Publish NativePort web-access API benchmark dataset

Initial public dataset release with CC BY 4.0 licensing, dataset card, normalized CSV/JSONL artifacts, provenance, reproducibility scripts, and tests.

LICENSE

@@ -0,0 +1,157 @@
+Creative Commons Attribution 4.0 International
+
+ Creative Commons Corporation (“Creative Commons”) is not a law firm and does not provide legal services or legal advice. Distribution of Creative Commons public licenses does not create a lawyer-client or other relationship. Creative Commons makes its licenses and related information available on an “as-is” basis. Creative Commons gives no warranties regarding its licenses, any material licensed under their terms and conditions, or any related information. Creative Commons disclaims all liability for damages resulting from their use to the fullest extent possible.
+
+Using Creative Commons Public Licenses
+
+Creative Commons public licenses provide a standard set of terms and conditions that creators and other rights holders may use to share original works of authorship and other material subject to copyright and certain other rights specified in the public license below. The following considerations are for informational purposes only, are not exhaustive, and do not form part of our licenses.
+
+Considerations for licensors: Our public licenses are intended for use by those authorized to give the public permission to use material in ways otherwise restricted by copyright and certain other rights. Our licenses are irrevocable. Licensors should read and understand the terms and conditions of the license they choose before applying it. Licensors should also secure all rights necessary before applying our licenses so that the public can reuse the material as expected. Licensors should clearly mark any material not subject to the license. This includes other CC-licensed material, or material used under an exception or limitation to copyright. More considerations for licensors.
+
+Considerations for the public: By using one of our public licenses, a licensor grants the public permission to use the licensed material under specified terms and conditions. If the licensor’s permission is not necessary for any reason–for example, because of any applicable exception or limitation to copyright–then that use is not regulated by the license. Our licenses grant only permissions under copyright and certain other rights that a licensor has authority to grant. Use of the licensed material may still be restricted for other reasons, including because others have copyright or other rights in the material. A licensor may make special requests, such as asking that all changes be marked or described. Although not required by our licenses, you are encouraged to respect those requests where reasonable. More considerations for the public.
+
+Creative Commons Attribution 4.0 International Public License
+
+By exercising the Licensed Rights (defined below), You accept and agree to be bound by the terms and conditions of this Creative Commons Attribution 4.0 International Public License ("Public License"). To the extent this Public License may be interpreted as a contract, You are granted the Licensed Rights in consideration of Your acceptance of these terms and conditions, and the Licensor grants You such rights in consideration of benefits the Licensor receives from making the Licensed Material available under these terms and conditions.
+
+Section 1 – Definitions.
+
+     a. Adapted Material means material subject to Copyright and Similar Rights that is derived from or based upon the Licensed Material and in which the Licensed Material is translated, altered, arranged, transformed, or otherwise modified in a manner requiring permission under the Copyright and Similar Rights held by the Licensor. For purposes of this Public License, where the Licensed Material is a musical work, performance, or sound recording, Adapted Material is always produced where the Licensed Material is synched in timed relation with a moving image.
+
+     b. Adapter's License means the license You apply to Your Copyright and Similar Rights in Your contributions to Adapted Material in accordance with the terms and conditions of this Public License.
+
+     c. Copyright and Similar Rights means copyright and/or similar rights closely related to copyright including, without limitation, performance, broadcast, sound recording, and Sui Generis Database Rights, without regard to how the rights are labeled or categorized. For purposes of this Public License, the rights specified in Section 2(b)(1)-(2) are not Copyright and Similar Rights.
+
+     d. Effective Technological Measures means those measures that, in the absence of proper authority, may not be circumvented under laws fulfilling obligations under Article 11 of the WIPO Copyright Treaty adopted on December 20, 1996, and/or similar international agreements.
+
+     e. Exceptions and Limitations means fair use, fair dealing, and/or any other exception or limitation to Copyright and Similar Rights that applies to Your use of the Licensed Material.
+
+     f. Licensed Material means the artistic or literary work, database, or other material to which the Licensor applied this Public License.
+
+     g. Licensed Rights means the rights granted to You subject to the terms and conditions of this Public License, which are limited to all Copyright and Similar Rights that apply to Your use of the Licensed Material and that the Licensor has authority to license.
+
+     h. Licensor means the individual(s) or entity(ies) granting rights under this Public License.
+
+     i. Share means to provide material to the public by any means or process that requires permission under the Licensed Rights, such as reproduction, public display, public performance, distribution, dissemination, communication, or importation, and to make material available to the public including in ways that members of the public may access the material from a place and at a time individually chosen by them.
+
+     j. Sui Generis Database Rights means rights other than copyright resulting from Directive 96/9/EC of the European Parliament and of the Council of 11 March 1996 on the legal protection of databases, as amended and/or succeeded, as well as other essentially equivalent rights anywhere in the world.
+
+     k. You means the individual or entity exercising the Licensed Rights under this Public License. Your has a corresponding meaning.
+
+Section 2 – Scope.
+
+     a. License grant.
+
+          1. Subject to the terms and conditions of this Public License, the Licensor hereby grants You a worldwide, royalty-free, non-sublicensable, non-exclusive, irrevocable license to exercise the Licensed Rights in the Licensed Material to:
+
+               A. reproduce and Share the Licensed Material, in whole or in part; and
+
+               B. produce, reproduce, and Share Adapted Material.
+
+          2. Exceptions and Limitations. For the avoidance of doubt, where Exceptions and Limitations apply to Your use, this Public License does not apply, and You do not need to comply with its terms and conditions.
+
+          3. Term. The term of this Public License is specified in Section 6(a).
+
+          4. Media and formats; technical modifications allowed. The Licensor authorizes You to exercise the Licensed Rights in all media and formats whether now known or hereafter created, and to make technical modifications necessary to do so. The Licensor waives and/or agrees not to assert any right or authority to forbid You from making technical modifications necessary to exercise the Licensed Rights, including technical modifications necessary to circumvent Effective Technological Measures. For purposes of this Public License, simply making modifications authorized by this Section 2(a)(4) never produces Adapted Material.
+
+          5. Downstream recipients.
+
+               A. Offer from the Licensor – Licensed Material. Every recipient of the Licensed Material automatically receives an offer from the Licensor to exercise the Licensed Rights under the terms and conditions of this Public License.
+
+               B. No downstream restrictions. You may not offer or impose any additional or different terms or conditions on, or apply any Effective Technological Measures to, the Licensed Material if doing so restricts exercise of the Licensed Rights by any recipient of the Licensed Material.
+
+          6.  No endorsement. Nothing in this Public License constitutes or may be construed as permission to assert or imply that You are, or that Your use of the Licensed Material is, connected with, or sponsored, endorsed, or granted official status by, the Licensor or others designated to receive attribution as provided in Section 3(a)(1)(A)(i).
+
+     b. Other rights.
+
+          1. Moral rights, such as the right of integrity, are not licensed under this Public License, nor are publicity, privacy, and/or other similar personality rights; however, to the extent possible, the Licensor waives and/or agrees not to assert any such rights held by the Licensor to the limited extent necessary to allow You to exercise the Licensed Rights, but not otherwise.
+
+          2. Patent and trademark rights are not licensed under this Public License.
+
+          3. To the extent possible, the Licensor waives any right to collect royalties from You for the exercise of the Licensed Rights, whether directly or through a collecting society under any voluntary or waivable statutory or compulsory licensing scheme. In all other cases the Licensor expressly reserves any right to collect such royalties.
+
+Section 3 – License Conditions.
+
+Your exercise of the Licensed Rights is expressly made subject to the following conditions.
+
+     a. Attribution.
+
+          1. If You Share the Licensed Material (including in modified form), You must:
+
+               A. retain the following if it is supplied by the Licensor with the Licensed Material:
+
+                    i. identification of the creator(s) of the Licensed Material and any others designated to receive attribution, in any reasonable manner requested by the Licensor (including by pseudonym if designated);
+
+                    ii. a copyright notice;
+
+                    iii. a notice that refers to this Public License;
+
+                    iv. a notice that refers to the disclaimer of warranties;
+
+                    v. a URI or hyperlink to the Licensed Material to the extent reasonably practicable;
+
+               B. indicate if You modified the Licensed Material and retain an indication of any previous modifications; and
+
+               C. indicate the Licensed Material is licensed under this Public License, and include the text of, or the URI or hyperlink to, this Public License.
+
+          2. You may satisfy the conditions in Section 3(a)(1) in any reasonable manner based on the medium, means, and context in which You Share the Licensed Material. For example, it may be reasonable to satisfy the conditions by providing a URI or hyperlink to a resource that includes the required information.
+
+          3. If requested by the Licensor, You must remove any of the information required by Section 3(a)(1)(A) to the extent reasonably practicable.
+
+          4. If You Share Adapted Material You produce, the Adapter's License You apply must not prevent recipients of the Adapted Material from complying with this Public License.
+
+Section 4 – Sui Generis Database Rights.
+
+Where the Licensed Rights include Sui Generis Database Rights that apply to Your use of the Licensed Material:
+
+     a. for the avoidance of doubt, Section 2(a)(1) grants You the right to extract, reuse, reproduce, and Share all or a substantial portion of the contents of the database;
+
+     b. if You include all or a substantial portion of the database contents in a database in which You have Sui Generis Database Rights, then the database in which You have Sui Generis Database Rights (but not its individual contents) is Adapted Material; and
+
+     c. You must comply with the conditions in Section 3(a) if You Share all or a substantial portion of the contents of the database.
+
+For the avoidance of doubt, this Section 4 supplements and does not replace Your obligations under this Public License where the Licensed Rights include other Copyright and Similar Rights.
+
+Section 5 – Disclaimer of Warranties and Limitation of Liability.
+
+     a. Unless otherwise separately undertaken by the Licensor, to the extent possible, the Licensor offers the Licensed Material as-is and as-available, and makes no representations or warranties of any kind concerning the Licensed Material, whether express, implied, statutory, or other. This includes, without limitation, warranties of title, merchantability, fitness for a particular purpose, non-infringement, absence of latent or other defects, accuracy, or the presence or absence of errors, whether or not known or discoverable. Where disclaimers of warranties are not allowed in full or in part, this disclaimer may not apply to You.
+
+     b. To the extent possible, in no event will the Licensor be liable to You on any legal theory (including, without limitation, negligence) or otherwise for any direct, special, indirect, incidental, consequential, punitive, exemplary, or other losses, costs, expenses, or damages arising out of this Public License or use of the Licensed Material, even if the Licensor has been advised of the possibility of such losses, costs, expenses, or damages. Where a limitation of liability is not allowed in full or in part, this limitation may not apply to You.
+
+     c. The disclaimer of warranties and limitation of liability provided above shall be interpreted in a manner that, to the extent possible, most closely approximates an absolute disclaimer and waiver of all liability.
+
+Section 6 – Term and Termination.
+
+     a. This Public License applies for the term of the Copyright and Similar Rights licensed here. However, if You fail to comply with this Public License, then Your rights under this Public License terminate automatically.
+
+     b. Where Your right to use the Licensed Material has terminated under Section 6(a), it reinstates:
+
+          1. automatically as of the date the violation is cured, provided it is cured within 30 days of Your discovery of the violation; or
+
+          2. upon express reinstatement by the Licensor.
+
+     c. For the avoidance of doubt, this Section 6(b) does not affect any right the Licensor may have to seek remedies for Your violations of this Public License.
+
+     d. For the avoidance of doubt, the Licensor may also offer the Licensed Material under separate terms or conditions or stop distributing the Licensed Material at any time; however, doing so will not terminate this Public License.
+
+     e. Sections 1, 5, 6, 7, and 8 survive termination of this Public License.
+
+Section 7 – Other Terms and Conditions.
+
+     a. The Licensor shall not be bound by any additional or different terms or conditions communicated by You unless expressly agreed.
+
+     b. Any arrangements, understandings, or agreements regarding the Licensed Material not stated herein are separate from and independent of the terms and conditions of this Public License.
+
+Section 8 – Interpretation.
+
+     a. For the avoidance of doubt, this Public License does not, and shall not be interpreted to, reduce, limit, restrict, or impose conditions on any use of the Licensed Material that could lawfully be made without permission under this Public License.
+
+     b. To the extent possible, if any provision of this Public License is deemed unenforceable, it shall be automatically reformed to the minimum extent necessary to make it enforceable. If the provision cannot be reformed, it shall be severed from this Public License without affecting the enforceability of the remaining terms and conditions.
+
+     c. No term or condition of this Public License will be waived and no failure to comply consented to unless expressly agreed to by the Licensor.
+
+     d. Nothing in this Public License constitutes or may be interpreted as a limitation upon, or waiver of, any privileges and immunities that apply to the Licensor or You, including from the legal processes of any jurisdiction or authority.
+
+Creative Commons is not a party to its public licenses. Notwithstanding, Creative Commons may elect to apply one of its public licenses to material it publishes and in those instances will be considered the “Licensor.” Except for the limited purpose of indicating that material is shared under a Creative Commons public license or as otherwise permitted by the Creative Commons policies published at creativecommons.org/policies, Creative Commons does not authorize the use of the trademark “Creative Commons” or any other trademark or logo of Creative Commons without its prior written consent including, without limitation, in connection with any unauthorized modifications to any of its public licenses or any other arrangements, understandings, or agreements concerning use of licensed material. For the avoidance of doubt, this paragraph does not form part of the public licenses.
+
+Creative Commons may be contacted at creativecommons.org.

Makefile

@@ -0,0 +1,41 @@
+# NativePort Web-Access API Benchmarks — build, test and validate.
+# Standard library only: no pip install, no third-party tools, no network access.
+#
+#   make build      regenerate data/ from the snapshot
+#   make test       run the unittest suite
+#   make validate   parse, hash and scan the package, print a report
+#   make check      all three, in order
+#   make hashes     print SHA-256 for the source snapshot and every artifact
+#
+# Point the build at another snapshot with:  make check EVALS=/path/to/evals.json
+#
+# scripts/validate.py is an internal review tool and is not in the public upload set,
+# so "make validate" reports and skips when it is absent instead of failing.
+
+PYTHON ?= python3
+EVALS  ?= /tmp/nativeport-evals-current.json
+DATA   ?= data
+
+.PHONY: check build test validate hashes help
+
+check: build test validate
+
+build:
+	$(PYTHON) scripts/build_dataset.py --input $(EVALS) --output-dir $(DATA)
+
+test:
+	NP_EVALS_JSON=$(EVALS) $(PYTHON) -m unittest discover -s tests -v
+
+validate:
+	@if [ -f scripts/validate.py ]; then \
+		$(PYTHON) scripts/validate.py --input $(EVALS); \
+	else \
+		echo "validate: scripts/validate.py is internal and not part of the public"; \
+		echo "          upload set; skipping package validation."; \
+	fi
+
+hashes:
+	@sha256sum $(EVALS) $(DATA)/benchmarks.csv $(DATA)/benchmarks.jsonl $(DATA)/summary.json README.md LICENSE
+
+help:
+	@sed -n '1,13p' Makefile

README.md

@@ -0,0 +1,427 @@
+---
+license: cc-by-4.0
+pretty_name: NativePort Web-Access API Benchmarks
+language:
+  - en
+tags:
+  - benchmark
+  - web-search
+  - web-scraping
+  - browser-automation
+  - agents
+  - tool-use
+  - api-evaluation
+  - latency
+  - cost
+size_categories:
+  - n<1K
+configs:
+  - config_name: metric_rows
+    data_files:
+      - split: train
+        path: data/benchmarks.csv
+  - config_name: evaluations
+    data_files:
+      - split: train
+        path: data/benchmarks.jsonl
+---
+
+# NativePort Web-Access API Benchmarks
+
+Measured quality, latency, cost and error-rate figures for **22 commercial web-access
+APIs** — search, SERP, scraping, crawling, extraction, sourced answers, screenshots,
+document parsing, browser actions and change watching — scored per capability on a
+fixed task corpus. This is the `2026-08-05` run: **67 provider × capability
+scorecards** across **13 capabilities**, flattened into **297 metric rows**.
+
+It exists for one practical decision: *when an AI agent needs to reach the live web,
+which API should the call go to, and what will that cost in latency and dollars?*
+Composite scores alone rarely answer that. The raw per-metric values do, and they are
+all here.
