puffinparse 0.1.4__tar.gz → 0.1.6__tar.gz

{puffinparse-0.1.4 → puffinparse-0.1.6}/Cargo.lock

@@ -1395,7 +1395,7 @@ dependencies = [
 
 [[package]]
 name = "puffinparse-cli"
-version = "0.1.4"
+version = "0.1.6"
 dependencies = [
  "anyhow",
  "chrono",
@@ -1415,7 +1415,7 @@ dependencies = [
 
 [[package]]
 name = "puffinparse-core"
-version = "0.1.4"
+version = "0.1.6"
 dependencies = [
  "async-trait",
  "bytes",
@@ -1439,7 +1439,7 @@ dependencies = [
 
 [[package]]
 name = "puffinparse-node"
-version = "0.1.4"
+version = "0.1.6"
 dependencies = [
  "bytes",
  "napi",
@@ -1453,7 +1453,7 @@ dependencies = [
 
 [[package]]
 name = "puffinparse-python"
-version = "0.1.4"
+version = "0.1.6"
 dependencies = [
  "bytes",
  "puffinparse-core",
@@ -1468,7 +1468,7 @@ dependencies = [
 
 [[package]]
 name = "puffinparse-server"
-version = "0.1.4"
+version = "0.1.6"
 dependencies = [
  "axum",
  "bytes",

{puffinparse-0.1.4 → puffinparse-0.1.6}/Cargo.toml

@@ -3,7 +3,7 @@ resolver = "2"
 members = ["crates/puffinparse-core", "crates/puffinparse-python"]
 
 [workspace.package]
-version = "0.1.4"
+version = "0.1.6"
 edition = "2021"
 rust-version = "1.80"
 license = "MIT"
@@ -15,7 +15,7 @@ keywords = ["ocr", "document-parsing", "pdf", "api", "benchmark"]
 categories = ["api-bindings", "text-processing"]
 
 [workspace.dependencies]
-puffinparse-core = { path = "crates/puffinparse-core", version = "0.1.4" }
+puffinparse-core = { path = "crates/puffinparse-core", version = "0.1.6" }
 reqwest = { version = "0.13", default-features = false, features = ["json", "multipart", "rustls", "stream", "gzip"] }
 tokio = { version = "1", features = ["rt-multi-thread", "macros", "fs", "time", "sync"] }
 serde = { version = "1", features = ["derive"] }

{puffinparse-0.1.4 → puffinparse-0.1.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: puffinparse
-Version: 0.1.4
+Version: 0.1.6
 Classifier: Development Status :: 3 - Alpha
 Classifier: Intended Audience :: Developers
 Classifier: License :: OSI Approved :: MIT License
@@ -61,7 +61,7 @@ Two things make switching real rather than aspirational. **Modes**: every call n
 
