining whole-day behavior for FSRS-6. BindingMemoryState(stability, difficulty, stabilityFast?) adds a stabilityFast getter; the optional constructor argument defaults to stability for compatibility. (#530) Thanks @ishiko732!
• Add the threadless @open-spaced-repetition/binding-wasm32-wasip1 package, including its /workerd and /wasm exports. The trainer can run across Node.js, browser Workers, and Cloudflare Workers/workerd using the appropriate native or WebAssembly entry. The threadless WASIp1 build requires no COOP/COEP headers or shared memory, so static sites hosted on GitHub Pages can run training locally in a browser Worker without a training backend. See the optimizer guide and live playground. Training and time-series evaluation retain Promise-based APIs but execute synchronously on the current worker; use a dedicated Worker in browsers. Progress callbacks are supported, and returning false stops the operation. (#432, #433) Thanks @ishiko732!
• Accept ReadableStream<Uint8Array> input for CSV conversion. (#447) Thanks @ishiko732!
• Add convertCsvToFsrsItemsWithCardIds and card IDs in training/evaluation workflows, preserving card identity across converted review items. (#553) Thanks @ishiko732!
• Add enableSchedPenalties for FSRS-7 training and evaluation. (#558) Thanks @ishiko732!

#Fixes and Dependencies

• Validate optimizer inputs and bound review-history expansion. FSRSBindingItem rejects review ratings outside 1–4, including Manual, and supplied cardIds must match the item count. (#550, #553) Thanks @ishiko732!
• Fix dynamic loader cleanup after asynchronous work and update native/WASI dependencies and binding validation. (#536, #594) Thanks @ishiko732!
• Incorporate upstream training/evaluation fixes and update fsrs-rs to revision d7f337b. (#571, #600) Thanks @ishiko732!

#Minor Changes

• #381 18562c4 Thanks @ishiko732! - chore(binding): upgrade fsrs to the published 6.5.0 crate and support passing trainingConfig to parameter training APIs

#Minor Changes

• #379 7479e74 Thanks @ishiko732! - chore(binding): remove the Burn dependency. Upgrade fsrs to the published 6.4.0 crate (no longer requires the custom burn fork), drop the [patch.crates-io.burn]/[patch.crates-io.burn-train] workspace patches, and delete the burn-v0.17.1-sys-metrics.wasm32 patch.

#Patch Changes

• #364 c50e027 Thanks @ishiko732! - feat(binding): implement memoryStateFromSM2 method for FSRSBinding

#Minor Changes

• #325 948e3cd Thanks @ishiko732! - feat(binding): add computeOptimalSteps to analyze revlog and recommend optimal learning/relearning steps

  Compute optimal learning/relearning steps from CSV revlog data. Analyzes review history by rating groups, fits forgetting curves via golden-section search, and recommends step durations based on desired retention and decay parameters.

#Patch Changes

• #327 e089f2f Thanks @ishiko732! - fix: Build the binding packages with Rust 1.94.1 and invalidate Turborepo caches when CI configuration changes.

• #329 5cd8949 Thanks @ishiko732! - fix: include README, localized package docs, and license files in the published binding tarball

#Patch Changes

• #319 d80230e Thanks @ishiko732! - fix(binding): worker and WASM path issues in the export flow to prevent Webpack and Turbopack resolution errors.

• #319 d80230e Thanks @ishiko732! - fix(binding): bump @emnapi/core and @emnapi/runtime to ^1.9.0 to avoid Cannot read properties of undefined (reading 'whenLoaded') error.

  ref: https://github.com/toyobayashi/emnapi/issues/202

• #319 d80230e Thanks @ishiko732! - feat(binding): support errorEvent option to emit custom events for errors in WASI worker.

  When enabled, worker errors are captured and dispatched as napi-rs-worker-error CustomEvent on window, allowing applications to handle WASM panics gracefully in the browser.

  ref: https://github.com/toyobayashi/emnapi/issues/203

#Minor Changes

• #272 23ec7a9 Thanks @ishiko732! - refactor(binding): refactor progress callback to support training interruption

  BREAKING CHANGES:

  • progress callback no longer supports async functions, because call_with_return_value does not support asynchronous execution. See Node.js N-API documentation: https://nodejs.org/api/n-api.html#n_api_napi_call_threadsafe_function

• #309 7c0d97c Thanks @ishiko732! - feat(binding): add initOptimizer dynamic loader for custom wasm/worker resources

  • Export ./dynamic-wasi entry from binding with initOptimizer(options) for Node.js and browser
  • Export ./wasi-worker and ./wasm entries from binding-wasm32-wasi for resolving worker and wasm paths
  • Allows users to provide wasm binary (Buffer, URL, Response) and worker (factory, instance, path) externally

#Patch Changes

• #271 1f8cf5e Thanks @ishiko732! - add universal_metrics method for FSRS evaluation

• #313 36c6066 Thanks @ishiko732! - fix(binding): avoid wasm dead-code warnings in compute_parameters task

• #269 ae7775c Thanks @ishiko732! - add evaluation method

• #295 014ad22 Thanks @ishiko732! - add evaluate_with_time_series_splits

#Patch Changes

• #241 f2b700d Thanks @ishiko732! - Fix missing build artifacts in the binding package and ensure .js, .cjs, .mjs, and .d.ts files are properly included in the distributed package.

#Minor Changes

• #237 c1d61fe Thanks @ishiko732! - # @open-spaced-repetition/binding v0.1.0

  🎉 Release the first version of the high-performance FSRS optimizer based on fsrs-rs and napi-rs.

  #Features

  • High-Performance Parameter Optimization: Native Rust implementation for computationally intensive FSRS parameter optimization
  • Multi-Platform Support: Pre-built binaries for 11 platforms:
    • Windows: x64, ia32, arm64 (MSVC)
    • macOS: x64, arm64 (Darwin)
    • Linux: x64/arm64 (GNU/MUSL)
    • Android: arm64
    • WebAssembly: wasm32-wasip1-threads
  • Core APIs:
    • computeParameters(): Optimize FSRS parameters from review history
    • convertCsvToFsrsItems(): Convert CSV review logs to FSRS data format
    • FSRSBinding: FSRS scheduler instance for memory state calculations

  #Requirements

  • Node.js >= 20.0.0
  • For WebAssembly: Additional package manager configuration required

  #Limitations

  • ⚠️ Beta Notice: API may change in future releases
  • Cannot run in edge-runtime environments (edge-runtime does not support WASI)

  #Documentation

  See README.md for installation instructions and usage examples.