+
+## Disclosure
+
+This dataset is produced by **NativePort** from its **own first-party benchmark
+runs**. NativePort operates a commercial gateway that routes to many of the providers
+scored here. It is **not an independent third-party evaluation**, and it should not be
+cited as one. The measurement protocol, including what NativePort explicitly declines
+to claim, is published in full (linked under [Methodology](#methodology)) so readers
+can weigh the numbers accordingly. Weak and last-place results are published
+unchanged — three scorecards in this snapshot sit below 2.0 out of 10, and two record a
+100% error rate.
+
+## Dataset structure
+
+Two views of the same 67 scorecards. Nothing is aggregated or re-scored between them.
+
+| Config | File | Rows | Grain |
+| --- | --- | --- | --- |
+| `metric_rows` | `data/benchmarks.csv` | 297 | one row per provider × capability × metric (tidy/long) |
+| `evaluations` | `data/benchmarks.jsonl` | 67 | one record per provider × capability, metrics nested |
+
+`data/summary.json` carries the computed snapshot facts (counts, date range, source
+hash, per-capability inventories). Every figure in it is derived by the generator.
+
+### `data/benchmarks.csv` — tidy metric rows
+
+| Column | Type | Description |
+| --- | --- | --- |
+| `evaluation_id` | string | `{provider_id}:{capability_id}`, unique per scorecard |
+| `provider_id` | string | Provider slug as published by the source |
+| `provider_name` | string | Display name, e.g. `Firecrawl` |
+| `provider_group` | string | `Search`, `Scraping & Crawling` or `Browser Automation` |
+| `provider_category` | string | Source's short category, e.g. `Google SERP scrape` |
+| `capability_id` | string | Capability slug, e.g. `search`, `scrape`, `extract_ai` |
+| `capability_label` | string | Human label, e.g. `Extract · AI/schema` |
+| `metric_key` | string | Metric identifier, e.g. `recall_at_10` |
+| `metric_label` | string | Source's short label for the metric |
+| `metric_raw` | number | **The measured value**, copied verbatim from the source |
+| `metric_display` | string | Source's formatted rendering, e.g. `816 ms`, `$0.0003 / call` |
+| `metric_index` | int | Position of this metric within its scorecard (source order) |
+| `composite_score` | number | Scorecard composite for this provider × capability |
+| `composite_scale_max` | int | `10` for every row |
+| `rank` | int | Rank within the capability, 1 = best |
+| `rank_of` | int | Number of providers ranked in that capability |
+| `is_capability_top` | bool | `true` where `rank == 1` |
+| `measured_date` | date | Date of the run that produced this scorecard |
+| `note` | string | Source's scorecard note; empty string where absent |
+| `provider_page_url` | url | Public scorecard page for this provider |
+| `run_id` | string | Benchmark run identifier (`2026-08-05`) |
+| `source_url` | url | `https://nativeport.ai/evals.json` |
+| `snapshot_sha256` | string | SHA-256 of the exact source snapshot these rows came from |
+
+Provenance columns repeat on every row so that a filtered slice stays interpretable
+and verifiable on its own.
+
+### `data/benchmarks.jsonl` — one record per evaluation
+
+```json
+{
+  "evaluation_id": "serper:serp",
+  "provider_id": "serper",
+  "provider_name": "Serper",
+  "provider_group": "Search",
+  "provider_category": "Google SERP scrape",
+  "capability_id": "serp",
+  "capability_label": "SERP verticals",
+  "capability_description": "Google SERP verticals (web, news, images, places, scholar) as structured JSON",
+  "composite_score": 9.13,
+  "composite_scale_max": 10,
+  "rank": 1,
+  "rank_of": 4,
+  "is_capability_top": true,
+  "measured_date": "2026-08-05",
+  "note": "Lowest cost and latency; Google only.",
+  "metric_count": 4,
+  "metrics": [
+    {"metric_index": 0, "metric_key": "quality", "metric_label": "Quality", "raw_value": 0.938, "display_value": "0.94"},
+    {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 886.0, "display_value": "886 ms"},
+    {"metric_index": 2, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.0003, "display_value": "$0.0003 / call"},
+    {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}
+  ],
+  "provider_page_url": "https://nativeport.ai/providers/serper/",
+  "run_id": "2026-08-05",
+  "source_url": "https://nativeport.ai/evals.json",
+  "source_schema_version": 1,
+  "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"
+}
+```
+
+`metrics` is a **list of structs**, not a key-value map, because metric sets differ by
+capability — a map would force a sparse union column across the whole dataset.
+
+### Capabilities in this snapshot
+
+Each capability is a separate leaderboard with its own task corpus, its own quality
+metric and its own cost denominator.
+
+| `capability_id` | Label | Evaluations | Metric rows | Metric keys |
+| --- | --- | --- | --- | --- |
+| `act` | Act · declarative | 6 | 24 | `cost_per_call_usd`, `error_rate_pct`, `latency_p50_ms`, `task_success` |
+| `act_agent` | Act · NL-agent | 1 | 4 | `cost_per_call_usd`, `error_rate_pct`, `latency_p50_ms`, `task_success` |
+| `answer` | Answer | 6 | 30 | `answer_correctness`, `citation_faithfulness`, `cost_per_call_usd`, `error_rate_pct`, `latency_p50_ms` |
+| `crawl` | Crawl | 4 | 16 | `coverage`, `cost_per_useful_usd`, `error_rate_pct`, `latency_p50_ms` |
+| `extract_ai` | Extract · AI/schema | 4 | 16 | `field_accuracy`, `cost_per_useful_usd`, `error_rate_pct`, `latency_p50_ms` |
+| `extract_rules` | Extract · CSS rules | 3 | 12 | `field_accuracy`, `cost_per_useful_usd`, `error_rate_pct`, `latency_p50_ms` |
+| `parse` | Parse · PDF/doc | 3 | 12 | `text_accuracy`, `cost_per_useful_usd`, `error_rate_pct`, `latency_p50_ms` |
+| `scrape` | Scrape | 10 | 50 | `block_bypass_success_rate`, `markdown_cleanliness`, `cost_per_successful_page_usd`, `error_rate_pct`, `latency_p50_ms` |
+| `scrape_domain` | Scrape-domain | 6 | 30 | `value_accuracy`, `field_fill`, `cost_per_useful_usd`, `error_rate_pct`, `latency_p50_ms` |
+| `screenshot` | Screenshot | 8 | 38 | `valid_image_rate`, `full_page_support`, `cost_per_call_usd`, `error_rate_pct`, `latency_p50_ms` |
+| `search` | Search | 11 | 44 | `recall_at_10`, `cost_per_useful_usd`, `error_rate_pct`, `latency_p50_ms` |
+| `serp` | SERP verticals | 4 | 16 | `quality`, `cost_per_call_usd`, `error_rate_pct`, `latency_p50_ms` |
+| `watch` | Watch | 1 | 5 | `classification_accuracy`, `diff_quality`, `cost_per_call_usd`, `error_rate_pct`, `latency_p50_ms` |
+
+Metric sets are stable within a capability but **not guaranteed uniform**: one
+`screenshot` scorecard carries three metrics rather than five — its source note records
+that every capture errored, and no cost-per-call or full-page value is published for
+it. Consumers should key on `metric_key` rather than on metric position or count, and
+treat a missing metric as absent rather than as zero.
+
+### Metric keys
+
+| `metric_key` | Label | Occurrences | Direction | Unit |
+| --- | --- | --- | --- | --- |
+| `answer_correctness` | Correctness | 6 | higher is better | 0–1 |
+| `block_bypass_success_rate` | Anti-bot bypass | 10 | higher is better | percent |
+| `citation_faithfulness` | Citation faithfulness | 6 | higher is better | 0–1 |
+| `classification_accuracy` | Class. accuracy | 1 | higher is better | 0–1 |
+| `coverage` | Coverage | 4 | higher is better | 0–1 |
+| `cost_per_call_usd` | Cost | 25 | lower is better | USD per call |
+| `cost_per_successful_page_usd` | Cost | 10 | lower is better | USD per successful page |
+| `cost_per_useful_usd` | Cost | 31 | lower is better | USD per useful result |
+| `diff_quality` | Diff quality | 1 | higher is better | 0–1 |
+| `error_rate_pct` | Errors | 67 | lower is better | percent |
+| `field_accuracy` | Field accuracy | 7 | higher is better | 0–1 |
+| `field_fill` | Field fill | 6 | higher is better | 0–1 |
+| `full_page_support` | Full-page | 7 | higher is better | 0–1 |
+| `latency_p50_ms` | Latency p50 | 67 | lower is better | milliseconds |
+| `markdown_cleanliness` | Markdown clean | 10 | higher is better | 0–10 |
+| `quality` | Quality | 4 | higher is better | 0–1 |
+| `recall_at_10` | Recall@10 | 11 | higher is better | 0–1 |
+| `task_success` | Task success | 7 | higher is better | 0–1 |
+| `text_accuracy` | Text accuracy | 3 | higher is better | 0–1 |
+| `valid_image_rate` | Valid image | 8 | higher is better | 0–1 |
+| `value_accuracy` | Accuracy | 6 | higher is better | 0–1 |
+
+Units and direction are read off the source's own `metric_display` strings (`816 ms`,
+`$0.0003 / call`, `0%`, `3.4 / 10`); the machine-readable value always lives in
+`metric_raw` / `raw_value`.
+
+## Usage
+
+```python
+from datasets import load_dataset
+
+rows = load_dataset("nativeport/web-access-api-benchmarks", "metric_rows", split="train")
+evals = load_dataset("nativeport/web-access-api-benchmarks", "evaluations", split="train")
+```
+
+Cheapest search provider that clears a recall floor — the typical routing question,
+answered from the tidy view with pandas:
+
+```python
+import pandas as pd
+
+df = pd.read_csv("hf://datasets/nativeport/web-access-api-benchmarks/data/benchmarks.csv")
+search = df[df.capability_id == "search"]
+
+wide = search.pivot_table(
+    index=["provider_id", "composite_score", "rank"],
+    columns="metric_key",
+    values="metric_raw",
+).reset_index()
+
+eligible = wide[(wide.recall_at_10 >= 0.55) & (wide.error_rate_pct == 0)]
+print(eligible.sort_values("cost_per_useful_usd")[
+    ["provider_id", "recall_at_10", "latency_p50_ms", "cost_per_useful_usd", "rank"]
+])
+```
+
+The same query with no third-party dependencies, from the nested view:
+
+```python
+import json
+
+with open("data/benchmarks.jsonl", encoding="utf-8") as handle:
+    evaluations = [json.loads(line) for line in handle]
+
+def metric(evaluation, key):
+    for entry in evaluation["metrics"]:
+        if entry["metric_key"] == key:
+            return entry["raw_value"]
+    return None
+
+search = [e for e in evaluations if e["capability_id"] == "search"]
+eligible = [e for e in search if metric(e, "recall_at_10") >= 0.55]
+for evaluation in sorted(eligible, key=lambda e: metric(e, "cost_per_useful_usd")):
+    print(
+        evaluation["provider_id"],
+        metric(evaluation, "recall_at_10"),
+        f'{metric(evaluation, "latency_p50_ms"):.0f} ms',
+        f'${metric(evaluation, "cost_per_useful_usd"):.5f}/useful',
+    )
+```
+
+Two shapes worth knowing before you write a query:
+
+- **Latency and cost are not comparable across capabilities.** A `serp` call and an
+  `act_agent` call differ by three orders of magnitude in wall time by nature.
+- **Cost denominators differ by capability.** `cost_per_call_usd` charges every
+  attempt; `cost_per_useful_usd` and `cost_per_successful_page_usd` divide by usable
+  output, so failures inflate them. Do not mix the three in one ordering.
+
+## Methodology
+
+Each capability has a versioned task corpus held fixed across every provider — the
+same URLs, queries, target schemas and pass criteria — and four dimensions are
+recorded per provider × capability pair: a capability-specific quality metric, median
+wall-clock latency measured from the runner, track-specific cost computed from the
+provider's real metered price, and error rate across the run. Where quality needs
+judgment rather than string comparison, grading is done by an LLM panel working from
+written rubrics. These fold into a composite out of 10, and providers are ranked
+within each capability.
+
+The full protocol — corpus construction, the four measured dimensions, how grading
+works, and what NativePort explicitly does **not** claim (no uptime, SLA or
+throughput figures) — is documented in
+[How we measure](https://nativeport.ai/methodology/?utm_source=huggingface&utm_medium=referral&utm_campaign=backlink_hf_web_access_benchmarks_20260811).
+Human-readable ranked tables per capability are at the
+[leaderboards](https://nativeport.ai/leaderboards/).
+
+The source snapshot summarises its own protocol as:
+
+> One fixed task corpus per capability, identical for every provider; scorecards carry
+> their run dates. Weak scores stay published, and the gateway's flat top-up fee means
+> the ranking earns nothing from steering you toward pricier providers.
+
+## Provenance and reproducibility
+
+| Field | Value |
+| --- | --- |
+| Source | `https://nativeport.ai/evals.json` |
+| Source schema version | `1` |
+| Source SHA-256 | `f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b` |
+| Source size | 89894 bytes |
+| Run | `2026-08-05` |
+| Measured date range | `2026-08-05` to `2026-08-05` |
+
+Rebuild the data files from that snapshot:
+
+```bash
+curl -sSfL https://nativeport.ai/evals.json -o evals.json
+sha256sum evals.json          # must match the SHA-256 above
+python3 scripts/build_dataset.py --input evals.json --output-dir data
+```
+
+`scripts/build_dataset.py` is standard-library-only and deterministic: identical input
+bytes produce byte-identical outputs. No wall-clock timestamp is written anywhere, so
+a rebuild can be diffed directly against the published files. Numeric values are
+carried across as exact source tokens — the generator verifies that each emitted
+number serialises character-for-character back to the token it read, and aborts rather
+than emit a rounded stand-in. `data/summary.json` records the SHA-256 of both data
+files.
+
+Fields present in the source but **excluded by design**: gateway routing and
+authentication strings, list prices, latency prose, marketing summaries,
+`choose_if` / `avoid_if` guidance, and catalog tier labels. Only benchmark
+measurements and the provenance needed to interpret them are published here.
+
+## Limitations and scope
+
+- **Coverage is partial.** 22 of the 32 providers in the source catalog carry
+  scorecards in this run; the other 10 have no eval entries and therefore no rows
+  here. The source additionally flags 5 provider × capability pairs as unscored in
+  this run. Absence from this dataset means *not measured in the `2026-08-05` run* —
+  not "failed", and not "unavailable".
+- **Composites are capability-local.** A composite is only meaningful against other
+  providers in the same capability. A `9.13` on `serp` and a `9.13` on `scrape` are
+  not the same achievement, and averaging a provider's composites across capabilities
+  produces a number with no defined meaning.
+- **Thin capabilities.** `act_agent` and `watch` contain a single scored provider
+  each; `extract_rules` and `parse` contain three. A rank of 1 out of 1 is not
+  evidence of superiority. Always read `rank_of` alongside `rank`.
+- **Single point in time.** Every row in this snapshot was measured on `2026-08-05`.
+  Provider behaviour, pricing and anti-bot posture change; these figures age.
+- **First-party measurement.** Runs are operated by NativePort, which has a commercial
+  relationship with providers in the catalog. See [Disclosure](#disclosure).
+- **Judged metrics carry model bias.** Quality metrics that require judgment are
+  graded by an LLM panel against rubrics, not by human annotators.
+- **Not measured at all:** uptime, SLA conformance, throughput ceilings, regional
+  performance, concurrency behaviour, and long-run stability. No row in this dataset
+  speaks to any of them.
+- **Cost is a measurement, not a quote.** Figures are computed from metered prices at
+  run time for the calls in the corpus. They are not an offer, a rate card, or a
+  prediction of any particular workload's bill.
+
+## Update policy
+
+- The dataset tracks NativePort benchmark runs. A new run publishes as a new revision
+  of this repository, with `run_id`, `measured_date` and `snapshot_sha256` changing
+  together.
+- Prior revisions stay reachable through the repository's commit history; superseded
+  numbers are not silently rewritten in place.
+- Schema changes that are not backward compatible will be described in the commit that
+  makes them and reflected in the tables above.
+- No update cadence is promised here. `measured_date` and `run_id` are on every row
+  precisely so a consumer can decide for itself whether the snapshot is still fresh
+  enough to act on.
+
+## Licensing
+
+This dataset is licensed by NativePort under the **Creative Commons Attribution 4.0
+International licence (CC BY 4.0)**.