 | | |
 |---|---|
-| **Providers (v0.1)** | 18 providers · 60 models · 3 modes. **Live-verified** against the real APIs: [Reducto](https://reducto.ai), [Extend](https://extend.ai), [LlamaParse](https://cloud.llamaindex.ai). **Verified locally**: the self-hosted Tesseract and Docling. **Docs-only** (implemented from the provider's API documentation and tested against fixture payloads, not yet run live): Mistral, Azure, Textract, Gemini, OpenAI, Anthropic, Mathpix, Datalab, Unstructured, Upstage, Landing AI, Google Document AI and the self-hosted PaddleOCR ([help verify them](https://github.com/ajinkyashejul/puffinparse/issues/10)). [Full table](#model-names) |
+| **Providers (v0.1)** | 20 providers · 79 models · 3 modes. **Live-verified** against the real APIs: [Reducto](https://reducto.ai), [Extend](https://extend.ai), [LlamaParse](https://cloud.llamaindex.ai). **Verified locally**: the self-hosted Tesseract and Docling. **Docs-only** (implemented from the provider's API documentation and tested against fixture payloads, not yet run live): Mistral, Azure, Textract, Gemini, OpenAI, Anthropic, Mathpix, Datalab, Unstructured, Upstage, Landing AI, Google Document AI, OpenDocRouter and the self-hosted PaddleOCR and vLLM presets ([help verify them](https://github.com/ajinkyashejul/puffinparse/issues/10)). [Full table](#model-names) |
 | **Modes** | `parse` (markdown + blocks), `ocr` (plain text + boxes), `extract` (JSON from a schema) |
 | **Core** | Rust (`puffinparse-core`): `reqwest` + `tokio`, no vendor SDKs, `#![forbid(unsafe_code)]` |
 | **SDKs** | Python 3.9+ (sync + async, fully typed) and Node.js / TypeScript ([`js/`](https://github.com/ajinkyashejul/puffinparse/blob/main/js/README.md)), both on the same Rust core |
@@ -216,8 +216,9 @@ result = puffinparse.retrieve(job)              # Job (still pending) or ParseRe
 result = puffinparse.handle_webhook(request.json(), model="reducto")  # verify the signature first
 ```
 
-Reducto, Extend and LlamaParse support jobs; `webhook_url` maps to each provider's per-job webhook
-where one exists (Extend only has workspace-level webhooks, so it is rejected there). See
+Reducto, Extend, LlamaParse and OpenDocRouter support jobs; `webhook_url` maps to each provider's
+per-job webhook where one exists (Extend only has workspace-level webhooks and OpenDocRouter has
+none, so it is rejected there). See
 [`docs/SPEC.md`](https://github.com/ajinkyashejul/puffinparse/blob/main/docs/SPEC.md) §15.
 
 ### TypeScript / Node.js
@@ -332,6 +333,8 @@ page per provider.
 | `gemini/3.5-flash` | parse, ocr, extract | $0.00945 · $0.00945 · $0.00495 | Gemini 3.5 Flash: frontier Flash generation (GA 2026-05-19) |
 | `gemini/3.5-flash-lite` | parse, ocr, extract | $0.00245 · $0.00245 · $0.0012 | Gemini 3.5 Flash-Lite: low-latency 3.x tier (GA 2026-07-21) |
 | `gemini/3.8-flash` | parse, ocr, extract | $0.004125 · $0.004125 · $0.00225 | Gemini 3.8 Flash: newest Flash model (GA 2026-09-02, introductory pricing) |
+| `gemini/3.8-flash-low` | parse, ocr, extract | $0.004125 · $0.004125 · $0.00225 | Gemini 3.8 Flash with thinkingConfig.thinkingLevel=low (cheaper, faster transcription) |
+| `gemini/3-flash-preview` | parse, ocr, extract | $0.00315 · $0.00315 · $0.00165 | Gemini 3 Flash preview (gemini-3-flash-preview; PDF and image input) |
 
 **OpenAI** · `OPENAI_API_KEY` · docs-only — Vision transcription through the Responses API. Per-page prices are estimates, as for Gemini.
 
@@ -341,6 +344,7 @@ page per provider.
 | `openai/gpt-5.6-terra` | parse, ocr, extract | $0.0114 · $0.0114 · $0.0114 | OpenAI Responses API, gpt-5.6-terra (balanced capability/price) |
 | `openai/gpt-5.6-sol` | parse, ocr, extract | $0.02 · $0.02 · $0.02 | OpenAI Responses API, gpt-5.6-sol (flagship GPT-5.6) |
 | `openai/gpt-6-astra` | parse, ocr, extract | $0.05 · $0.05 · $0.05 | OpenAI Responses API, gpt-6-astra (most capable, most expensive) |
+| `openai/gpt-6-luna` | parse, ocr, extract | $0.0005 · $0.0005 · $0.0005 | OpenAI Responses API, gpt-6-luna (cheapest GPT-6, image and PDF input) |
 
 **Anthropic (Claude)** · `ANTHROPIC_API_KEY` · docs-only — Vision transcription through the Messages API. Per-page prices are estimates.
 
@@ -349,6 +353,8 @@ page per provider.
 | `anthropic/claude-sonnet-5` | parse `*`, ocr `*`, extract `*` | $0.01 · $0.01 · $0.01 | Claude Messages API, claude-sonnet-5 (balanced vision transcription) |
 | `anthropic/claude-haiku-4-5` | parse, ocr, extract | $0.005 · $0.005 · $0.005 | Claude Messages API, claude-haiku-4-5 (cheapest, 200K context) |
 | `anthropic/claude-opus-5` | parse, ocr, extract | $0.025 · $0.025 · $0.025 | Claude Messages API, claude-opus-5 (highest accuracy) |
+| `anthropic/claude-opus-5-5` | parse, ocr, extract | $0.02 · $0.02 · $0.02 | Claude Messages API, claude-opus-5-5 (current Opus; tool_choice auto, no forced tool) |
+| `anthropic/claude-haiku-5-5` | parse, ocr, extract | $0.0005 · $0.0005 · $0.0005 | Claude Messages API, claude-haiku-5-5 (cheapest Claude, 1M context) |
 
 **Mathpix** · `MATHPIX_APP_KEY` (+ `MATHPIX_APP_ID`) · docs-only — Maths-first OCR: Mathpix Markdown with LaTeX, line and word polygons.
 
@@ -395,13 +401,33 @@ page per provider.
 | `google_documentai/form` | parse, ocr, extract `*` | $0.03 · $0.03 · $0.03 | Form Parser processor: paragraphs + tables, entities as extraction |
 | `google_documentai/prebuilt` | parse, ocr, extract | $0.03 · $0.03 · $0.03 | Prebuilt or custom extractor (invoice, W2, ...): entities as extraction |
 
