Proof
Claims you can re-run.
Every benchmark number on this page comes from a report in the repository — with the fixtures and the command to reproduce it. If a measurement can’t be re-run, it doesn’t go on this page.
Re-run from the repository root
pnpm bench:e2eWhat is checked, and how
- 3benchmark corpora, from 80 to 1,500 files, with reports in the repo
- 25example apps built and smoke-checked on relevant changes
- 25of them driven in a real browser on a weekly schedule
- 30decision records, alternatives included
01 — Benchmarks
Extraction, measured in the frameworks you use.
The benchmark times the whole job: scanning source, extracting messages, writing output. Tools that also update catalogs are compared like for like; React Intl only extracts, so it is marked as narrower scope. Every run is checked for the same messages, not just timed. Palamedes’ cached re-run is shown on its own — a capability, not a race.
End-to-end extract + catalog update
React · Three checked corpora
2026-08-14 · Node v24.19.0 · darwin/arm64 · median of 7
| Corpus | Palamedes cold | Palamedes warm · 5 files touched | Same-scope tools |
|---|---|---|---|
| 80 files, 640 messages | 12 ms | 9 ms | 27–51× slower |
| 240 files, 1920 messages | 21 ms | 13 ms | 24–50× slower |
| 1,500 files (750 with i18n), ~400k lines, 6,000 messages | 73 ms | 47 ms | 30–100× slower |
Checked result ledger · React
Realistic corpus — 1,500 files across ~400k lines, 6,000 messages (median of 7 runs)
| Workflow | Result | Relative time |
|---|---|---|
| Palamedes v1.17.3extract + catalog update | 47–73 ms* | 1× |
| React Intl v6.16.16extraction only · narrower scope | 424 ms | 5× |
| Lingui v6.6.0extract + catalog update | 2.2 s | 30× |
| fbtee v3.0.1collect + two-catalog update · two CLI commands | 7.3 s | 100× |
| General Translation v2.16.4extract + catalog update | 5.1 s | 70× |
| i18next-cli v1.69.0extract + catalog update | 5.8 s | 80× |
* 47 ms is Palamedes on a cached re-run after 5 changed source files; 73 ms is the cold workflow result. No speedup factor is calculated for the non-comparable cached run. Cache details →
Times are rounded to display precision and relative factors are rounded down. Exact medians remain in the checked report. Machine-local run: darwin/arm64, Node v24.19.0, 2026-08-14, median of 7 runs.
Framework workflows
Vue, Svelte, and Solid.
Real .vue, .svelte, and .tsx components, with the same logical messages and existing English/German catalogs. Each timed sample checks the active message set, unchanged translations, and source immutability. This is a separate run from the React report above.
Wuchale also transforms components and compiles catalogs, so its command covers more work. Times include CLI startup, discovery, parsing, catalog merging, and writes. Local tool state is reset for each run; the OS page cache is not flushed. Plain strings and interpolation only; no plurals, rich text, runtime rendering, or watch-mode measurements. Solid uses the 1.9 adapters supported by these tools.
Apple M1 Ultra · darwin/arm64 · Node v24.22.0 · 2026-10-23
3 warmups · median of 7 runs · rounded display values
Vue
| Workflow | Median |
|---|---|
| Palamedes v1.25.1extract + catalog update | 103 ms |
| Lingui v6.6.0extract + catalog update | 2.8 s |
| i18next-cli v1.72.1extract + catalog update | 4.3 s |
| Intlayer v9.7.0extract + update + compile | Not timed Catalog checks failed |
Intlayer JSON receipt: 6,450 written entries for 6,000 current messages; 4,215/4,950 unchanged German translations retained.
Small corpus · 640 messages
| Workflow | Median |
|---|---|
| Palamedes v1.25.1extract + catalog update | 12 ms |
| Lingui v6.6.0extract + catalog update | 767 ms |
| i18next-cli v1.72.1extract + catalog update | 419 ms |
| Intlayer v9.7.0extract + update + compile | Not timed Catalog checks failed |
Intlayer JSON receipt: 688 written entries for 640 current messages; 450/528 unchanged German translations retained.