+
+| Field | Value |
+| --- | --- |
+| Licence | Creative Commons Attribution 4.0 International (CC BY 4.0) |
+| Canonical licence URL | <https://creativecommons.org/licenses/by/4.0/> |
+| Full legal code | <https://creativecommons.org/licenses/by/4.0/legalcode> — reproduced verbatim in [`LICENSE`](LICENSE) |
+| SPDX identifier | `CC-BY-4.0` (Hub metadata key: `license: cc-by-4.0`) |
+
+**What the licence covers.** NativePort licenses what it is in a position to license:
+this dataset as a compilation — its selection, arrangement, schema, documentation and
+this card — together with the benchmark measurements NativePort itself produced and
+any database rights NativePort holds in them. The grant extends only to those rights
+and only to the extent NativePort holds them. Where a jurisdiction treats an individual
+measured figure as an unprotectable fact, the licence simply does not reach it: CC BY
+4.0 places no conditions on a use that is lawful without permission (legal code
+§ 2(a)(2) and § 8(a)). Where sui generis database rights do apply, § 4 of the legal
+code grants extraction and reuse of all or a substantial part of the contents, subject
+to the same attribution condition.
+
+**What the licence does not cover.** CC BY 4.0 does not license patent or trademark
+rights (legal code § 2(b)(2)). The provider, product and company names and marks that
+appear in this dataset — including every mark listed under [Trademark
+notice](#trademark-notice) — remain the property of their respective owners. They are
+**not** licensed, sublicensed or otherwise granted to you here, by NativePort or by
+this licence; NativePort has no authority to grant rights in another party's marks and
+does not purport to. Reusing this dataset under CC BY 4.0 therefore gives you no right
+to use those marks beyond whatever nominative, descriptive or fair use your own
+jurisdiction independently allows. Nor does the licence permit you to assert or imply a
+connection with, sponsorship by, or endorsement from NativePort as licensor
+(legal code § 2(a)(6)); no trademark owner named here has endorsed, reviewed or
+sponsored these results.
+
+**Attribution.** Credit *NativePort*, name the dataset and the run (`2026-08-05`),
+link to this repository or to `https://nativeport.ai/evals.json`, state that the
+material is under CC BY 4.0 with a link to the licence, and indicate whether you
+modified it. The [Citation](#citation) block below carries everything needed.
+
+**No warranty.** The material is offered as-is and as-available, without warranties or
+conditions of any kind, and NativePort's liability is limited, as set out in § 5 of the
+legal code. Read it alongside [Limitations and scope](#limitations-and-scope): these
+are measurements from one run on one date, not a guarantee of any provider's future
+behaviour.
+
+## Trademark notice
+
+Provider names, product names and logos referenced here — including Serper, SerpApi,
+SearchAPI.io, Brave Search, You.com, DataForSEO, Tavily, Exa, Linkup, Parallel, Jina,
+ScraperAPI, Firecrawl, ScrapingBee, Scrapfly, ZenRows, Crawlbase, Oxylabs, Bright
+Data, Spider, Zyte and Apify — are trademarks of their respective owners. They are
+used here for identification and factual comparison only. Their appearance does not
+imply any affiliation with, sponsorship by, endorsement by, or review of these results
+by the trademark owners. NativePort is a trademark of its owner. Hugging Face is a
+trademark of Hugging Face, Inc.
+
+The CC BY 4.0 licence described under [Licensing](#licensing) grants **no** rights in
+any of these marks — trademark rights are outside what that licence conveys (legal code
+§ 2(b)(2)) and outside what NativePort could convey in the first place.
+
+## Citation
+
+```bibtex
+@misc{nativeport_web_access_api_benchmarks_2026_08_05,
+  title        = {NativePort Web-Access API Benchmarks},
+  author       = {{NativePort}},
+  year         = {2026},
+  note         = {Benchmark run 2026-08-05; 67 provider-capability scorecards across 13 capabilities.
+                  Source snapshot SHA-256 f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b},
+  license      = {CC BY 4.0, \url{https://creativecommons.org/licenses/by/4.0/}},
+  howpublished = {\url{https://nativeport.ai/evals.json}}
+}
+```
+
+Plain text: NativePort. *NativePort Web-Access API Benchmarks*, run `2026-08-05`.
+Retrieved from `https://nativeport.ai/evals.json`. Licensed under CC BY 4.0
+(<https://creativecommons.org/licenses/by/4.0/>); indicate if you modified it.

data/benchmarks.jsonl

@@ -0,0 +1,67 @@
+{"evaluation_id": "scrapingbee:act", "provider_id": "scrapingbee", "provider_name": "ScrapingBee", "provider_group": "Scraping & Crawling", "provider_category": "Single-call scraping API", "capability_id": "act", "capability_label": "Act · declarative", "capability_description": "Declarative in-page interactions (click, fill, scroll) in one call", "composite_score": 8.68, "composite_scale_max": 10, "rank": 1, "rank_of": 6, "is_capability_top": true, "measured_date": "2026-08-05", "note": "Declarative browser actions: strongest showing.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "task_success", "metric_label": "Task success", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 5887.9, "display_value": "5.9 s"}, {"metric_index": 2, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.006, "display_value": "$0.006 / call"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/scrapingbee/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "zenrows:act", "provider_id": "zenrows", "provider_name": "ZenRows", "provider_group": "Scraping & Crawling", "provider_category": "Scraper API + CDP browser", "capability_id": "act", "capability_label": "Act · declarative", "capability_description": "Declarative in-page interactions (click, fill, scroll) in one call", "composite_score": 8.68, "composite_scale_max": 10, "rank": 2, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "task_success", "metric_label": "Task success", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 6084.8, "display_value": "6.1 s"}, {"metric_index": 2, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.005, "display_value": "$0.005 / call"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/zenrows/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "zyte:act", "provider_id": "zyte", "provider_name": "Zyte", "provider_group": "Scraping & Crawling", "provider_category": "Single-endpoint extraction", "capability_id": "act", "capability_label": "Act · declarative", "capability_description": "Declarative in-page interactions (click, fill, scroll) in one call", "composite_score": 8.5, "composite_scale_max": 10, "rank": 3, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "task_success", "metric_label": "Task success", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 39968.1, "display_value": "40.0 s"}, {"metric_index": 2, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.001, "display_value": "$0.001 / call"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/zyte/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "firecrawl:act", "provider_id": "firecrawl", "provider_name": "Firecrawl", "provider_group": "Scraping & Crawling", "provider_category": "Scrape + crawl + extract", "capability_id": "act", "capability_label": "Act · declarative", "capability_description": "Declarative in-page interactions (click, fill, scroll) in one call", "composite_score": 7.46, "composite_scale_max": 10, "rank": 4, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "task_success", "metric_label": "Task success", "raw_value": 0.8, "display_value": "0.80"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 4336.7, "display_value": "4.3 s"}, {"metric_index": 2, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.005, "display_value": "$0.005 / call"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/firecrawl/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "scrapfly:act", "provider_id": "scrapfly", "provider_name": "Scrapfly", "provider_group": "Scraping & Crawling", "provider_category": "Single-call scraping API", "capability_id": "act", "capability_label": "Act · declarative", "capability_description": "Declarative in-page interactions (click, fill, scroll) in one call", "composite_score": 5.74, "composite_scale_max": 10, "rank": 5, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "task_success", "metric_label": "Task success", "raw_value": 0.6, "display_value": "0.60"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 8698.2, "display_value": "8.7 s"}, {"metric_index": 2, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.002, "display_value": "$0.002 / call"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/scrapfly/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "spider:act", "provider_id": "spider", "provider_name": "Spider", "provider_group": "Scraping & Crawling", "provider_category": "Crawl / AI-focused platform", "capability_id": "act", "capability_label": "Act · declarative", "capability_description": "Declarative in-page interactions (click, fill, scroll) in one call", "composite_score": 1.5, "composite_scale_max": 10, "rank": 6, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Live declarative clicks do nothing.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "task_success", "metric_label": "Task success", "raw_value": 0.0, "display_value": "0.00"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 105130.2, "display_value": "105.1 s"}, {"metric_index": 2, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.001, "display_value": "$0.001 / call"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/spider/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "firecrawl:act_agent", "provider_id": "firecrawl", "provider_name": "Firecrawl", "provider_group": "Scraping & Crawling", "provider_category": "Scrape + crawl + extract", "capability_id": "act_agent", "capability_label": "Act · NL-agent", "capability_description": "A natural-language goal handed to an autonomous browser agent", "composite_score": 7.99, "composite_scale_max": 10, "rank": 1, "rank_of": 1, "is_capability_top": true, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "task_success", "metric_label": "Task success", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 43196.7, "display_value": "43.2 s"}, {"metric_index": 2, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.05, "display_value": "$0.05 / call"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/firecrawl/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "parallel:answer", "provider_id": "parallel", "provider_name": "Parallel", "provider_group": "Search", "provider_category": "Agent-optimized search index", "capability_id": "answer", "capability_label": "Answer", "capability_description": "A question in, a synthesized answer with cited sources out", "composite_score": 8.33, "composite_scale_max": 10, "rank": 1, "rank_of": 6, "is_capability_top": true, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "answer_correctness", "metric_label": "Correctness", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "citation_faithfulness", "metric_label": "Citation faithfulness", "raw_value": 0.735, "display_value": "0.73"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 16325.6, "display_value": "16.3 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.01, "display_value": "$0.01 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/parallel/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "exa:answer", "provider_id": "exa", "provider_name": "Exa", "provider_group": "Search", "provider_category": "Neural + keyword index", "capability_id": "answer", "capability_label": "Answer", "capability_description": "A question in, a synthesized answer with cited sources out", "composite_score": 7.99, "composite_scale_max": 10, "rank": 2, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Second on citation faithfulness, at a fraction of the leader's latency.", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "answer_correctness", "metric_label": "Correctness", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "citation_faithfulness", "metric_label": "Citation faithfulness", "raw_value": 0.515, "display_value": "0.52"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 1357.5, "display_value": "1.4 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.005, "display_value": "$0.005 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/exa/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "tavily:answer", "provider_id": "tavily", "provider_name": "Tavily", "provider_group": "Search", "provider_category": "Agent-native search index", "capability_id": "answer", "capability_label": "Answer", "capability_description": "A question in, a synthesized answer with cited sources out", "composite_score": 7.28, "composite_scale_max": 10, "rank": 3, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "answer_correctness", "metric_label": "Correctness", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "citation_faithfulness", "metric_label": "Citation faithfulness", "raw_value": 0.297, "display_value": "0.30"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 1924.5, "display_value": "1.9 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.008, "display_value": "$0.008 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/tavily/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "youcom:answer", "provider_id": "youcom", "provider_name": "You.com", "provider_group": "Search", "provider_category": "Own crawl index + Research API", "capability_id": "answer", "capability_label": "Answer", "capability_description": "A question in, a synthesized answer with cited sources out", "composite_score": 7.06, "composite_scale_max": 10, "rank": 4, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "answer_correctness", "metric_label": "Correctness", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "citation_faithfulness", "metric_label": "Citation faithfulness", "raw_value": 0.261, "display_value": "0.26"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 3476.8, "display_value": "3.5 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.02, "display_value": "$0.02 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/youcom/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "linkup:answer", "provider_id": "linkup", "provider_name": "Linkup", "provider_group": "Search", "provider_category": "Own crawl index", "capability_id": "answer", "capability_label": "Answer", "capability_description": "A question in, a synthesized answer with cited sources out", "composite_score": 6.84, "composite_scale_max": 10, "rank": 5, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "answer_correctness", "metric_label": "Correctness", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "citation_faithfulness", "metric_label": "Citation faithfulness", "raw_value": 0.138, "display_value": "0.14"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 2345.7, "display_value": "2.3 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.005, "display_value": "$0.005 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/linkup/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "jina:answer", "provider_id": "jina", "provider_name": "Jina", "provider_group": "Search", "provider_category": "Search + read-to-markdown", "capability_id": "answer", "capability_label": "Answer", "capability_description": "A question in, a synthesized answer with cited sources out", "composite_score": 6.65, "composite_scale_max": 10, "rank": 6, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "answer_correctness", "metric_label": "Correctness", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "citation_faithfulness", "metric_label": "Citation faithfulness", "raw_value": 0.125, "display_value": "0.12"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 3625.1, "display_value": "3.6 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.02, "display_value": "$0.02 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/jina/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "tavily:crawl", "provider_id": "tavily", "provider_name": "Tavily", "provider_group": "Search", "provider_category": "Agent-native search index", "capability_id": "crawl", "capability_label": "Crawl", "capability_description": "Enumerate and fetch many pages across a site", "composite_score": 8.91, "composite_scale_max": 10, "rank": 1, "rank_of": 4, "is_capability_top": true, "measured_date": "2026-08-05", "note": "Performs well through its /map variant.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "coverage", "metric_label": "Coverage", "raw_value": 0.88, "display_value": "0.88"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 557.5, "display_value": "558 ms"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.0125, "display_value": "$0.013 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/tavily/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "firecrawl:crawl", "provider_id": "firecrawl", "provider_name": "Firecrawl", "provider_group": "Scraping & Crawling", "provider_category": "Scrape + crawl + extract", "capability_id": "crawl", "capability_label": "Crawl", "capability_description": "Enumerate and fetch many pages across a site", "composite_score": 7.95, "composite_scale_max": 10, "rank": 2, "rank_of": 4, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Clean path-scoped enumeration; Tavily's map now covers more of the corpus.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "coverage", "metric_label": "Coverage", "raw_value": 0.8, "display_value": "0.80"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 29740.2, "display_value": "29.7 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.02675, "display_value": "$0.027 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/firecrawl/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "spider:crawl", "provider_id": "spider", "provider_name": "Spider", "provider_group": "Scraping & Crawling", "provider_category": "Crawl / AI-focused platform", "capability_id": "crawl", "capability_label": "Crawl", "capability_description": "Enumerate and fetch many pages across a site", "composite_score": 7.28, "composite_scale_max": 10, "rank": 3, "rank_of": 4, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Quick at recursive crawling.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "coverage", "metric_label": "Coverage", "raw_value": 0.537, "display_value": "0.54"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 2308.4, "display_value": "2.3 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 9.1e-05, "display_value": "$9.1e-05 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/spider/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "brightdata:crawl", "provider_id": "brightdata", "provider_name": "Bright Data", "provider_group": "Scraping & Crawling", "provider_category": "Web Unlocker + Scraper API", "capability_id": "crawl", "capability_label": "Crawl", "capability_description": "Enumerate and fetch many pages across a site", "composite_score": 3.88, "composite_scale_max": 10, "rank": 4, "rank_of": 4, "is_capability_top": false, "measured_date": "2026-08-05", "note": "No link discovery: url_collection hands back just the seed.