+**OpenDocRouter** · `OPEN_DOC_ROUTER_API_KEY` · docs-only — LlamaIndex's hosted router: frontier VLMs and open OCR models behind one endpoint, billed per token at the providers' prices, per-page status and charge, layout boxes on request. Model ids keep the router's own `<vendor>/<model>` after the provider name. Prices here are the site's average charge per page (price version 2026-10-06); `cost_usd` is the actual `charge_usd`.
+
+| Model | Modes | List price / page (parse · ocr · extract) | Notes |
+|---|---|---|---|
+| `opendocrouter/google/gemini-3.8-flash-low` | parse `*`, ocr `*` | $0.005908 · $0.005908 · — | Gemini 3.8 Flash (low thinking); the router's quickstart model |
+| `opendocrouter/google/gemini-3-flash` | parse, ocr | $0.019667 · $0.019667 · — | Gemini 3 Flash |
+| `opendocrouter/anthropic/claude-opus-5-5` | parse, ocr | $0.04882 · $0.04882 · — | Claude Opus 5.5 (frontier, most expensive) |
+| `opendocrouter/anthropic/claude-haiku-5-5` | parse, ocr | $0.001225 · $0.001225 · — | Claude Haiku 5.5 |
+| `opendocrouter/openai/gpt-5.6-terra` | parse, ocr | $0.019886 · $0.019886 · — | GPT-5.6 Terra |
+| `opendocrouter/openai/gpt-6-luna` | parse, ocr | $0.000798 · $0.000798 · — | GPT-6 Luna (cheapest frontier model) |
+| `opendocrouter/infly/infinity-parser2-flash` | parse, ocr | $0.004344 · $0.004344 · — | Infinity-Parser2-Flash (open model hosted by the router) |
+| `opendocrouter/opendatalab/mineru2.5-pro` | parse, ocr | $0.000861 · $0.000861 · — | MinerU2.5-Pro (open model hosted by the router) |
+| `opendocrouter/xingchen-agi/teleocr` | parse, ocr | $0.002702 · $0.002702 · — | TeleOCR (open model hosted by the router) |
+| `opendocrouter/rednote-hilab/dots.mocr` | parse, ocr | $0.00397 · $0.00397 · — | dots.mocr (open model hosted by the router) |
+| `opendocrouter/paddlepaddle/paddleocr-vl-1.6` | parse, ocr | $0.002111 · $0.002111 · — | PaddleOCR-VL-1.6 (open model hosted by the router) |
+
 **Self-hosted engines** · no key · $0/page — out-of-process: a local binary or a server you run. Tesseract and Docling are verified locally; PaddleOCR is docs-only.
+**Self-hosted engines** · no key · $0/page — out-of-process: a local binary or a server you run. Tesseract and Docling are verified locally; PaddleOCR and vLLM are docs-only.
 
 | Model | Modes | List price / page (parse · ocr · extract) | Notes |
 |---|---|---|---|
 | `tesseract/default` | ocr `*` (native), parse `*` | $0 · $0 · — | local `tesseract` binary (`TESSERACT_CMD`), PDFs via `pdftoppm`; word/line boxes + confidences, no layout model; verified locally |
 | `docling/default` | parse `*`, ocr `*` | $0 · $0 · — | your docling-serve (`DOCLING_BASE_URL`): layout, tables, OCR; verified locally |
 | `paddleocr/default` | ocr `*` (native), parse `*` (PP-StructureV3) | $0 · $0 · — | your PaddleOCR serving (`PADDLEOCR_BASE_URL`); docs-only |
+| `paddleocr/vl` | parse, ocr | $0 · $0 · — | your PaddleOCR-VL pipeline serving (`PADDLEOCR_VL_BASE_URL`; PaddleOCR-VL-1.6 by default): layout + 0.9B VLM; docs-only |
+| `vllm/infinity-parser2-flash` | parse `*`, ocr `*` | $0 · $0 · — | `infly/Infinity-Parser2-Flash` on your `vllm serve` (`VLLM_BASE_URL`): page images, layout blocks with boxes; docs-only |
+| `vllm/dots.mocr` | parse, ocr | $0 · $0 · — | `rednote-hilab/dots.mocr` on your `vllm serve` (`VLLM_BASE_URL`): page images, layout blocks with boxes; docs-only |
 
 ### Keep your Reducto / Extend / LlamaParse code
 

{puffinparse-0.1.4 → puffinparse-0.1.6}/README.md

@@ -27,7 +27,7 @@ Two things make switching real rather than aspirational. **Modes**: every call n
 
 | | |
 |---|---|
-| **Providers (v0.1)** | 18 providers · 60 models · 3 modes. **Live-verified** against the real APIs: [Reducto](https://reducto.ai), [Extend](https://extend.ai), [LlamaParse](https://cloud.llamaindex.ai). **Verified locally**: the self-hosted Tesseract and Docling. **Docs-only** (implemented from the provider's API documentation and tested against fixture payloads, not yet run live): Mistral, Azure, Textract, Gemini, OpenAI, Anthropic, Mathpix, Datalab, Unstructured, Upstage, Landing AI, Google Document AI and the self-hosted PaddleOCR ([help verify them](https://github.com/ajinkyashejul/puffinparse/issues/10)). [Full table](#model-names) |
+| **Providers (v0.1)** | 20 providers · 79 models · 3 modes. **Live-verified** against the real APIs: [Reducto](https://reducto.ai), [Extend](https://extend.ai), [LlamaParse](https://cloud.llamaindex.ai). **Verified locally**: the self-hosted Tesseract and Docling. **Docs-only** (implemented from the provider's API documentation and tested against fixture payloads, not yet run live): Mistral, Azure, Textract, Gemini, OpenAI, Anthropic, Mathpix, Datalab, Unstructured, Upstage, Landing AI, Google Document AI, OpenDocRouter and the self-hosted PaddleOCR and vLLM presets ([help verify them](https://github.com/ajinkyashejul/puffinparse/issues/10)). [Full table](#model-names) |
 | **Modes** | `parse` (markdown + blocks), `ocr` (plain text + boxes), `extract` (JSON from a schema) |
 | **Core** | Rust (`puffinparse-core`): `reqwest` + `tokio`, no vendor SDKs, `#![forbid(unsafe_code)]` |
 | **SDKs** | Python 3.9+ (sync + async, fully typed) and Node.js / TypeScript ([`js/`](https://github.com/ajinkyashejul/puffinparse/blob/main/js/README.md)), both on the same Rust core |
@@ -182,8 +182,9 @@ result = puffinparse.retrieve(job)              # Job (still pending) or ParseRe
 result = puffinparse.handle_webhook(request.json(), model="reducto")  # verify the signature first
 ```
 