Medium corpus · 1,920 messages
| Workflow | Median |
|---|---|
| Palamedes v1.25.1extract + catalog update | 19 ms |
| Lingui v6.6.0extract + catalog update | 906 ms |
| i18next-cli v1.72.1extract + catalog update | 623 ms |
| Intlayer v9.7.0extract + update + compile | Not timed Catalog checks failed |
Intlayer JSON receipt: 2,064 written entries for 1,920 current messages; 1,344/1,584 unchanged German translations retained.
Svelte
| Workflow | Median |
|---|---|
| Palamedes v1.25.1extract + catalog update | 107 ms |
| Lingui v6.6.0extract + catalog update | 2.3 s |
| i18next-cli v1.72.1extract + catalog update | 11.9 s |
| Wuchale v0.26.7extract + update + compile | 5.6 s |
| Intlayer v9.7.0extract + update + compile | Not timed Catalog checks failed |
Intlayer JSON receipt: 6,450 written entries for 6,000 current messages; 4,215/4,950 unchanged German translations retained.
Small corpus · 640 messages
| Workflow | Median |
|---|---|
| Palamedes v1.25.1extract + catalog update | 12 ms |
| Lingui v6.6.0extract + catalog update | 946 ms |
| i18next-cli v1.72.1extract + catalog update | 552 ms |
| Wuchale v0.26.7extract + update + compile | 305 ms |
| Intlayer v9.7.0extract + update + compile | Not timed Catalog checks failed |
Intlayer JSON receipt: 688 written entries for 640 current messages; 450/528 unchanged German translations retained.
Medium corpus · 1,920 messages
| Workflow | Median |
|---|---|
| Palamedes v1.25.1extract + catalog update | 18 ms |
| Lingui v6.6.0extract + catalog update | 1.1 s |
| i18next-cli v1.72.1extract + catalog update | 811 ms |
| Wuchale v0.26.7extract + update + compile | 407 ms |
| Intlayer v9.7.0extract + update + compile | Not timed Catalog checks failed |
Intlayer JSON receipt: 2,064 written entries for 1,920 current messages; 1,344/1,584 unchanged German translations retained.
Solid
| Workflow | Median |
|---|---|
| Palamedes v1.25.1extract + catalog update | 109 ms |
| Lingui v6.6.0extract + catalog update | 2.6 s |
| i18next-cli v1.72.1extract + catalog update | 6.8 s |
| Wuchale v0.26.7extract + update + compile | 4.2 s |
| Intlayer v9.7.0extract + update + compile | Not timed Catalog checks failed |
Intlayer JSON receipt: 6,450 written entries for 6,000 current messages; 4,215/4,950 unchanged German translations retained.
Small corpus · 640 messages
| Workflow | Median |
|---|---|
| Palamedes v1.25.1extract + catalog update | 13 ms |
| Lingui v6.6.0extract + catalog update | 557 ms |
| i18next-cli v1.72.1extract + catalog update | 358 ms |
| Wuchale v0.26.7extract + update + compile | 187 ms |
| Intlayer v9.7.0extract + update + compile | Not timed Catalog checks failed |
Intlayer JSON receipt: 688 written entries for 640 current messages; 450/528 unchanged German translations retained.
Medium corpus · 1,920 messages
| Workflow | Median |
|---|---|
| Palamedes v1.25.1extract + catalog update | 20 ms |
| Lingui v6.6.0extract + catalog update | 643 ms |
| i18next-cli v1.72.1extract + catalog update | 567 ms |
| Wuchale v0.26.7extract + update + compile | 268 ms |
| Intlayer v9.7.0extract + update + compile | Not timed Catalog checks failed |
Intlayer JSON receipt: 2,064 written entries for 1,920 current messages; 1,344/1,584 unchanged German translations retained.