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "coverage", "metric_label": "Coverage", "raw_value": 0.1, "display_value": "0.10"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 6469.8, "display_value": "6.5 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.01, "display_value": "$0.01 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 20.0, "display_value": "20%"}], "provider_page_url": "https://nativeport.ai/providers/brightdata/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "scrapfly:extract_ai", "provider_id": "scrapfly", "provider_name": "Scrapfly", "provider_group": "Scraping & Crawling", "provider_category": "Single-call scraping API", "capability_id": "extract_ai", "capability_label": "Extract · AI/schema", "capability_description": "Structured fields pulled from a page against an AI/natural-language schema", "composite_score": 9.0, "composite_scale_max": 10, "rank": 1, "rank_of": 4, "is_capability_top": true, "measured_date": "2026-08-05", "note": "Extraction accuracy ranks first among the set.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "field_accuracy", "metric_label": "Field accuracy", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 4057.2, "display_value": "4.1 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.002, "display_value": "$0.002 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/scrapfly/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "firecrawl:extract_ai", "provider_id": "firecrawl", "provider_name": "Firecrawl", "provider_group": "Scraping & Crawling", "provider_category": "Scrape + crawl + extract", "capability_id": "extract_ai", "capability_label": "Extract · AI/schema", "capability_description": "Structured fields pulled from a page against an AI/natural-language schema", "composite_score": 8.75, "composite_scale_max": 10, "rank": 2, "rank_of": 4, "is_capability_top": false, "measured_date": "2026-08-05", "note": "On the precision test it hallucinated a null field.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "field_accuracy", "metric_label": "Field accuracy", "raw_value": 0.957, "display_value": "0.96"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 2356.8, "display_value": "2.4 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.005225, "display_value": "$0.0052 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/firecrawl/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "scrapingbee:extract_ai", "provider_id": "scrapingbee", "provider_name": "ScrapingBee", "provider_group": "Scraping & Crawling", "provider_category": "Single-call scraping API", "capability_id": "extract_ai", "capability_label": "Extract · AI/schema", "capability_description": "Structured fields pulled from a page against an AI/natural-language schema", "composite_score": 8.21, "composite_scale_max": 10, "rank": 3, "rank_of": 4, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "field_accuracy", "metric_label": "Field accuracy", "raw_value": 0.893, "display_value": "0.89"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 3112.7, "display_value": "3.1 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.006719, "display_value": "$0.0067 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/scrapingbee/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "zyte:extract_ai", "provider_id": "zyte", "provider_name": "Zyte", "provider_group": "Scraping & Crawling", "provider_category": "Single-endpoint extraction", "capability_id": "extract_ai", "capability_label": "Extract · AI/schema", "capability_description": "Structured fields pulled from a page against an AI/natural-language schema", "composite_score": 5.17, "composite_scale_max": 10, "rank": 4, "rank_of": 4, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Custom fields aren't possible under its fixed schema.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "field_accuracy", "metric_label": "Field accuracy", "raw_value": 0.461, "display_value": "0.46"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 5552.9, "display_value": "5.6 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.002169, "display_value": "$0.0022 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 7.1, "display_value": "7%"}], "provider_page_url": "https://nativeport.ai/providers/zyte/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "scrapingbee:extract_rules", "provider_id": "scrapingbee", "provider_name": "ScrapingBee", "provider_group": "Scraping & Crawling", "provider_category": "Single-call scraping API", "capability_id": "extract_rules", "capability_label": "Extract · CSS rules", "capability_description": "Structured fields pulled from a page via CSS/XPath rules", "composite_score": 9.23, "composite_scale_max": 10, "rank": 1, "rank_of": 3, "is_capability_top": true, "measured_date": "2026-08-05", "note": "Top-ranked rule-based extractor, at a fraction of AI-extract cost.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "field_accuracy", "metric_label": "Field accuracy", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 2698.8, "display_value": "2.7 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.00098, "display_value": "$0.00098 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/scrapingbee/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "oxylabs:extract_rules", "provider_id": "oxylabs", "provider_name": "Oxylabs", "provider_group": "Scraping & Crawling", "provider_category": "Scraping API + proxy network", "capability_id": "extract_rules", "capability_label": "Extract · CSS rules", "capability_description": "Structured fields pulled from a page via CSS/XPath rules", "composite_score": 8.65, "composite_scale_max": 10, "rank": 2, "rank_of": 3, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "field_accuracy", "metric_label": "Field accuracy", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 8217.8, "display_value": "8.2 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.0012, "display_value": "$0.0012 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/oxylabs/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "zenrows:extract_rules", "provider_id": "zenrows", "provider_name": "ZenRows", "provider_group": "Scraping & Crawling", "provider_category": "Scraper API + CDP browser", "capability_id": "extract_rules", "capability_label": "Extract · CSS rules", "capability_description": "Structured fields pulled from a page via CSS/XPath rules", "composite_score": 5.77, "composite_scale_max": 10, "rank": 3, "rank_of": 3, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Every *.toscrape.com domain is blocklisted — an openly stated gap.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "field_accuracy", "metric_label": "Field accuracy", "raw_value": 0.5, "display_value": "0.50"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 1841.0, "display_value": "1.8 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.0028, "display_value": "$0.0028 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 50.0, "display_value": "50%"}], "provider_page_url": "https://nativeport.ai/providers/zenrows/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "firecrawl:parse", "provider_id": "firecrawl", "provider_name": "Firecrawl", "provider_group": "Scraping & Crawling", "provider_category": "Scrape + crawl + extract", "capability_id": "parse", "capability_label": "Parse · PDF/doc", "capability_description": "PDF/document to text, with OCR where the provider supports it", "composite_score": 8.81, "composite_scale_max": 10, "rank": 1, "rank_of": 3, "is_capability_top": true, "measured_date": "2026-08-05", "note": "Scanned-PDF OCR sets it apart.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "text_accuracy", "metric_label": "Text accuracy", "raw_value": 0.907, "display_value": "0.91"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 354.8, "display_value": "355 ms"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.001103, "display_value": "$0.0011 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/firecrawl/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "jina:parse", "provider_id": "jina", "provider_name": "Jina", "provider_group": "Search", "provider_category": "Search + read-to-markdown", "capability_id": "parse", "capability_label": "Parse · PDF/doc", "capability_description": "PDF/document to text, with OCR where the provider supports it", "composite_score": 8.28, "composite_scale_max": 10, "rank": 2, "rank_of": 3, "is_capability_top": false, "measured_date": "2026-08-05", "note": "No OCR: scanned PDFs come back empty.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "text_accuracy", "metric_label": "Text accuracy", "raw_value": 0.78, "display_value": "0.78"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 1819.0, "display_value": "1.8 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.0, "display_value": "$0 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/jina/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "spider:parse", "provider_id": "spider", "provider_name": "Spider", "provider_group": "Scraping & Crawling", "provider_category": "Crawl / AI-focused platform", "capability_id": "parse", "capability_label": "Parse · PDF/doc", "capability_description": "PDF/document to text, with OCR where the provider supports it", "composite_score": 6.96, "composite_scale_max": 10, "rank": 3, "rank_of": 3, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Scanned PDFs yield raw bytes.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "text_accuracy", "metric_label": "Text accuracy", "raw_value": 0.6, "display_value": "0.60"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 2329.7, "display_value": "2.3 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 3.8e-05, "display_value": "$3.8e-05 / useful"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/spider/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "firecrawl:scrape", "provider_id": "firecrawl", "provider_name": "Firecrawl", "provider_group": "Scraping & Crawling", "provider_category": "Scrape + crawl + extract", "capability_id": "scrape", "capability_label": "Scrape", "capability_description": "Fetch one URL as clean content/markdown, anti-bot handled", "composite_score": 9.85, "composite_scale_max": 10, "rank": 1, "rank_of": 10, "is_capability_top": true, "measured_date": "2026-08-05", "note": "Class-leading result; every anti-bot vendor cleared.", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "block_bypass_success_rate", "metric_label": "Anti-bot bypass", "raw_value": 1.0, "display_value": "100%"}, {"metric_index": 1, "metric_key": "markdown_cleanliness", "metric_label": "Markdown clean", "raw_value": 9.62, "display_value": "9.6 / 10"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 369.0, "display_value": "369 ms"}, {"metric_index": 3, "metric_key": "cost_per_successful_page_usd", "metric_label": "Cost", "raw_value": 0.002094, "display_value": "$0.0021 / success"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/firecrawl/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "scrapfly:scrape", "provider_id": "scrapfly", "provider_name": "Scrapfly", "provider_group": "Scraping & Crawling", "provider_category": "Single-call scraping API", "capability_id": "scrape", "capability_label": "Scrape", "capability_description": "Fetch one URL as clean content/markdown, anti-bot handled", "composite_score": 8.9, "composite_scale_max": 10, "rank": 2, "rank_of": 10, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Cleared both anti-bot vendors behind the single asp=true flag; the one miss was an unprotected page.", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "block_bypass_success_rate", "metric_label": "Anti-bot bypass", "raw_value": 0.875, "display_value": "88%"}, {"metric_index": 1, "metric_key": "markdown_cleanliness", "metric_label": "Markdown clean", "raw_value": 9.57, "display_value": "9.6 / 10"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 4223.9, "display_value": "4.2 s"}, {"metric_index": 3, "metric_key": "cost_per_successful_page_usd", "metric_label": "Cost", "raw_value": 0.003363, "display_value": "$0.0034 / success"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/scrapfly/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "zyte:scrape", "provider_id": "zyte", "provider_name": "Zyte", "provider_group": "Scraping & Crawling", "provider_category": "Single-endpoint extraction", "capability_id": "scrape", "capability_label": "Scrape", "capability_description": "Fetch one URL as clean content/markdown, anti-bot handled", "composite_score": 6.92, "composite_scale_max": 10, "rank": 3, "rank_of": 10, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Cleared Cloudflare, but the DataDome target came back blocked.", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "block_bypass_success_rate", "metric_label": "Anti-bot bypass", "raw_value": 0.75, "display_value": "75%"}, {"metric_index": 1, "metric_key": "markdown_cleanliness", "metric_label": "Markdown clean", "raw_value": 4.17, "display_value": "4.2 / 10"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 9400.8, "display_value": "9.4 s"}, {"metric_index": 3, "metric_key": "cost_per_successful_page_usd", "metric_label": "Cost", "raw_value": 0.000917, "display_value": "$0.00092 / success"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 25.0, "display_value": "25%"}], "provider_page_url": "https://nativeport.ai/providers/zyte/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "scrapingbee:scrape", "provider_id": "scrapingbee", "provider_name": "ScrapingBee", "provider_group": "Scraping & Crawling", "provider_category": "Single-call scraping API", "capability_id": "scrape", "capability_label": "Scrape", "capability_description": "Fetch one URL as clean content/markdown, anti-bot handled", "composite_score": 6.75, "composite_scale_max": 10, "rank": 4, "rank_of": 10, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "block_bypass_success_rate", "metric_label": "Anti-bot bypass", "raw_value": 0.75, "display_value": "75%"}, {"metric_index": 1, "metric_key": "markdown_cleanliness", "metric_label": "Markdown clean", "raw_value": 3.67, "display_value": "3.7 / 10"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 11132.5, "display_value": "11.1 s"}, {"metric_index": 3, "metric_key": "cost_per_successful_page_usd", "metric_label": "Cost", "raw_value": 0.005735, "display_value": "$0.0057 / success"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 12.5, "display_value": "12%"}], "provider_page_url": "https://nativeport.ai/providers/scrapingbee/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "brightdata:scrape", "provider_id": "brightdata", "provider_name": "Bright Data", "provider_group": "Scraping & Crawling", "provider_category": "Web Unlocker + Scraper API", "capability_id": "scrape", "capability_label": "Scrape", "capability_description": "Fetch one URL as clean content/markdown, anti-bot handled", "composite_score": 6.42, "composite_scale_max": 10, "rank": 5, "rank_of": 10, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "block_bypass_success_rate", "metric_label": "Anti-bot bypass", "raw_value": 0.5, "display_value": "50%"}, {"metric_index": 1, "metric_key": "markdown_cleanliness", "metric_label": "Markdown clean", "raw_value": 4.62, "display_value": "4.6 / 10"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 6355.0, "display_value": "6.4 s"}, {"metric_index": 3, "metric_key": "cost_per_successful_page_usd", "metric_label": "Cost", "raw_value": 0.0029, "display_value": "$0.0029 / success"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/brightdata/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "crawlbase:scrape", "provider_id": "crawlbase", "provider_name": "Crawlbase", "provider_group": "Scraping & Crawling", "provider_category": "Single-call + managed crawl", "capability_id": "scrape", "capability_label": "Scrape", "capability_description": "Fetch one URL as clean content/markdown, anti-bot handled", "composite_score": 6.04, "composite_scale_max": 10, "rank": 6, "rank_of": 10, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "block_bypass_success_rate", "metric_label": "Anti-bot bypass", "raw_value": 0.625, "display_value": "62%"}, {"metric_index": 1, "metric_key": "markdown_cleanliness", "metric_label": "Markdown clean", "raw_value": 3.4, "display_value": "3.4 / 10"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 6710.1, "display_value": "6.7 s"}, {"metric_index": 3, "metric_key": "cost_per_successful_page_usd", "metric_label": "Cost", "raw_value": 0.0072, "display_value": "$0.0072 / success"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 37.5, "display_value": "38%"}], "provider_page_url": "https://nativeport.ai/providers/crawlbase/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "spider:scrape", "provider_id": "spider", "provider_name": "Spider", "provider_group": "Scraping & Crawling", "provider_category": "Crawl / AI-focused platform", "capability_id": "scrape", "capability_label": "Scrape", "capability_description": "Fetch one URL as clean content/markdown, anti-bot handled", "composite_score": 6.0, "composite_scale_max": 10, "rank": 7, "rank_of": 10, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "block_bypass_success_rate", "metric_label": "Anti-bot bypass", "raw_value": 0.375, "display_value": "38%"}, {"metric_index": 1, "metric_key": "markdown_cleanliness", "metric_label": "Markdown clean", "raw_value": 9.5, "display_value": "9.5 / 10"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 14476.4, "display_value": "14.5 s"}, {"metric_index": 3, "metric_key": "cost_per_successful_page_usd", "metric_label": "Cost", "raw_value": 0.000173, "display_value": "$0.00017 / success"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/spider/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "oxylabs:scrape", "provider_id": "oxylabs", "provider_name": "Oxylabs", "provider_group": "Scraping & Crawling", "provider_category": "Scraping API + proxy network", "capability_id": "scrape", "capability_label": "Scrape", "capability_description": "Fetch one URL as clean content/markdown, anti-bot handled", "composite_score": 5.98, "composite_scale_max": 10, "rank": 8, "rank_of": 10, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "block_bypass_success_rate", "metric_label": "Anti-bot bypass", "raw_value": 0.625, "display_value": "62%"}, {"metric_index": 1, "metric_key": "markdown_cleanliness", "metric_label": "Markdown clean", "raw_value": 3.4, "display_value": "3.4 / 10"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 14490.6, "display_value": "14.5 s"}, {"metric_index": 3, "metric_key": "cost_per_successful_page_usd", "metric_label": "Cost", "raw_value": 0.00152, "display_value": "$0.0015 / success"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 12.5, "display_value": "12%"}], "provider_page_url": "https://nativeport.ai/providers/oxylabs/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "scraperapi:scrape", "provider_id": "scraperapi", "provider_name": "ScraperAPI", "provider_group": "Scraping & Crawling", "provider_category": "Single-call scraping API", "capability_id": "scrape", "capability_label": "Scrape", "capability_description": "Fetch one URL as clean