-Reducto, Extend and LlamaParse support jobs; `webhook_url` maps to each provider's per-job webhook
-where one exists (Extend only has workspace-level webhooks, so it is rejected there). See
+Reducto, Extend, LlamaParse and OpenDocRouter support jobs; `webhook_url` maps to each provider's
+per-job webhook where one exists (Extend only has workspace-level webhooks and OpenDocRouter has
+none, so it is rejected there). See
 [`docs/SPEC.md`](https://github.com/ajinkyashejul/puffinparse/blob/main/docs/SPEC.md) §15.
 
 ### TypeScript / Node.js
@@ -298,6 +299,8 @@ page per provider.
 | `gemini/3.5-flash` | parse, ocr, extract | $0.00945 · $0.00945 · $0.00495 | Gemini 3.5 Flash: frontier Flash generation (GA 2026-05-19) |
 | `gemini/3.5-flash-lite` | parse, ocr, extract | $0.00245 · $0.00245 · $0.0012 | Gemini 3.5 Flash-Lite: low-latency 3.x tier (GA 2026-07-21) |
 | `gemini/3.8-flash` | parse, ocr, extract | $0.004125 · $0.004125 · $0.00225 | Gemini 3.8 Flash: newest Flash model (GA 2026-09-02, introductory pricing) |
+| `gemini/3.8-flash-low` | parse, ocr, extract | $0.004125 · $0.004125 · $0.00225 | Gemini 3.8 Flash with thinkingConfig.thinkingLevel=low (cheaper, faster transcription) |
+| `gemini/3-flash-preview` | parse, ocr, extract | $0.00315 · $0.00315 · $0.00165 | Gemini 3 Flash preview (gemini-3-flash-preview; PDF and image input) |
 