Why Intlayer has a receipt instead of a timing
The timed lane uses Intlayer 9.7.0 with .content.json dictionaries. It retains removed entries and loses unchanged interpolated translations in this fixture. A separate untimed smoke check also covers .content.ts: TypeScript preserves those translations, while both formats retain removed entries. These findings cover the pinned version and configurations.
| Framework / format | Retained | Stale entries |
|---|---|---|
| Vue.content.json | 17/25 | 3 |
| Vue.content.ts | 25/25 | 3 |
| Svelte.content.json | 17/25 | 3 |
| Svelte.content.ts | 25/25 | 3 |
| Solid.content.json | 17/25 | 3 |
| Solid.content.ts | 25/25 | 3 |
Re-run all three frameworks
pnpm benchmark:framework-workflow02 — Verification
25 apps built on every relevant change. 25 driven in a real browser every week.
“Works with your framework” is easy to say. Here it means an application that is built and smoke-checked — and, for the browser-verified examples, loaded, switched between languages, and captured as a screenshot you can diff.
01
Build
All 25 example apps build and smoke-test against the workspace packages on relevant PRs and main pushes — no mocked integrations.
02
Drive
25 browser-capable examples run the Playwright flow weekly or on manual dispatch: load, SSR output, locale switch, and localized server actions.
03
Capture
Screenshots are versioned in the repo, so 'works across frameworks' is a diffable artifact, not a slide.




03 — Catalog quality
Fast would be worthless if the catalogs were wrong.
Catalog semantics live in one dedicated engine (ferrocat): parsing, merging, structured audits, and ICU authoring diagnostics. The benchmark harness validates every tool run semantically — message inventories are compared, not just timed.
Structured audits
Machine-readable catalog audits catch missing translations, stale entries, and metadata drift in CI.
ICU diagnostics
Authoring mistakes in plural/select syntax are flagged at extract time, not at runtime in production.
Semantic merging
A Git merge driver resolves catalog conflicts by meaning, not by line — no more broken .po files after rebases.
04 — ICU semantics
ICU stays ICU from source to runtime.
A checked-in nested select + plural message runs through extraction, macro transformation, PO catalog update, catalog compilation, and six executions of the transformed runtime function. The proof compares exact messages and selector structure instead of treating ICU as a yes/no feature checkbox.
05 — Decision trail
30 decisions, written down before you depend on them.
The ADRs cover message identity, the native boundary, adapter architecture — and, just as deliberately, what Palamedes refuses to own. Reading them is the fastest way to know if our tradeoffs match yours.
- 001Project Scope and Open PositioningAccepted
- 002Rust-First Core with Thin Host AdaptersAccepted
- 003Source-String-First Message IdentityAccepted
- 004Generated Catalogs Use Executable Message FunctionsAccepted
- 005Public Application Runtime Is Parser-FreeAccepted
- 006Universal `getI18n()` Runtime ModelAccepted
- 007Locale Is Fixed for a Browser DocumentAccepted
- 008Framework Adapter ArchitectureAccepted
- 009Ferrocat as Catalog and ICU FoundationAccepted
- 010Translation Augmentation BoundaryAccepted
- 011Editorial Visual Brand SystemAccepted
- 012Native CLI And YAML-First ConfigurationAccepted
- 013Native Boundary and DistributionAccepted
- 014The npm Launcher Is A Packaging Bridge With A Node-Free Escape HatchAccepted
- 015Host Explicit Binary CLI Plugins In RustAccepted
- 016Independent Update Checks and Opt-In CLI Usage ReportingAccepted
- 017Typed N-API Boundary with Workflow-First Native OperationsAccepted
- 018Generated TypeScript Types Derived from the Native Binding SurfaceAccepted
- 019Native Catalog Rendering Produces Bundler ModulesAccepted
- 020Internal Compiled Lookup KeysAccepted
- 021Runtime Formatter Subset DiagnosticsAccepted
- 022Extraction CacheAccepted
- 023Bounded Parallel ExtractionAccepted
- 024Native Transform Source MapsAccepted
- 025Binary Plugin Protocol For Rust-Hosted ExtensionsAccepted
- 026No React Router RSC AdapterAccepted
- 027Shared Cross-Repository Site UIAccepted
- 028README Theme CompositionAccepted
- 029Effective Flow Project SetupAccepted
- 030Next Production Client Catalogs Outside the Bundler GraphProposed