content/markdown, anti-bot handled", "composite_score": 5.68, "composite_scale_max": 10, "rank": 9, "rank_of": 10, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "block_bypass_success_rate", "metric_label": "Anti-bot bypass", "raw_value": 0.5, "display_value": "50%"}, {"metric_index": 1, "metric_key": "markdown_cleanliness", "metric_label": "Markdown clean", "raw_value": 3.75, "display_value": "3.8 / 10"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 2044.0, "display_value": "2.0 s"}, {"metric_index": 3, "metric_key": "cost_per_successful_page_usd", "metric_label": "Cost", "raw_value": 0.01274, "display_value": "$0.013 / success"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 50.0, "display_value": "50%"}], "provider_page_url": "https://nativeport.ai/providers/scraperapi/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "zenrows:scrape", "provider_id": "zenrows", "provider_name": "ZenRows", "provider_group": "Scraping & Crawling", "provider_category": "Scraper API + CDP browser", "capability_id": "scrape", "capability_label": "Scrape", "capability_description": "Fetch one URL as clean content/markdown, anti-bot handled", "composite_score": 1.1, "composite_scale_max": 10, "rank": 10, "rank_of": 10, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Every target errored this run — nothing came back to score.", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "block_bypass_success_rate", "metric_label": "Anti-bot bypass", "raw_value": 0.0, "display_value": "0%"}, {"metric_index": 1, "metric_key": "markdown_cleanliness", "metric_label": "Markdown clean", "raw_value": 0.0, "display_value": "0.0 / 10"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 0.0, "display_value": "0 ms"}, {"metric_index": 3, "metric_key": "cost_per_successful_page_usd", "metric_label": "Cost", "raw_value": 0.0588, "display_value": "$0.059 / success"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 100.0, "display_value": "100%"}], "provider_page_url": "https://nativeport.ai/providers/zenrows/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "oxylabs:scrape_domain", "provider_id": "oxylabs", "provider_name": "Oxylabs", "provider_group": "Scraping & Crawling", "provider_category": "Scraping API + proxy network", "capability_id": "scrape_domain", "capability_label": "Scrape-domain", "capability_description": "Purpose-built scrapers for hard, high-value domains (e-commerce, social)", "composite_score": 8.38, "composite_scale_max": 10, "rank": 1, "rank_of": 6, "is_capability_top": true, "measured_date": "2026-08-05", "note": "Leading scraper, though official first-party APIs win where available.", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "value_accuracy", "metric_label": "Accuracy", "raw_value": 0.86, "display_value": "0.86"}, {"metric_index": 1, "metric_key": "field_fill", "metric_label": "Field fill", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 2568.4, "display_value": "2.6 s"}, {"metric_index": 3, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.002326, "display_value": "$0.0023 / useful"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/oxylabs/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "scrapingbee:scrape_domain", "provider_id": "scrapingbee", "provider_name": "ScrapingBee", "provider_group": "Scraping & Crawling", "provider_category": "Single-call scraping API", "capability_id": "scrape_domain", "capability_label": "Scrape-domain", "capability_description": "Purpose-built scrapers for hard, high-value domains (e-commerce, social)", "composite_score": 8.1, "composite_scale_max": 10, "rank": 2, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "value_accuracy", "metric_label": "Accuracy", "raw_value": 0.877, "display_value": "0.88"}, {"metric_index": 1, "metric_key": "field_fill", "metric_label": "Field fill", "raw_value": 0.947, "display_value": "0.95"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 3486.9, "display_value": "3.5 s"}, {"metric_index": 3, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.011174, "display_value": "$0.011 / useful"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/scrapingbee/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "apify:scrape_domain", "provider_id": "apify", "provider_name": "Apify", "provider_group": "Browser Automation", "provider_category": "Actor marketplace", "capability_id": "scrape_domain", "capability_label": "Scrape-domain", "capability_description": "Purpose-built scrapers for hard, high-value domains (e-commerce, social)", "composite_score": 7.99, "composite_scale_max": 10, "rank": 3, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "value_accuracy", "metric_label": "Accuracy", "raw_value": 0.933, "display_value": "0.93"}, {"metric_index": 1, "metric_key": "field_fill", "metric_label": "Field fill", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 11182.2, "display_value": "11.2 s"}, {"metric_index": 3, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.003215, "display_value": "$0.0032 / useful"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/apify/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "crawlbase:scrape_domain", "provider_id": "crawlbase", "provider_name": "Crawlbase", "provider_group": "Scraping & Crawling", "provider_category": "Single-call + managed crawl", "capability_id": "scrape_domain", "capability_label": "Scrape-domain", "capability_description": "Purpose-built scrapers for hard, high-value domains (e-commerce, social)", "composite_score": 7.04, "composite_scale_max": 10, "rank": 4, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Instagram's anti-bot triggers non-deterministic behavior.", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "value_accuracy", "metric_label": "Accuracy", "raw_value": 0.746, "display_value": "0.75"}, {"metric_index": 1, "metric_key": "field_fill", "metric_label": "Field fill", "raw_value": 0.877, "display_value": "0.88"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 7911.4, "display_value": "7.9 s"}, {"metric_index": 3, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.002681, "display_value": "$0.0027 / useful"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/crawlbase/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "scraperapi:scrape_domain", "provider_id": "scraperapi", "provider_name": "ScraperAPI", "provider_group": "Scraping & Crawling", "provider_category": "Single-call scraping API", "capability_id": "scrape_domain", "capability_label": "Scrape-domain", "capability_description": "Purpose-built scrapers for hard, high-value domains (e-commerce, social)", "composite_score": 6.64, "composite_scale_max": 10, "rank": 5, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "value_accuracy", "metric_label": "Accuracy", "raw_value": 0.667, "display_value": "0.67"}, {"metric_index": 1, "metric_key": "field_fill", "metric_label": "Field fill", "raw_value": 0.778, "display_value": "0.78"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 4850.5, "display_value": "4.9 s"}, {"metric_index": 3, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.007346, "display_value": "$0.0073 / useful"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/scraperapi/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "brightdata:scrape_domain", "provider_id": "brightdata", "provider_name": "Bright Data", "provider_group": "Scraping & Crawling", "provider_category": "Web Unlocker + Scraper API", "capability_id": "scrape_domain", "capability_label": "Scrape-domain", "capability_description": "Purpose-built scrapers for hard, high-value domains (e-commerce, social)", "composite_score": 6.32, "composite_scale_max": 10, "rank": 6, "rank_of": 6, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "value_accuracy", "metric_label": "Accuracy", "raw_value": 0.694, "display_value": "0.69"}, {"metric_index": 1, "metric_key": "field_fill", "metric_label": "Field fill", "raw_value": 0.75, "display_value": "0.75"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 13905.6, "display_value": "13.9 s"}, {"metric_index": 3, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.003602, "display_value": "$0.0036 / useful"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 25.0, "display_value": "25%"}], "provider_page_url": "https://nativeport.ai/providers/brightdata/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "firecrawl:screenshot", "provider_id": "firecrawl", "provider_name": "Firecrawl", "provider_group": "Scraping & Crawling", "provider_category": "Scrape + crawl + extract", "capability_id": "screenshot", "capability_label": "Screenshot", "capability_description": "Rendered image of a URL", "composite_score": 9.27, "composite_scale_max": 10, "rank": 1, "rank_of": 8, "is_capability_top": true, "measured_date": "2026-08-05", "note": "Highest rate of valid images.", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "valid_image_rate", "metric_label": "Valid image", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "full_page_support", "metric_label": "Full-page", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 3583.6, "display_value": "3.6 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.001, "display_value": "$0.001 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/firecrawl/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "zyte:screenshot", "provider_id": "zyte", "provider_name": "Zyte", "provider_group": "Scraping & Crawling", "provider_category": "Single-endpoint extraction", "capability_id": "screenshot", "capability_label": "Screenshot", "capability_description": "Rendered image of a URL", "composite_score": 9.11, "composite_scale_max": 10, "rank": 2, "rank_of": 8, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "valid_image_rate", "metric_label": "Valid image", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 1, "metric_key": "full_page_support", "metric_label": "Full-page", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 5115.4, "display_value": "5.1 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.001, "display_value": "$0.001 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/zyte/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "scrapingbee:screenshot", "provider_id": "scrapingbee", "provider_name": "ScrapingBee", "provider_group": "Scraping & Crawling", "provider_category": "Single-call scraping API", "capability_id": "screenshot", "capability_label": "Screenshot", "capability_description": "Rendered image of a URL", "composite_score": 8.56, "composite_scale_max": 10, "rank": 3, "rank_of": 8, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "valid_image_rate", "metric_label": "Valid image", "raw_value": 0.875, "display_value": "0.88"}, {"metric_index": 1, "metric_key": "full_page_support", "metric_label": "Full-page", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 2760.0, "display_value": "2.8 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.0028, "display_value": "$0.0028 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 12.5, "display_value": "12%"}], "provider_page_url": "https://nativeport.ai/providers/scrapingbee/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "scrapfly:screenshot", "provider_id": "scrapfly", "provider_name": "Scrapfly", "provider_group": "Scraping & Crawling", "provider_category": "Single-call scraping API", "capability_id": "screenshot", "capability_label": "Screenshot", "capability_description": "Rendered image of a URL", "composite_score": 7.87, "composite_scale_max": 10, "rank": 4, "rank_of": 8, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "valid_image_rate", "metric_label": "Valid image", "raw_value": 0.875, "display_value": "0.88"}, {"metric_index": 1, "metric_key": "full_page_support", "metric_label": "Full-page", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 10490.8, "display_value": "10.5 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.002, "display_value": "$0.002 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 12.5, "display_value": "12%"}], "provider_page_url": "https://nativeport.ai/providers/scrapfly/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "crawlbase:screenshot", "provider_id": "crawlbase", "provider_name": "Crawlbase", "provider_group": "Scraping & Crawling", "provider_category": "Single-call + managed crawl", "capability_id": "screenshot", "capability_label": "Screenshot", "capability_description": "Rendered image of a URL", "composite_score": 7.84, "composite_scale_max": 10, "rank": 5, "rank_of": 8, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "valid_image_rate", "metric_label": "Valid image", "raw_value": 0.75, "display_value": "0.75"}, {"metric_index": 1, "metric_key": "full_page_support", "metric_label": "Full-page", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 3694.6, "display_value": "3.7 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.0015, "display_value": "$0.0015 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 12.5, "display_value": "12%"}], "provider_page_url": "https://nativeport.ai/providers/crawlbase/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "zenrows:screenshot", "provider_id": "zenrows", "provider_name": "ZenRows", "provider_group": "Scraping & Crawling", "provider_category": "Scraper API + CDP browser", "capability_id": "screenshot", "capability_label": "Screenshot", "capability_description": "Rendered image of a URL", "composite_score": 7.16, "composite_scale_max": 10, "rank": 6, "rank_of": 8, "is_capability_top": false, "measured_date": "2026-08-05", "note": "books.toscrape blocks it — a gap openly noted.", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "valid_image_rate", "metric_label": "Valid image", "raw_value": 0.625, "display_value": "0.62"}, {"metric_index": 1, "metric_key": "full_page_support", "metric_label": "Full-page", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 3681.1, "display_value": "3.7 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.0014, "display_value": "$0.0014 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 37.5, "display_value": "38%"}], "provider_page_url": "https://nativeport.ai/providers/zenrows/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "scraperapi:screenshot", "provider_id": "scraperapi", "provider_name": "ScraperAPI", "provider_group": "Scraping & Crawling", "provider_category": "Single-call scraping API", "capability_id": "screenshot", "capability_label": "Screenshot", "capability_description": "Rendered image of a URL", "composite_score": 7.09, "composite_scale_max": 10, "rank": 7, "rank_of": 8, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "valid_image_rate", "metric_label": "Valid image", "raw_value": 0.75, "display_value": "0.75"}, {"metric_index": 1, "metric_key": "full_page_support", "metric_label": "Full-page", "raw_value": 1.0, "display_value": "1.00"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 30994.7, "display_value": "31.0 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.0049, "display_value": "$0.0049 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 25.0, "display_value": "25%"}], "provider_page_url": "https://nativeport.ai/providers/scraperapi/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "spider:screenshot", "provider_id": "spider", "provider_name": "Spider", "provider_group": "Scraping & Crawling", "provider_category": "Crawl / AI-focused platform", "capability_id": "screenshot", "capability_label": "Screenshot", "capability_description": "Rendered image of a URL", "composite_score": 1.75, "composite_scale_max": 10, "rank": 8, "rank_of": 8, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Every capture errored; no image came back.", "metric_count": 3, "metrics": [{"metric_index": 0, "metric_key": "valid_image_rate", "metric_label": "Valid image", "raw_value": 0.0, "display_value": "0.00"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 0.0, "display_value": "0 ms"}, {"metric_index": 2, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 100.0, "display_value": "100%"}], "provider_page_url": "https://nativeport.ai/providers/spider/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "jina:search", "provider_id": "jina", "provider_name": "Jina", "provider_group": "Search", "provider_category": "Search + read-to-markdown", "capability_id": "search", "capability_label": "Search", "capability_description": "Web search: a query in, ranked result links out", "composite_score": 7.96, "composite_scale_max": 10, "rank": 1, "rank_of": 11, "is_capability_top": true, "measured_date": "2026-08-05", "note": "First across all three query classes.