 **OpenAI** · `OPENAI_API_KEY` · docs-only — Vision transcription through the Responses API. Per-page prices are estimates, as for Gemini.
 
@@ -307,6 +310,7 @@ page per provider.
 | `openai/gpt-5.6-terra` | parse, ocr, extract | $0.0114 · $0.0114 · $0.0114 | OpenAI Responses API, gpt-5.6-terra (balanced capability/price) |
 | `openai/gpt-5.6-sol` | parse, ocr, extract | $0.02 · $0.02 · $0.02 | OpenAI Responses API, gpt-5.6-sol (flagship GPT-5.6) |
 | `openai/gpt-6-astra` | parse, ocr, extract | $0.05 · $0.05 · $0.05 | OpenAI Responses API, gpt-6-astra (most capable, most expensive) |
+| `openai/gpt-6-luna` | parse, ocr, extract | $0.0005 · $0.0005 · $0.0005 | OpenAI Responses API, gpt-6-luna (cheapest GPT-6, image and PDF input) |
 
 **Anthropic (Claude)** · `ANTHROPIC_API_KEY` · docs-only — Vision transcription through the Messages API. Per-page prices are estimates.
 
@@ -315,6 +319,8 @@ page per provider.
 | `anthropic/claude-sonnet-5` | parse `*`, ocr `*`, extract `*` | $0.01 · $0.01 · $0.01 | Claude Messages API, claude-sonnet-5 (balanced vision transcription) |
 | `anthropic/claude-haiku-4-5` | parse, ocr, extract | $0.005 · $0.005 · $0.005 | Claude Messages API, claude-haiku-4-5 (cheapest, 200K context) |
 | `anthropic/claude-opus-5` | parse, ocr, extract | $0.025 · $0.025 · $0.025 | Claude Messages API, claude-opus-5 (highest accuracy) |
+| `anthropic/claude-opus-5-5` | parse, ocr, extract | $0.02 · $0.02 · $0.02 | Claude Messages API, claude-opus-5-5 (current Opus; tool_choice auto, no forced tool) |
+| `anthropic/claude-haiku-5-5` | parse, ocr, extract | $0.0005 · $0.0005 · $0.0005 | Claude Messages API, claude-haiku-5-5 (cheapest Claude, 1M context) |
 
 **Mathpix** · `MATHPIX_APP_KEY` (+ `MATHPIX_APP_ID`) · docs-only — Maths-first OCR: Mathpix Markdown with LaTeX, line and word polygons.
 
@@ -361,13 +367,33 @@ page per provider.
 | `google_documentai/form` | parse, ocr, extract `*` | $0.03 · $0.03 · $0.03 | Form Parser processor: paragraphs + tables, entities as extraction |
 | `google_documentai/prebuilt` | parse, ocr, extract | $0.03 · $0.03 · $0.03 | Prebuilt or custom extractor (invoice, W2, ...): entities as extraction |
 
+**OpenDocRouter** · `OPEN_DOC_ROUTER_API_KEY` · docs-only — LlamaIndex's hosted router: frontier VLMs and open OCR models behind one endpoint, billed per token at the providers' prices, per-page status and charge, layout boxes on request. Model ids keep the router's own `<vendor>/<model>` after the provider name. Prices here are the site's average charge per page (price version 2026-10-06); `cost_usd` is the actual `charge_usd`.
+
+| Model | Modes | List price / page (parse · ocr · extract) | Notes |
+|---|---|---|---|
+| `opendocrouter/google/gemini-3.8-flash-low` | parse `*`, ocr `*` | $0.005908 · $0.005908 · — | Gemini 3.8 Flash (low thinking); the router's quickstart model |
+| `opendocrouter/google/gemini-3-flash` | parse, ocr | $0.019667 · $0.019667 · — | Gemini 3 Flash |
+| `opendocrouter/anthropic/claude-opus-5-5` | parse, ocr | $0.04882 · $0.04882 · — | Claude Opus 5.5 (frontier, most expensive) |
+| `opendocrouter/anthropic/claude-haiku-5-5` | parse, ocr | $0.001225 · $0.001225 · — | Claude Haiku 5.5 |
+| `opendocrouter/openai/gpt-5.6-terra` | parse, ocr | $0.019886 · $0.019886 · — | GPT-5.6 Terra |
+| `opendocrouter/openai/gpt-6-luna` | parse, ocr | $0.000798 · $0.000798 · — | GPT-6 Luna (cheapest frontier model) |
+| `opendocrouter/infly/infinity-parser2-flash` | parse, ocr | $0.004344 · $0.004344 · — | Infinity-Parser2-Flash (open model hosted by the router) |
+| `opendocrouter/opendatalab/mineru2.5-pro` | parse, ocr | $0.000861 · $0.000861 · — | MinerU2.5-Pro (open model hosted by the router) |
+| `opendocrouter/xingchen-agi/teleocr` | parse, ocr | $0.002702 · $0.002702 · — | TeleOCR (open model hosted by the router) |
+| `opendocrouter/rednote-hilab/dots.mocr` | parse, ocr | $0.00397 · $0.00397 · — | dots.mocr (open model hosted by the router) |
+| `opendocrouter/paddlepaddle/paddleocr-vl-1.6` | parse, ocr | $0.002111 · $0.002111 · — | PaddleOCR-VL-1.6 (open model hosted by the router) |
+
 **Self-hosted engines** · no key · $0/page — out-of-process: a local binary or a server you run. Tesseract and Docling are verified locally; PaddleOCR is docs-only.
+**Self-hosted engines** · no key · $0/page — out-of-process: a local binary or a server you run. Tesseract and Docling are verified locally; PaddleOCR and vLLM are docs-only.
 
 | Model | Modes | List price / page (parse · ocr · extract) | Notes |
 |---|---|---|---|
 | `tesseract/default` | ocr `*` (native), parse `*` | $0 · $0 · — | local `tesseract` binary (`TESSERACT_CMD`), PDFs via `pdftoppm`; word/line boxes + confidences, no layout model; verified locally |
 | `docling/default` | parse `*`, ocr `*` | $0 · $0 · — | your docling-serve (`DOCLING_BASE_URL`): layout, tables, OCR; verified locally |
 | `paddleocr/default` | ocr `*` (native), parse `*` (PP-StructureV3) | $0 · $0 · — | your PaddleOCR serving (`PADDLEOCR_BASE_URL`); docs-only |
+| `paddleocr/vl` | parse, ocr | $0 · $0 · — | your PaddleOCR-VL pipeline serving (`PADDLEOCR_VL_BASE_URL`; PaddleOCR-VL-1.6 by default): layout + 0.9B VLM; docs-only |
+| `vllm/infinity-parser2-flash` | parse `*`, ocr `*` | $0 · $0 · — | `infly/Infinity-Parser2-Flash` on your `vllm serve` (`VLLM_BASE_URL`): page images, layout blocks with boxes; docs-only |
+| `vllm/dots.mocr` | parse, ocr | $0 · $0 · — | `rednote-hilab/dots.mocr` on your `vllm serve` (`VLLM_BASE_URL`): page images, layout blocks with boxes; docs-only |
 
 ### Keep your Reducto / Extend / LlamaParse code
 

{puffinparse-0.1.4 → puffinparse-0.1.6}/crates/puffinparse-core/src/bench.rs

@@ -20,7 +20,10 @@ pub use tables::{extract_tables, teds_grid, Grid};
 /// * `1` — initial scorer (pipe tables only, exact rule matching, headline in `char_similarity`).
 /// * `2` — HTML tables, `teds_grid`, punctuation-spacing-insensitive rule matching, fuzzy
 ///   `bag_of_sentences`, entity decoding, explicit `Summary::score`.
-pub const SCORER_VERSION: u32 = 2;
+/// * `3` — single `*` / `_` emphasis stripped, inline HTML tags (`<sup>`, `<i>`) no longer split
+///   words, dot leaders collapsed, figure markup and image descriptions dropped from transcript
+///   predictions ([`strip_figures`]), and `table_cell` rules can match the header row.
+pub const SCORER_VERSION: u32 = 3;
 