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "recall_at_10", "metric_label": "Recall@10", "raw_value": 0.62, "display_value": "0.62"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 1023.8, "display_value": "1.0 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.000322, "display_value": "$0.00032 / useful result"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/jina/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "serper:search", "provider_id": "serper", "provider_name": "Serper", "provider_group": "Search", "provider_category": "Google SERP scrape", "capability_id": "search", "capability_label": "Search", "capability_description": "Web search: a query in, ranked result links out", "composite_score": 7.79, "composite_scale_max": 10, "rank": 2, "rank_of": 11, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "recall_at_10", "metric_label": "Recall@10", "raw_value": 0.62, "display_value": "0.62"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 815.5, "display_value": "816 ms"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.000484, "display_value": "$0.00048 / useful result"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/serper/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "youcom:search", "provider_id": "youcom", "provider_name": "You.com", "provider_group": "Search", "provider_category": "Own crawl index + Research API", "capability_id": "search", "capability_label": "Search", "capability_description": "Web search: a query in, ranked result links out", "composite_score": 7.6, "composite_scale_max": 10, "rank": 3, "rank_of": 11, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "recall_at_10", "metric_label": "Recall@10", "raw_value": 0.606, "display_value": "0.61"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 566.2, "display_value": "566 ms"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.008244, "display_value": "$0.0082 / useful result"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/youcom/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "serpapi:search", "provider_id": "serpapi", "provider_name": "SerpApi", "provider_group": "Search", "provider_category": "Multi-engine SERP scrape", "capability_id": "search", "capability_label": "Search", "capability_description": "Web search: a query in, ranked result links out", "composite_score": 7.42, "composite_scale_max": 10, "rank": 4, "rank_of": 11, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "recall_at_10", "metric_label": "Recall@10", "raw_value": 0.588, "display_value": "0.59"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 181.2, "display_value": "181 ms"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.015307, "display_value": "$0.015 / useful result"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/serpapi/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "brave:search", "provider_id": "brave", "provider_name": "Brave Search", "provider_group": "Search", "provider_category": "Independent crawler index", "capability_id": "search", "capability_label": "Search", "capability_description": "Web search: a query in, ranked result links out", "composite_score": 7.31, "composite_scale_max": 10, "rank": 5, "rank_of": 11, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "recall_at_10", "metric_label": "Recall@10", "raw_value": 0.62, "display_value": "0.62"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 470.1, "display_value": "470 ms"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.00806, "display_value": "$0.0081 / useful result"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/brave/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "exa:search", "provider_id": "exa", "provider_name": "Exa", "provider_group": "Search", "provider_category": "Neural + keyword index", "capability_id": "search", "capability_label": "Search", "capability_description": "Web search: a query in, ranked result links out", "composite_score": 6.9, "composite_scale_max": 10, "rank": 6, "rank_of": 11, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "recall_at_10", "metric_label": "Recall@10", "raw_value": 0.593, "display_value": "0.59"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 1078.9, "display_value": "1.1 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.014344, "display_value": "$0.014 / useful result"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/exa/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "parallel:search", "provider_id": "parallel", "provider_name": "Parallel", "provider_group": "Search", "provider_category": "Agent-optimized search index", "capability_id": "search", "capability_label": "Search", "capability_description": "Web search: a query in, ranked result links out", "composite_score": 6.77, "composite_scale_max": 10, "rank": 7, "rank_of": 11, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "recall_at_10", "metric_label": "Recall@10", "raw_value": 0.588, "display_value": "0.59"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 2151.2, "display_value": "2.2 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.006803, "display_value": "$0.0068 / useful result"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/parallel/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "tavily:search", "provider_id": "tavily", "provider_name": "Tavily", "provider_group": "Search", "provider_category": "Agent-native search index", "capability_id": "search", "capability_label": "Search", "capability_description": "Web search: a query in, ranked result links out", "composite_score": 6.63, "composite_scale_max": 10, "rank": 8, "rank_of": 11, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "recall_at_10", "metric_label": "Recall@10", "raw_value": 0.551, "display_value": "0.55"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 1882.1, "display_value": "1.9 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.014521, "display_value": "$0.015 / useful result"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/tavily/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "searchapi:search", "provider_id": "searchapi", "provider_name": "SearchAPI.io", "provider_group": "Search", "provider_category": "Multi-engine SERP scrape", "capability_id": "search", "capability_label": "Search", "capability_description": "Web search: a query in, ranked result links out", "composite_score": 6.27, "composite_scale_max": 10, "rank": 9, "rank_of": 11, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "recall_at_10", "metric_label": "Recall@10", "raw_value": 0.579, "display_value": "0.58"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 2879.7, "display_value": "2.9 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.006912, "display_value": "$0.0069 / useful result"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/searchapi/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "dataforseo:search", "provider_id": "dataforseo", "provider_name": "DataForSEO", "provider_group": "Search", "provider_category": "SEO & SERP data platform", "capability_id": "search", "capability_label": "Search", "capability_description": "Web search: a query in, ranked result links out", "composite_score": 5.86, "composite_scale_max": 10, "rank": 10, "rank_of": 11, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "recall_at_10", "metric_label": "Recall@10", "raw_value": 0.588, "display_value": "0.59"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 9929.0, "display_value": "9.9 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.003402, "display_value": "$0.0034 / useful result"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/dataforseo/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "linkup:search", "provider_id": "linkup", "provider_name": "Linkup", "provider_group": "Search", "provider_category": "Own crawl index", "capability_id": "search", "capability_label": "Search", "capability_description": "Web search: a query in, ranked result links out", "composite_score": 4.86, "composite_scale_max": 10, "rank": 11, "rank_of": 11, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "recall_at_10", "metric_label": "Recall@10", "raw_value": 0.236, "display_value": "0.24"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 2786.3, "display_value": "2.8 s"}, {"metric_index": 2, "metric_key": "cost_per_useful_usd", "metric_label": "Cost", "raw_value": 0.021176, "display_value": "$0.021 / useful result"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/linkup/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "serper:serp", "provider_id": "serper", "provider_name": "Serper", "provider_group": "Search", "provider_category": "Google SERP scrape", "capability_id": "serp", "capability_label": "SERP verticals", "capability_description": "Google SERP verticals (web, news, images, places, scholar) as structured JSON", "composite_score": 9.13, "composite_scale_max": 10, "rank": 1, "rank_of": 4, "is_capability_top": true, "measured_date": "2026-08-05", "note": "Lowest cost and latency; Google only.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "quality", "metric_label": "Quality", "raw_value": 0.938, "display_value": "0.94"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 886.0, "display_value": "886 ms"}, {"metric_index": 2, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.0003, "display_value": "$0.0003 / call"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/serper/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "searchapi:serp", "provider_id": "searchapi", "provider_name": "SearchAPI.io", "provider_group": "Search", "provider_category": "Multi-engine SERP scrape", "capability_id": "serp", "capability_label": "SERP verticals", "capability_description": "Google SERP verticals (web, news, images, places, scholar) as structured JSON", "composite_score": 9.01, "composite_scale_max": 10, "rank": 2, "rank_of": 4, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Widest vertical coverage, with SerpApi-compatible JSON.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "quality", "metric_label": "Quality", "raw_value": 0.977, "display_value": "0.98"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 1529.8, "display_value": "1.5 s"}, {"metric_index": 2, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.004, "display_value": "$0.004 / call"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/searchapi/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "serpapi:serp", "provider_id": "serpapi", "provider_name": "SerpApi", "provider_group": "Search", "provider_category": "Multi-engine SERP scrape", "capability_id": "serp", "capability_label": "SERP verticals", "capability_description": "Google SERP verticals (web, news, images, places, scholar) as structured JSON", "composite_score": 8.62, "composite_scale_max": 10, "rank": 3, "rank_of": 4, "is_capability_top": false, "measured_date": "2026-08-05", "note": "Every tested vertical is covered.", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "quality", "metric_label": "Quality", "raw_value": 0.977, "display_value": "0.98"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 3635.9, "display_value": "3.6 s"}, {"metric_index": 2, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.015, "display_value": "$0.015 / call"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/serpapi/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "dataforseo:serp", "provider_id": "dataforseo", "provider_name": "DataForSEO", "provider_group": "Search", "provider_category": "SEO & SERP data platform", "capability_id": "serp", "capability_label": "SERP verticals", "capability_description": "Google SERP verticals (web, news, images, places, scholar) as structured JSON", "composite_score": 7.41, "composite_scale_max": 10, "rank": 4, "rank_of": 4, "is_capability_top": false, "measured_date": "2026-08-05", "note": "", "metric_count": 4, "metrics": [{"metric_index": 0, "metric_key": "quality", "metric_label": "Quality", "raw_value": 0.857, "display_value": "0.86"}, {"metric_index": 1, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 24600.2, "display_value": "24.6 s"}, {"metric_index": 2, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.002, "display_value": "$0.002 / call"}, {"metric_index": 3, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 14.3, "display_value": "14%"}], "provider_page_url": "https://nativeport.ai/providers/dataforseo/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}
+{"evaluation_id": "firecrawl:watch", "provider_id": "firecrawl", "provider_name": "Firecrawl", "provider_group": "Scraping & Crawling", "provider_category": "Scrape + crawl + extract", "capability_id": "watch", "capability_label": "Watch", "capability_description": "Change detection / monitoring on a URL", "composite_score": 7.38, "composite_scale_max": 10, "rank": 1, "rank_of": 1, "is_capability_top": true, "measured_date": "2026-08-05", "note": "Uniquely offers synchronous change detection.", "metric_count": 5, "metrics": [{"metric_index": 0, "metric_key": "classification_accuracy", "metric_label": "Class. accuracy", "raw_value": 0.875, "display_value": "0.88"}, {"metric_index": 1, "metric_key": "diff_quality", "metric_label": "Diff quality", "raw_value": 0.5, "display_value": "0.50"}, {"metric_index": 2, "metric_key": "latency_p50_ms", "metric_label": "Latency p50", "raw_value": 5539.8, "display_value": "5.5 s"}, {"metric_index": 3, "metric_key": "cost_per_call_usd", "metric_label": "Cost", "raw_value": 0.001, "display_value": "$0.001 / call"}, {"metric_index": 4, "metric_key": "error_rate_pct", "metric_label": "Errors", "raw_value": 0.0, "display_value": "0%"}], "provider_page_url": "https://nativeport.ai/providers/firecrawl/", "run_id": "2026-08-05", "source_url": "https://nativeport.ai/evals.json", "source_schema_version": 1, "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b"}

data/summary.json

@@ -0,0 +1,919 @@
+{
+  "dataset_name": "NativePort Web-Access API Benchmarks",
+  "intended_dataset_id": "nativeport/web-access-api-benchmarks",
+  "generator": "scripts/build_dataset.py",
+  "generator_version": "1.0.0",
+  "source_url": "https://nativeport.ai/evals.json",
+  "methodology_url": "https://nativeport.ai/methodology/",
+  "leaderboards_url": "https://nativeport.ai/leaderboards/",
+  "source_schema_version": 1,
+  "source_bytes": 89894,
+  "snapshot_sha256": "f46bf416803d5adf696c66ac14a6cbf06f4dfa39727cca164e8f3872abd6ed9b",
+  "latest_run": "2026-08-05",
+  "run_id": "2026-08-05",
+  "provider_count_represented": 22,
+  "provider_count_in_source_catalog": 32,
+  "provider_count_in_catalog_without_evaluations": 10,
+  "evaluation_count": 67,
+  "capability_count": 13,
+  "metric_row_count": 297,
+  "unscored_pair_count_in_source": 5,
+  "measured_date_min": "2026-08-05",
+  "measured_date_max": "2026-08-05",
+  "measured_dates": [
+    "2026-08-05"
+  ],
+  "composite_scale_max": 10,
+  "source_reported_run_totals": {
+    "scored_providers": 22,
+    "scorecards": 67,
+    "capabilities": 13,
+    "catalog_providers": 32
+  },
+  "cross_check": {
+    "scored_providers_matches": true,
+    "scorecards_matches": true,
+    "capabilities_matches": true,
+    "catalog_providers_matches": true
+  },
+  "capabilities": [
+    {
+      "capability_id": "act",
+      "capability_label": "Act · declarative",
+      "capability_description": "Declarative in-page interactions (click, fill, scroll) in one call",
+      "evaluation_count": 6,
+      "rank_of_values": [
+        6
+      ],
+      "rank_of_matches_evaluation_count": true,
+      "provider_ids": [
+        "firecrawl",
+        "scrapfly",
+        "scrapingbee",
+        "spider",
+        "zenrows",
+        "zyte"
+      ],
+      "metric_keys": [
+        "cost_per_call_usd",
+        "error_rate_pct",
+        "latency_p50_ms",
+        "task_success"
+      ],
+      "metric_row_count": 24,
+      "composite_score_min": 1.5,
+      "composite_score_max": 8.68
+    },
+    {
+      "capability_id": "act_agent",
+      "capability_label": "Act · NL-agent",
+      "capability_description": "A natural-language goal handed to an autonomous browser agent",
+      "evaluation_count": 1,
+      "rank_of_values": [
+        1
+      ],
+      "rank_of_matches_evaluation_count": true,
+      "provider_ids": [
+        "firecrawl"
+      ],
+      "metric_keys": [
+        "cost_per_call_usd",
+        "error_rate_pct",
+        "latency_p50_ms",
+        "task_success"
+      ],
+      "metric_row_count": 4,
+      "composite_score_min": 7.99,
+      "composite_score_max": 7.99
+    },
+    {
+      "capability_id": "answer",
+      "capability_label": "Answer",
+      "capability_description": "A question in, a synthesized answer with cited sources out",
+      "evaluation_count": 6,
+      "rank_of_values": [
+        6
+      ],
+      "rank_of_matches_evaluation_count": true,
+      "provider_ids": [
+        "exa",
+        "jina",
+        "linkup",
+        "parallel",
+        "tavily",
+        "youcom"
+      ],
+      "metric_keys": [
+        "answer_correctness",
+        "citation_faithfulness",
+        "cost_per_call_usd",
+        "error_rate_pct",
+        "latency_p50_ms"
+      ],
+      "metric_row_count": 30,
+      "composite_score_min": 6.65,
+      "composite_score_max": 8.33
+    },
+    {
+      "capability_id": "crawl",
+      "capability_label": "Crawl",
+      "capability_description": "Enumerate and fetch many pages across a site",
+      "evaluation_count": 4,
+      "rank_of_values": [
+        4
+      ],
+      "rank_of_matches_evaluation_count": true,
+      "provider_ids": [
+        "brightdata",
+        "firecrawl",
+        "spider",
+        "tavily"
+      ],
+      "metric_keys": [
+        "cost_per_useful_usd",
+        "coverage",
+        "error_rate_pct",
+        "latency_p50_ms"
+      ],
+      "metric_row_count": 16,
+      "composite_score_min": 3.88,
+      "composite_score_max": 8.91
+    },
+    {
+      "capability_id": "extract_ai",
+      "capability_label": "Extract · AI/schema",
+      "capability_description": "Structured fields pulled from a page against an AI/natural-language schema",
+      "evaluation_count": 4,
+      "rank_of_values": [
+        4
+      ],
+      "rank_of_matches_evaluation_count": true,
+      "provider_ids": [
+        "firecrawl",
+        "scrapfly",
+        "scrapingbee",
+        "zyte"
+      ],
+      "metric_keys": [
+        "cost_per_useful_usd",
+        "error_rate_pct",
+        "field_accuracy",
+        "latency_p50_ms"
+      ],
+      "metric_row_count": 16,
+      "composite_score_min": 5.17,
+      "composite_score_max": 9.0
+    },
+    {
+      "capability_id": "extract_rules",
+      "capability_label": "Extract · CSS rules",
+      "capability_description": "Structured fields pulled from a page via CSS/XPath rules",
+      "evaluation_count": 3,
+      "rank_of_values": [
+        3
+      ],
+      "rank_of_matches_evaluation_count": true,
+      "provider_ids": [
+        "oxylabs",
+        "scrapingbee",
+        "zenrows"
+      ],
+      "metric_keys": [
+        "cost_per_useful_usd",
+        "error_rate_pct",
+        "field_accuracy",
+        "latency_p50_ms"
+      ],
+      "metric_row_count": 12,
+      "composite_score_min": 5.77,
+      "composite_score_max": 9.23
+    },
+    {
+      "capability_id": "parse",
+      "capability_label": "Parse · PDF/doc",
+      "capability_description": "PDF/document to text, with OCR where the provider supports it",
+      "evaluation_count": 3,
+      "rank_of_values": [
+        3
+      ],
+      "rank_of_matches_evaluation_count": true,
+      "provider_ids": [
+        "firecrawl",
+        "jina",
+        "spider"
+      ],
+      "metric_keys": [
+        "cost_per_useful_usd",
+        "error_rate_pct",
+        "latency_p50_ms",
+        "text_accuracy"
+      ],
+      "metric_row_count": 12,
+      "composite_score_min": 6.96,
+      "composite_score_max": 8.81
+    },
+    {
+      "capability_id": "scrape",
+      "capability_label": "Scrape",
+      "capability_description": "Fetch one URL as clean content/markdown, anti-bot handled",
+      "evaluation_count": 10,
+      "rank_of_values": [
+        10
+      ],
+      "rank_of_matches_evaluation_count": true,
+      "provider_ids": [
+        "brightdata",
+        "crawlbase",
+        "firecrawl",
+        "oxylabs",
+        "scraperapi",
+        "scrapfly",
+        "scrapingbee",
+        "spider",
+        "zenrows",
+        "zyte"
+      ],
+      "metric_keys": [
+        "block_bypass_success_rate",
+        "cost_per_successful_page_usd",
+        "error_rate_pct",
+        "latency_p50_ms",
+        "markdown_cleanliness"
+      ],
+      "metric_row_count": 50,
+      "composite_score_min": 1.1,
+      "composite_score_max": 9.85
+    },
+    {
+      "capability_id": "scrape_domain",
+      "capability_label": "Scrape-domain",
+      "capability_description": "Purpose-built scrapers for hard, high-value domains (e-commerce, social)",
+      "evaluation_count": 6,
+      "rank_of_values": [
+        6
+      ],
+      "rank_of_matches_evaluation_count": true,
+      "provider_ids": [
+        "apify",
+        "brightdata",
+        "crawlbase",
+        "oxylabs",
+        "scraperapi",
+        "scrapingbee"
+      ],
+      "metric_keys": [
+        "cost_per_useful_usd",
+        "error_rate_pct",
+        "field_fill",
+        "latency_p50_ms",
+        "value_accuracy"
+      ],
+      "metric_row_count": 30,
+      "composite_score_min": 6.32,
+      "composite_score_max": 8.38
+    },
+    {
+      "capability_id": "screenshot",
+      "capability_label": "Screenshot",
+      "capability_description": "Rendered image of a URL",
+      "evaluation_count": 8,
+      "rank_of_values": [
+        8
+      ],
+      "rank_of_matches_evaluation_count": true,
+      "provider_ids": [
+        "crawlbase",
+        "firecrawl",
+        "scraperapi",
+        "scrapfly",