 /// A `bag_of_sentences` sentence counts as present when some window of the prediction is within
 /// this normalised character similarity of it (`1 - edit_distance / |sentence|`).
@@ -67,6 +70,7 @@ pub fn normalize(s: &str, opts: NormalizeOptions) -> String {
         .replace(['\u{2018}', '\u{2019}'], "'")
         .replace(['\u{201C}', '\u{201D}'], "\"")
         .replace(['\u{2013}', '\u{2014}'], "-");
+    let s = collapse_dot_leaders(&s);
     let s = if opts.case_insensitive { s.to_lowercase() } else { s };
     let mut out = String::with_capacity(s.len());
     let mut last_space = true;
@@ -86,6 +90,98 @@ pub fn normalize(s: &str, opts: NormalizeOptions) -> String {
     out.trim().to_string()
 }
 
+/// Dot leaders (`Chapter 1 ........ 7`, `. . . .`) become one space: whether a parser keeps the
+/// dots of a table of contents says nothing about its reading. Four or more dots, optionally
+/// single-spaced; a three-dot ellipsis is text and stays.
+fn collapse_dot_leaders(s: &str) -> String {
+    let chars: Vec<char> = s.chars().collect();
+    let mut out = String::with_capacity(s.len());
+    let mut i = 0;
+    while i < chars.len() {
+        if chars[i] == '.' {
+            // Walk a run of dots separated by at most one space.
+            let (mut j, mut dots, mut end) = (i, 0, i);
+            while j < chars.len() {
+                if chars[j] == '.' {
+                    dots += 1;
+                    j += 1;
+                    end = j;
+                } else if chars[j] == ' ' && chars.get(j + 1) == Some(&'.') {
+                    j += 1;
+                } else {
+                    break;
+                }
+            }
+            if dots >= 4 {
+                out.push(' ');
+                i = end;
+                continue;
+            }
+        }
+        out.push(chars[i]);
+        i += 1;
+    }
+    out
+}
+
+/// Figure content a parser *adds*: `<figure>…</figure>` blocks (a description or caption a vision
+/// model wrote) and markdown images (`![description](src)`). Transcript truths carry no figure or
+/// chart content (DP-Bench drops them, as upstream ignores predictions inside figure regions), so
+/// this text is not a reading of the page and is not scored. Tags are matched case-insensitively;
+/// an unclosed `<figure>` runs to the end.
+pub fn strip_figures(s: &str) -> String {
+    let lower = s.to_ascii_lowercase();
+    let mut out = String::with_capacity(s.len());
+    let mut i = 0;
+    while let Some(rel) = lower[i..].find("<figure") {
+        let start = i + rel;
+        // `<figure>` or `<figure ...>`, not `<figures>`.
+        let next = lower.as_bytes().get(start + 7).copied();
+        if !matches!(next, Some(b'>' | b' ' | b'\t' | b'\n' | b'\r' | b'/')) {
+            out.push_str(&s[i..start + 7]);
+            i = start + 7;
+            continue;
+        }
+        out.push_str(&s[i..start]);
+        out.push(' ');
+        i = match lower[start..].find("</figure>") {
+            Some(e) => start + e + "</figure>".len(),
+            None => s.len(),
+        };
+    }
+    out.push_str(&s[i..]);
+    strip_markdown_images(&out)
+}
+
+/// `![alt](src)` → nothing (`[` without a following `](…)` is kept as text).
+fn strip_markdown_images(s: &str) -> String {
+    let mut out = String::with_capacity(s.len());
+    let mut rest = s;
+    while let Some(at) = rest.find("![") {
+        let after = &rest[at + 2..];
+        let image_end = after.find("](").and_then(|close| {
+            let alt = &after[..close];
+            if alt.contains('\n') {
+                return None;
+            }
+            after[close + 2..].find(')').map(|p| at + 2 + close + 2 + p + 1)
+        });
+        match image_end {
+            Some(end) => {
+                out.push_str(&rest[..at]);
+                out.push(' ');
+                rest = &rest[end..];
+            }
+            None => {
+                out.push_str(&rest[..at + 2]);
+                rest = &rest[at + 2..];
+            }
+        }
+    }
+    out.push_str(rest);
+    out
+}
+
 /// Scores for one (prediction, truth) pair. All in `0..=1` unless noted.
 #[derive(Debug, Clone, Copy, PartialEq, Serialize, Deserialize, Default)]
 pub struct Metrics {
@@ -127,6 +223,7 @@ pub struct Metrics {
 
 /// Compute all metrics for a prediction vs truth (both raw; normalisation applied internally).
 pub fn score(pred: &str, truth: &str, opts: NormalizeOptions) -> Metrics {
+    let pred = &strip_figures(pred);
     let p = normalize(pred, opts);
     let t = normalize(truth, opts);
     let p_chars: Vec<char> = p.chars().collect();
@@ -918,7 +1015,9 @@ fn check_table_cell(cache: &PredictionCache<'_>, cell: &CellRule, cs: bool, k: u
             }
             None => None,
         };
-        for row in &table.rows {
+        // The header row is a row too: a cell rule may be about a header cell (olmOCR's neighbour
+        // tests are), and a parser that marks the header correctly must not lose that rule.
+        for row in std::iter::once(&table.header).chain(&table.rows) {
             if !row_matches(row, row_header.as_deref(), k) {
                 continue;
             }
@@ -1040,6 +1139,44 @@ pub fn metrics_from_rules(score: &RuleScore) -> Metrics {
 mod tests {
     use super::*;
 
+    #[test]
+    fn scorer_v3_dot_leaders() {
+        let o = NormalizeOptions::default();
+        assert_eq!(normalize("Intro ........ 7", o), "intro 7");
+        assert_eq!(normalize("Intro . . . . . 7", o), "intro 7");
+        assert_eq!(normalize("Wait... what", o), "wait... what");
+        assert_eq!(score("Intro 7", "Intro .......... 7", o).char_similarity, 1.0);
+    }
+
+    #[test]
+    fn scorer_v3_strips_figures_from_transcripts() {
+        assert_eq!(strip_figures("a <figure><caption>A bar chart</caption></figure> b"), "a   b");
+        assert_eq!(strip_figures("a <FIGURE id=\"f\">x</Figure> b"), "a   b");
+        assert_eq!(strip_figures("<figures> stay"), "<figures> stay");
+        assert_eq!(strip_figures("see ![chart of sales](img.png) here"), "see   here");
+        assert_eq!(strip_figures("[link](x) and ![ no close"), "[link](x) and ![ no close");
+        let o = NormalizeOptions::default();
+        let truth = "Revenue grew.";
+        let pred = "Revenue grew.\n<figure><caption>The chart shows growth</caption></figure>";
+        assert_eq!(score(pred, truth, o).char_similarity, 1.0);
+    }
+
+    #[test]
+    fn scorer_v3_emphasis_and_inline_tags() {
+        let o = NormalizeOptions::default();
+        assert_eq!(score("An *italic* word", "An italic word", o).char_similarity, 1.0);
+        assert_eq!(score("the 9<sup>th</sup> day", "the 9th day", o).char_similarity, 1.0);
+    }
+
+    #[test]
+    fn scorer_v3_table_cell_can_match_the_header_row() {
+        let pred = "| Name | Value |\n|---|---|\n| a | 1 |";
+        let c = CellRule { value: "Value".into(), row_header: Some("Name".into()), col_header: None };
+        let r = Rule { cell: Some(c), ..rule("hdr", RuleType::TableCell) };
+        let got = score_rules(pred, &[r], NormalizeOptions::default());
+        assert_eq!(got.passed, 1, "{got:?}");
+    }
+
     #[test]
     fn levenshtein_basic() {
         let a: Vec<char> = "kitten".chars().collect();

{puffinparse-0.1.4 → puffinparse-0.1.6}/crates/puffinparse-core/src/http.rs

@@ -106,7 +106,7 @@ where
     }
 }
 
-fn is_transient(e: &Error) -> bool {
+pub(crate) fn is_transient(e: &Error) -> bool {
     matches!(e.kind, ErrorKind::RateLimit | ErrorKind::Network)
         || (e.kind == ErrorKind::Provider && matches!(e.status_code, Some(500 | 502 | 503 | 504)))
 }