+        "scrapingbee",
+        "spider",
+        "zenrows",
+        "zyte"
+      ],
+      "metric_keys": [
+        "cost_per_call_usd",
+        "error_rate_pct",
+        "full_page_support",
+        "latency_p50_ms",
+        "valid_image_rate"
+      ],
+      "metric_row_count": 38,
+      "composite_score_min": 1.75,
+      "composite_score_max": 9.27
+    },
+    {
+      "capability_id": "search",
+      "capability_label": "Search",
+      "capability_description": "Web search: a query in, ranked result links out",
+      "evaluation_count": 11,
+      "rank_of_values": [
+        11
+      ],
+      "rank_of_matches_evaluation_count": true,
+      "provider_ids": [
+        "brave",
+        "dataforseo",
+        "exa",
+        "jina",
+        "linkup",
+        "parallel",
+        "searchapi",
+        "serpapi",
+        "serper",
+        "tavily",
+        "youcom"
+      ],
+      "metric_keys": [
+        "cost_per_useful_usd",
+        "error_rate_pct",
+        "latency_p50_ms",
+        "recall_at_10"
+      ],
+      "metric_row_count": 44,
+      "composite_score_min": 4.86,
+      "composite_score_max": 7.96
+    },
+    {
+      "capability_id": "serp",
+      "capability_label": "SERP verticals",
+      "capability_description": "Google SERP verticals (web, news, images, places, scholar) as structured JSON",
+      "evaluation_count": 4,
+      "rank_of_values": [
+        4
+      ],
+      "rank_of_matches_evaluation_count": true,
+      "provider_ids": [
+        "dataforseo",
+        "searchapi",
+        "serpapi",
+        "serper"
+      ],
+      "metric_keys": [
+        "cost_per_call_usd",
+        "error_rate_pct",
+        "latency_p50_ms",
+        "quality"
+      ],
+      "metric_row_count": 16,
+      "composite_score_min": 7.41,
+      "composite_score_max": 9.13
+    },
+    {
+      "capability_id": "watch",
+      "capability_label": "Watch",
+      "capability_description": "Change detection / monitoring on a URL",
+      "evaluation_count": 1,
+      "rank_of_values": [
+        1
+      ],
+      "rank_of_matches_evaluation_count": true,
+      "provider_ids": [
+        "firecrawl"
+      ],
+      "metric_keys": [
+        "classification_accuracy",
+        "cost_per_call_usd",
+        "diff_quality",
+        "error_rate_pct",
+        "latency_p50_ms"
+      ],
+      "metric_row_count": 5,
+      "composite_score_min": 7.38,
+      "composite_score_max": 7.38
+    }
+  ],
+  "providers": [
+    {
+      "provider_id": "apify",
+      "provider_name": "Apify",
+      "provider_group": "Browser Automation",
+      "evaluation_count": 1,
+      "capability_ids": [
+        "scrape_domain"
+      ]
+    },
+    {
+      "provider_id": "brave",
+      "provider_name": "Brave Search",
+      "provider_group": "Search",
+      "evaluation_count": 1,
+      "capability_ids": [
+        "search"
+      ]
+    },
+    {
+      "provider_id": "brightdata",
+      "provider_name": "Bright Data",
+      "provider_group": "Scraping & Crawling",
+      "evaluation_count": 3,
+      "capability_ids": [
+        "crawl",
+        "scrape",
+        "scrape_domain"
+      ]
+    },
+    {
+      "provider_id": "crawlbase",
+      "provider_name": "Crawlbase",
+      "provider_group": "Scraping & Crawling",
+      "evaluation_count": 3,
+      "capability_ids": [
+        "scrape",
+        "scrape_domain",
+        "screenshot"
+      ]
+    },
+    {
+      "provider_id": "dataforseo",
+      "provider_name": "DataForSEO",
+      "provider_group": "Search",
+      "evaluation_count": 2,
+      "capability_ids": [
+        "search",
+        "serp"
+      ]
+    },
+    {
+      "provider_id": "exa",
+      "provider_name": "Exa",
+      "provider_group": "Search",
+      "evaluation_count": 2,
+      "capability_ids": [
+        "answer",
+        "search"
+      ]
+    },
+    {
+      "provider_id": "firecrawl",
+      "provider_name": "Firecrawl",
+      "provider_group": "Scraping & Crawling",
+      "evaluation_count": 8,
+      "capability_ids": [
+        "act",
+        "act_agent",
+        "crawl",
+        "extract_ai",
+        "parse",
+        "scrape",
+        "screenshot",
+        "watch"
+      ]
+    },
+    {
+      "provider_id": "jina",
+      "provider_name": "Jina",
+      "provider_group": "Search",
+      "evaluation_count": 3,
+      "capability_ids": [
+        "answer",
+        "parse",
+        "search"
+      ]
+    },
+    {
+      "provider_id": "linkup",
+      "provider_name": "Linkup",
+      "provider_group": "Search",
+      "evaluation_count": 2,
+      "capability_ids": [
+        "answer",
+        "search"
+      ]
+    },
+    {
+      "provider_id": "oxylabs",
+      "provider_name": "Oxylabs",
+      "provider_group": "Scraping & Crawling",
+      "evaluation_count": 3,
+      "capability_ids": [
+        "extract_rules",
+        "scrape",
+        "scrape_domain"
+      ]
+    },
+    {
+      "provider_id": "parallel",
+      "provider_name": "Parallel",
+      "provider_group": "Search",
+      "evaluation_count": 2,
+      "capability_ids": [
+        "answer",
+        "search"
+      ]
+    },
+    {
+      "provider_id": "scraperapi",
+      "provider_name": "ScraperAPI",
+      "provider_group": "Scraping & Crawling",
+      "evaluation_count": 3,
+      "capability_ids": [
+        "scrape",
+        "scrape_domain",
+        "screenshot"
+      ]
+    },
+    {
+      "provider_id": "scrapfly",
+      "provider_name": "Scrapfly",
+      "provider_group": "Scraping & Crawling",
+      "evaluation_count": 4,
+      "capability_ids": [
+        "act",
+        "extract_ai",
+        "scrape",
+        "screenshot"
+      ]
+    },
+    {
+      "provider_id": "scrapingbee",
+      "provider_name": "ScrapingBee",
+      "provider_group": "Scraping & Crawling",
+      "evaluation_count": 6,
+      "capability_ids": [
+        "act",
+        "extract_ai",
+        "extract_rules",
+        "scrape",
+        "scrape_domain",
+        "screenshot"
+      ]
+    },
+    {
+      "provider_id": "searchapi",
+      "provider_name": "SearchAPI.io",
+      "provider_group": "Search",
+      "evaluation_count": 2,
+      "capability_ids": [
+        "search",
+        "serp"
+      ]
+    },
+    {
+      "provider_id": "serpapi",
+      "provider_name": "SerpApi",
+      "provider_group": "Search",
+      "evaluation_count": 2,
+      "capability_ids": [
+        "search",
+        "serp"
+      ]
+    },
+    {
+      "provider_id": "serper",
+      "provider_name": "Serper",
+      "provider_group": "Search",
+      "evaluation_count": 2,
+      "capability_ids": [
+        "search",
+        "serp"
+      ]
+    },
+    {
+      "provider_id": "spider",
+      "provider_name": "Spider",
+      "provider_group": "Scraping & Crawling",
+      "evaluation_count": 5,
+      "capability_ids": [
+        "act",
+        "crawl",
+        "parse",
+        "scrape",
+        "screenshot"
+      ]
+    },
+    {
+      "provider_id": "tavily",
+      "provider_name": "Tavily",
+      "provider_group": "Search",
+      "evaluation_count": 3,
+      "capability_ids": [
+        "answer",
+        "crawl",
+        "search"
+      ]
+    },
+    {
+      "provider_id": "youcom",
+      "provider_name": "You.com",
+      "provider_group": "Search",
+      "evaluation_count": 2,
+      "capability_ids": [
+        "answer",
+        "search"
+      ]
+    },
+    {
+      "provider_id": "zenrows",
+      "provider_name": "ZenRows",
+      "provider_group": "Scraping & Crawling",
+      "evaluation_count": 4,
+      "capability_ids": [
+        "act",
+        "extract_rules",
+        "scrape",
+        "screenshot"
+      ]
+    },
+    {
+      "provider_id": "zyte",
+      "provider_name": "Zyte",
+      "provider_group": "Scraping & Crawling",
+      "evaluation_count": 4,
+      "capability_ids": [
+        "act",
+        "extract_ai",
+        "scrape",
+        "screenshot"
+      ]
+    }
+  ],
+  "metric_keys": [
+    {
+      "metric_key": "answer_correctness",
+      "metric_labels": [
+        "Correctness"
+      ],
+      "occurrence_count": 6,
+      "capability_ids": [
+        "answer"
+      ]
+    },
+    {
+      "metric_key": "block_bypass_success_rate",
+      "metric_labels": [
+        "Anti-bot bypass"
+      ],
+      "occurrence_count": 10,
+      "capability_ids": [
+        "scrape"
+      ]
+    },
+    {
+      "metric_key": "citation_faithfulness",
+      "metric_labels": [
+        "Citation faithfulness"
+      ],
+      "occurrence_count": 6,
+      "capability_ids": [
+        "answer"
+      ]
+    },
+    {
+      "metric_key": "classification_accuracy",
+      "metric_labels": [
+        "Class. accuracy"
+      ],
+      "occurrence_count": 1,
+      "capability_ids": [
+        "watch"
+      ]
+    },
+    {
+      "metric_key": "cost_per_call_usd",
+      "metric_labels": [
+        "Cost"
+      ],
+      "occurrence_count": 25,
+      "capability_ids": [
+        "act",
+        "act_agent",
+        "answer",
+        "screenshot",
+        "serp",
+        "watch"
+      ]
+    },
+    {
+      "metric_key": "cost_per_successful_page_usd",
+      "metric_labels": [
+        "Cost"
+      ],
+      "occurrence_count": 10,
+      "capability_ids": [
+        "scrape"
+      ]
+    },
+    {
+      "metric_key": "cost_per_useful_usd",
+      "metric_labels": [
+        "Cost"
+      ],
+      "occurrence_count": 31,
+      "capability_ids": [
+        "crawl",
+        "extract_ai",
+        "extract_rules",
+        "parse",
+        "scrape_domain",
+        "search"
+      ]
+    },
+    {
+      "metric_key": "coverage",
+      "metric_labels": [
+        "Coverage"
+      ],
+      "occurrence_count": 4,
+      "capability_ids": [
+        "crawl"
+      ]
+    },
+    {
+      "metric_key": "diff_quality",
+      "metric_labels": [
+        "Diff quality"
+      ],
+      "occurrence_count": 1,
+      "capability_ids": [
+        "watch"
+      ]
+    },
+    {
+      "metric_key": "error_rate_pct",
+      "metric_labels": [
+        "Errors"
+      ],
+      "occurrence_count": 67,
+      "capability_ids": [
+        "act",
+        "act_agent",
+        "answer",
+        "crawl",
+        "extract_ai",
+        "extract_rules",
+        "parse",
+        "scrape",
+        "scrape_domain",
+        "screenshot",
+        "search",
+        "serp",
+        "watch"
+      ]
+    },
+    {
+      "metric_key": "field_accuracy",
+      "metric_labels": [
+        "Field accuracy"
+      ],
+      "occurrence_count": 7,
+      "capability_ids": [
+        "extract_ai",
+        "extract_rules"
+      ]
+    },
+    {
+      "metric_key": "field_fill",
+      "metric_labels": [
+        "Field fill"
+      ],
+      "occurrence_count": 6,
+      "capability_ids": [
+        "scrape_domain"
+      ]
+    },
+    {
+      "metric_key": "full_page_support",
+      "metric_labels": [
+        "Full-page"
+      ],
+      "occurrence_count": 7,
+      "capability_ids": [
+        "screenshot"
+      ]
+    },
+    {
+      "metric_key": "latency_p50_ms",
+      "metric_labels": [
+        "Latency p50"
+      ],
+      "occurrence_count": 67,
+      "capability_ids": [
+        "act",
+        "act_agent",
+        "answer",
+        "crawl",
+        "extract_ai",
+        "extract_rules",
+        "parse",
+        "scrape",
+        "scrape_domain",
+        "screenshot",
+        "search",
+        "serp",
+        "watch"
+      ]
+    },
+    {
+      "metric_key": "markdown_cleanliness",
+      "metric_labels": [
+        "Markdown clean"
+      ],
+      "occurrence_count": 10,
+      "capability_ids": [
+        "scrape"
+      ]
+    },
+    {
+      "metric_key": "quality",
+      "metric_labels": [
+        "Quality"
+      ],
+      "occurrence_count": 4,
+      "capability_ids": [
+        "serp"
+      ]
+    },
+    {
+      "metric_key": "recall_at_10",
+      "metric_labels": [
+        "Recall@10"
+      ],
+      "occurrence_count": 11,
+      "capability_ids": [
+        "search"
+      ]
+    },
+    {
+      "metric_key": "task_success",
+      "metric_labels": [
+        "Task success"
+      ],
+      "occurrence_count": 7,
+      "capability_ids": [
+        "act",
+        "act_agent"
+      ]
+    },
+    {
+      "metric_key": "text_accuracy",
+      "metric_labels": [
+        "Text accuracy"
+      ],
+      "occurrence_count": 3,
+      "capability_ids": [
+        "parse"
+      ]
+    },
+    {
+      "metric_key": "valid_image_rate",
+      "metric_labels": [
+        "Valid image"
+      ],
+      "occurrence_count": 8,
+      "capability_ids": [
+        "screenshot"
+      ]
+    },
+    {
+      "metric_key": "value_accuracy",
+      "metric_labels": [
+        "Accuracy"
+      ],
+      "occurrence_count": 6,
+      "capability_ids": [
+        "scrape_domain"
+      ]
+    }
+  ],
+  "outputs": {
+    "benchmarks.csv": {
+      "kind": "csv",
+      "data_rows": 297,
+      "columns": 23,
+      "column_names": [
+        "evaluation_id",
+        "provider_id",
+        "provider_name",
+        "provider_group",
+        "provider_category",
+        "capability_id",
+        "capability_label",
+        "metric_key",
+        "metric_label",
+        "metric_raw",
+        "metric_display",
+        "metric_index",
+        "composite_score",
+        "composite_scale_max",
+        "rank",
+        "rank_of",
+        "is_capability_top",
+        "measured_date",
+        "note",
+        "provider_page_url",
+        "run_id",
+        "source_url",
+        "snapshot_sha256"
+      ],
+      "bytes": 101064,
+      "sha256": "6153966709823465ffa1c369b4e48170d793bbece92a25938dd63fec9bd5ceb9"
+    },
+    "benchmarks.jsonl": {
+      "kind": "jsonl",
+      "records": 67,
+      "bytes": 91340,
+      "sha256": "6dad86653cbac5c10d9996aa36cf56c7c8708102cf62a833d9fc620f55b7e2e1"
+    }
+  }
+}

scripts/build_dataset.py

@@ -0,0 +1,606 @@
+#!/usr/bin/env python3
+"""Deterministic generator for the NativePort Web-Access API Benchmarks dataset.
+
+Reads a NativePort ``evals.json`` snapshot and writes three artifacts:
+
+  data/benchmarks.csv     tidy/long form, one row per provider x capability x metric
+  data/benchmarks.jsonl   one record per provider x capability evaluation
+  data/summary.json       computed snapshot facts (every count is derived, never typed)
+
+Design rules
+------------
+* Standard library only.
+* Deterministic: identical input bytes produce identical output bytes. No wall-clock
+  timestamps are recorded; the snapshot is identified by ``latest_run`` plus the
+  SHA-256 of the source file.
+* Numeric values are preserved exactly. The source is parsed with ``parse_float=str``
+  and ``parse_int=str`` so the original token text is kept, and every number is only
+  emitted after verifying that the JSON serialisation of the parsed value is
+  character-for-character identical to that token. A snapshot that cannot satisfy
+  this (for example a token written as ``1.10``) aborts the build instead of
+  silently rounding.
+* Only defensible benchmark and provenance fields are copied. Commercial and routing
+  fields present in the source (pricing, latency prose, marketing summaries,
+  choose_if / avoid_if, gateway routes, auth strings, catalog tier) are excluded.
+
+Usage
+-----
+    python3 scripts/build_dataset.py --input /path/to/evals.json --output-dir data
+"""
+
+from __future__ import annotations
+
+import argparse
+import csv
+import hashlib
+import json
+import sys
+from pathlib import Path
+
+GENERATOR_VERSION = "1.0.0"
+DATASET_NAME = "NativePort Web-Access API Benchmarks"
+DATASET_ID = "nativeport/web-access-api-benchmarks"
+METHODOLOGY_URL = "https://nativeport.ai/methodology/"
+LEADERBOARDS_URL = "https://nativeport.ai/leaderboards/"
+
+# The source states, and the methodology page confirms, that per-capability metrics
+# fold into a composite scored out of 10. It is recorded per row so a consumer never
+# has to assume the scale.
+COMPOSITE_SCALE_MAX = 10
+
+CSV_COLUMNS = [
+    "evaluation_id",
+    "provider_id",
+    "provider_name",
+    "provider_group",
+    "provider_category",
+    "capability_id",
+    "capability_label",
+    "metric_key",
+    "metric_label",
+    "metric_raw",
+    "metric_display",
+    "metric_index",
+    "composite_score",
+    "composite_scale_max",
+    "rank",
+    "rank_of",
+    "is_capability_top",
+    "measured_date",
+    "note",
+    "provider_page_url",
+    "run_id",
+    "source_url",
+    "snapshot_sha256",
+]
+
+
+class BuildError(Exception):
+    """Raised when the snapshot cannot be converted faithfully."""
+
+
+# --------------------------------------------------------------------------- #
+# exact numeric handling
+# --------------------------------------------------------------------------- #
+
+def json_number_text(value):
+    """Return the exact text ``json.dumps`` will emit for this number."""
+    return json.dumps(value)
+
+
+def exact_number(token, path):
+    """Convert a source numeric token to (text, value) without losing precision.
+
+    ``token`` is the untouched text from the snapshot. The returned ``text`` is that
+    same token, used verbatim in the CSV; the returned ``value`` is the parsed number
+    used in JSON output. The build aborts unless the two are provably identical.
+    """
+    if not isinstance(token, str):
+        raise BuildError(f"{path}: expected a numeric token, got {type(token).__name__}")
+    try:
+        if "." in token or "e" in token or "E" in token:
+            value = float(token)
+        else:
+            value = int(token)
+    except ValueError as exc:
+        raise BuildError(f"{path}: not a number: {token!r}") from exc
+    rendered = json_number_text(value)
+    if rendered != token:
+        raise BuildError(
+            f"{path}: cannot round-trip {token!r} exactly (would be written as "
+            f"{rendered!r}). Refusing to emit an altered numeric value."
+        )
+    return token, value
+
+
+def require(condition, message):
+    if not condition:
+        raise BuildError(message)
+
+
+def text_field(container, key, path):
+    value = container.get(key)
+    require(isinstance(value, str), f"{path}.{key}: expected a string")
+    return value
+
+
+def https_url(container, key, path):
+    value = text_field(container, key, path)
+    require(value.startswith("https://"), f"{path}.{key}: expected an https URL, got {value!r}")
+    return value
+
+
+# --------------------------------------------------------------------------- #
+# extraction
+# --------------------------------------------------------------------------- #
+
+def load_snapshot(input_path):
+    raw_bytes = input_path.read_bytes()
+    snapshot_sha256 = hashlib.sha256(raw_bytes).hexdigest()
+    document = json.loads(raw_bytes.decode("utf-8"), parse_float=str, parse_int=str)
+    require(isinstance(document, dict), "snapshot root must be a JSON object")
+    return document, snapshot_sha256, len(raw_bytes)
+
+
+def extract_evaluations(document, snapshot_sha256):
+    """Return the evaluation records in a stable order.
+
+    Ordering key is (capability_id, rank, provider_id): leaderboard order within each
+    capability, which is both meaningful and independent of source dict iteration.
+    """
+    source_url = https_url(document, "source", "$")
+    latest_run = text_field(document, "latest_run", "$")
+    schema_version_text, schema_version = exact_number(
+        document.get("schema_version"), "$.schema_version"
+    )
+    del schema_version_text
+
+    verbs = document.get("verbs")
+    require(isinstance(verbs, dict), "$.verbs must be an object")
+    providers = document.get("providers")
+    require(isinstance(providers, dict), "$.providers must be an object")
+
+    run = document.get("run") if isinstance(document.get("run"), dict) else {}
+    run_id = run.get("name") if isinstance(run.get("name"), str) else latest_run
+
+    evaluations = []
+    seen_keys = set()
+
+    for provider_id in providers:
+        provider = providers[provider_id]
+        path = f"$.providers.{provider_id}"
+        require(isinstance(provider, dict), f"{path}: expected an object")
+        evals = provider.get("evals")
+        if not evals:
+            continue
+        require(isinstance(evals, list), f"{path}.evals: expected a list")
+
+        provider_name = text_field(provider, "name", path)
+        provider_group = text_field(provider, "group", path)
+        provider_category = text_field(provider, "category", path)
+        provider_page_url = https_url(provider, "page", path)
+
+        for position, entry in enumerate(evals):
+            entry_path = f"{path}.evals[{position}]"
+            require(isinstance(entry, dict), f"{entry_path}: expected an object")
+
+            capability_id = text_field(entry, "verb", entry_path)
+            require(
+                capability_id in verbs,
+                f"{entry_path}.verb: {capability_id!r} is absent from $.verbs",
+            )
+            capability = verbs[capability_id]
+            capability_label = text_field(entry, "label", entry_path)
+            require(
+                capability_label == capability.get("label"),
+                f"{entry_path}.label: {capability_label!r} disagrees with "
+                f"$.verbs.{capability_id}.label",
+            )
+            capability_description = text_field(
+                capability, "description", f"$.verbs.{capability_id}"
+            )
+
+            evaluation_id = f"{provider_id}:{capability_id}"
+            require(
+                evaluation_id not in seen_keys,
+                f"{entry_path}: duplicate evaluation key {evaluation_id!r}",
+            )
+            seen_keys.add(evaluation_id)
+
+            composite_text, composite_value = exact_number(
+                entry.get("composite"), f"{entry_path}.composite"
+            )
+            rank_text, rank_value = exact_number(entry.get("rank"), f"{entry_path}.rank")
+            of_text, of_value = exact_number(entry.get("of"), f"{entry_path}.of")
+            require(
+                isinstance(rank_value, int) and isinstance(of_value, int),
+                f"{entry_path}: rank and of must be integers",
+            )
+            require(
+                1 <= rank_value <= of_value,
+                f"{entry_path}: rank {rank_value} outside 1..{of_value}",
+            )
+            require(
+                0 <= composite_value <= COMPOSITE_SCALE_MAX,
+                f"{entry_path}: composite {composite_value} outside 0..{COMPOSITE_SCALE_MAX}",
+            )
+
+            top = entry.get("top")
+            require(isinstance(top, bool), f"{entry_path}.top: expected a boolean")
+            require(
+                top == (rank_value == 1),
+                f"{entry_path}.top: {top} disagrees with rank {rank_value}",
+            )
+
+            measured_date = text_field(entry, "measured", entry_path)
+            note = text_field(entry, "note", entry_path)
+
+            metrics_source = entry.get("metrics")
+            require(
+                isinstance(metrics_source, list) and metrics_source,
+                f"{entry_path}.metrics: expected a non-empty list",
+            )
+            metrics = []
+            metric_keys = set()
+            for metric_index, metric in enumerate(metrics_source):
+                metric_path = f"{entry_path}.metrics[{metric_index}]"
+                require(isinstance(metric, dict), f"{metric_path}: expected an object")
+                metric_key = text_field(metric, "key", metric_path)
+                require(
+                    metric_key not in metric_keys,
+                    f"{metric_path}: duplicate metric key {metric_key!r}",
+                )
+                metric_keys.add(metric_key)
+                raw_text, raw_value = exact_number(metric.get("raw"), f"{metric_path}.raw")
+                metrics.append(
+                    {
+                        "metric_index": metric_index,
+                        "metric_key": metric_key,
+                        "metric_label": text_field(metric, "label", metric_path),
+                        "raw_value": raw_value,
+                        "raw_text": raw_text,
+                        "display_value": text_field(metric, "value", metric_path),
+                    }
+                )
+
+            evaluations.append(
+                {
+                    "evaluation_id": evaluation_id,
+                    "provider_id": provider_id,
+                    "provider_name": provider_name,
+                    "provider_group": provider_group,
+                    "provider_category": provider_category,
+                    "capability_id": capability_id,
+                    "capability_label": capability_label,
+                    "capability_description": capability_description,
+                    "composite_score": composite_value,
+                    "composite_text": composite_text,
+                    "composite_scale_max": COMPOSITE_SCALE_MAX,
+                    "rank": rank_value,
+                    "rank_text": rank_text,
+                    "rank_of": of_value,
+                    "rank_of_text": of_text,
+                    "is_capability_top": top,
+                    "measured_date": measured_date,
+                    "note": note,
+                    "metrics": metrics,
+                    "provider_page_url": provider_page_url,
+                    "run_id": run_id,
+                    "source_url": source_url,
+                    "source_schema_version": schema_version,
+                    "snapshot_sha256": snapshot_sha256,
+                }
+            )
+
+    evaluations.sort(key=lambda row: (row["capability_id"], row["rank"], row["provider_id"]))
+
+    meta = {
+        "source_url": source_url,
+        "latest_run": latest_run,
+        "run_id": run_id,
+        "schema_version": schema_version,
+        "catalog_provider_count": len(providers),
+        "run": run,
+    }
+    return evaluations, meta
+
+
+# --------------------------------------------------------------------------- #
+# writers
+# --------------------------------------------------------------------------- #
+
+def write_csv(evaluations, path):
+    rows = 0
+    with path.open("w", encoding="utf-8", newline="") as handle:
+        writer = csv.writer(handle, lineterminator="\n", quoting=csv.QUOTE_MINIMAL)
+        writer.writerow(CSV_COLUMNS)
+        for evaluation in evaluations:
+            for metric in evaluation["metrics"]:
+                writer.writerow(
+                    [
+                        evaluation["evaluation_id"],
+                        evaluation["provider_id"],
+                        evaluation["provider_name"],
+                        evaluation["provider_group"],
+                        evaluation["provider_category"],
+                        evaluation["capability_id"],
+                        evaluation["capability_label"],
+                        metric["metric_key"],
+                        metric["metric_label"],
+                        metric["raw_text"],
+                        metric["display_value"],
+                        metric["metric_index"],
+                        evaluation["composite_text"],
+                        evaluation["composite_scale_max"],
+                        evaluation["rank_text"],
+                        evaluation["rank_of_text"],
+                        "true" if evaluation["is_capability_top"] else "false",
+                        evaluation["measured_date"],
+                        evaluation["note"],
+                        evaluation["provider_page_url"],
+                        evaluation["run_id"],
+                        evaluation["source_url"],
+                        evaluation["snapshot_sha256"],
+                    ]
+                )
+                rows += 1
+    return rows
+
+
+def jsonl_record(evaluation):
+    """Public shape of one evaluation record. Key order is fixed for determinism."""
+    return {
+        "evaluation_id": evaluation["evaluation_id"],
+        "provider_id": evaluation["provider_id"],
+        "provider_name": evaluation["provider_name"],
+        "provider_group": evaluation["provider_group"],
+        "provider_category": evaluation["provider_category"],
+        "capability_id": evaluation["capability_id"],
+        "capability_label": evaluation["capability_label"],
+        "capability_description": evaluation["capability_description"],
+        "composite_score": evaluation["composite_score"],
+        "composite_scale_max": evaluation["composite_scale_max"],
+        "rank": evaluation["rank"],
+        "rank_of": evaluation["rank_of"],
+        "is_capability_top": evaluation["is_capability_top"],
+        "measured_date": evaluation["measured_date"],
+        "note": evaluation["note"],
+        "metric_count": len(evaluation["metrics"]),
+        "metrics": [
+            {
+                "metric_index": metric["metric_index"],
+                "metric_key": metric["metric_key"],
+                "metric_label": metric["metric_label"],
+                "raw_value": metric["raw_value"],
+                "display_value": metric["display_value"],
+            }
+            for metric in evaluation["metrics"]
+        ],
+        "provider_page_url": evaluation["provider_page_url"],
+        "run_id": evaluation["run_id"],
+        "source_url": evaluation["source_url"],
+        "source_schema_version": evaluation["source_schema_version"],
+        "snapshot_sha256": evaluation["snapshot_sha256"],
+    }
+
+
+def write_jsonl(evaluations, path):
+    with path.open("w", encoding="utf-8", newline="") as handle:
+        for evaluation in evaluations:
+            handle.write(json.dumps(jsonl_record(evaluation), ensure_ascii=False))
+            handle.write("\n")
+    return len(evaluations)
+
+
+def sha256_of(path):
+    return hashlib.sha256(path.read_bytes()).hexdigest()
+
+
+def build_summary(evaluations, meta, csv_path, jsonl_path, csv_rows, jsonl_records, source_bytes):
+    """Every figure here is computed from the extracted rows."""
+    provider_ids = sorted({e["provider_id"] for e in evaluations})
+    capability_ids = sorted({e["capability_id"] for e in evaluations})
+    measured_dates = sorted({e["measured_date"] for e in evaluations})
+    metric_row_count = sum(len(e["metrics"]) for e in evaluations)
+
+    capabilities = []
+    for capability_id in capability_ids:
+        rows = [e for e in evaluations if e["capability_id"] == capability_id]
+        metric_keys = sorted({m["metric_key"] for r in rows for m in r["metrics"]})
+        rank_of_values = sorted({r["rank_of"] for r in rows})
+        capabilities.append(
+            {
+                "capability_id": capability_id,
+                "capability_label": rows[0]["capability_label"],
+                "capability_description": rows[0]["capability_description"],
+                "evaluation_count": len(rows),
+                "rank_of_values": rank_of_values,
+                "rank_of_matches_evaluation_count": rank_of_values == [len(rows)],
+                "provider_ids": sorted(r["provider_id"] for r in rows),
+                "metric_keys": metric_keys,
+                "metric_row_count": sum(len(r["metrics"]) for r in rows),
+                "composite_score_min": min(r["composite_score"] for r in rows),
+                "composite_score_max": max(r["composite_score"] for r in rows),
+            }
+        )
+
+    providers = []
+    for provider_id in provider_ids:
+        rows = [e for e in evaluations if e["provider_id"] == provider_id]
+        providers.append(
+            {
+                "provider_id": provider_id,
+                "provider_name": rows[0]["provider_name"],
+                "provider_group": rows[0]["provider_group"],
+                "evaluation_count": len(rows),
+                "capability_ids": sorted(r["capability_id"] for r in rows),
+            }
+        )
+
+    metric_keys = []
+    for metric_key in sorted({m["metric_key"] for e in evaluations for m in e["metrics"]}):
+        occurrences = [
+            (e, m) for e in evaluations for m in e["metrics"] if m["metric_key"] == metric_key
+        ]
+        metric_keys.append(
+            {
+                "metric_key": metric_key,
+                "metric_labels": sorted({m["metric_label"] for _, m in occurrences}),
+                "occurrence_count": len(occurrences),
+                "capability_ids": sorted({e["capability_id"] for e, _ in occurrences}),
+            }
+        )
+
+    run = meta["run"]
+    source_reported = {
+        "scored_providers": run.get("scored_providers"),
+        "scorecards": run.get("scorecards"),
+        "capabilities": run.get("capabilities"),
+        "catalog_providers": run.get("catalog_providers"),
+    }
+
+    def as_int(value):
+        return int(value) if isinstance(value, str) and value.isdigit() else value
+
+    source_reported = {key: as_int(value) for key, value in source_reported.items()}
+
+    unscored_pairs = run.get("unscored_pairs")
+    unscored_pair_count = len(unscored_pairs) if isinstance(unscored_pairs, list) else 0
+
+    return {
+        "dataset_name": DATASET_NAME,
+        "intended_dataset_id": DATASET_ID,
+        "generator": "scripts/build_dataset.py",
+        "generator_version": GENERATOR_VERSION,
+        "source_url": meta["source_url"],
+        "methodology_url": METHODOLOGY_URL,
+        "leaderboards_url": LEADERBOARDS_URL,
+        "source_schema_version": meta["schema_version"],
+        "source_bytes": source_bytes,
+        "snapshot_sha256": evaluations[0]["snapshot_sha256"] if evaluations else None,
+        "latest_run": meta["latest_run"],
+        "run_id": meta["run_id"],
+        "provider_count_represented": len(provider_ids),
+        "provider_count_in_source_catalog": meta["catalog_provider_count"],
+        "provider_count_in_catalog_without_evaluations": (
+            meta["catalog_provider_count"] - len(provider_ids)
+        ),
+        "evaluation_count": len(evaluations),
+        "capability_count": len(capability_ids),
+        "metric_row_count": metric_row_count,
+        "unscored_pair_count_in_source": unscored_pair_count,
+        "measured_date_min": measured_dates[0] if measured_dates else None,
+        "measured_date_max": measured_dates[-1] if measured_dates else None,
+        "measured_dates": measured_dates,
+        "composite_scale_max": COMPOSITE_SCALE_MAX,
+        "source_reported_run_totals": source_reported,
+        "cross_check": {
+            "scored_providers_matches": source_reported.get("scored_providers")
+            == len(provider_ids),
+            "scorecards_matches": source_reported.get("scorecards") == len(evaluations),
+            "capabilities_matches": source_reported.get("capabilities") == len(capability_ids),
+            "catalog_providers_matches": source_reported.get("catalog_providers")
+            == meta["catalog_provider_count"],
+        },
+        "capabilities": capabilities,
+        "providers": providers,
+        "metric_keys": metric_keys,
+        "outputs": {
+            csv_path.name: {
+                "kind": "csv",
+                "data_rows": csv_rows,
+                "columns": len(CSV_COLUMNS),
+                "column_names": list(CSV_COLUMNS),
+                "bytes": csv_path.stat().st_size,
+                "sha256": sha256_of(csv_path),
+            },
+            jsonl_path.name: {
+                "kind": "jsonl",
+                "records": jsonl_records,
+                "bytes": jsonl_path.stat().st_size,
+                "sha256": sha256_of(jsonl_path),
+            },
+        },
+    }
+
+
+def write_summary(summary, path):
+    with path.open("w", encoding="utf-8", newline="") as handle:
+        json.dump(summary, handle, ensure_ascii=False, indent=2, sort_keys=False)
+        handle.write("\n")
+
+
+# --------------------------------------------------------------------------- #
+# entry point
+# --------------------------------------------------------------------------- #
+
+def build(input_path, output_dir):
+    input_path = Path(input_path)
+    output_dir = Path(output_dir)
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    document, snapshot_sha256, source_bytes = load_snapshot(input_path)
+    evaluations, meta = extract_evaluations(document, snapshot_sha256)
+    require(evaluations, "no evaluations found in the snapshot")
+
+    csv_path = output_dir / "benchmarks.csv"
+    jsonl_path = output_dir / "benchmarks.jsonl"
+    summary_path = output_dir / "summary.json"
+
+    csv_rows = write_csv(evaluations, csv_path)
+    jsonl_records = write_jsonl(evaluations, jsonl_path)
+    summary = build_summary(
+        evaluations, meta, csv_path, jsonl_path, csv_rows, jsonl_records, source_bytes
+    )
+    write_summary(summary, summary_path)
+    return summary
+
+
+def default_output_dir():
+    return Path(__file__).resolve().parent.parent / "data"
+
+
+def main(argv=None):
+    parser = argparse.ArgumentParser(
+        description="Build the NativePort Web-Access API Benchmarks dataset files."
+    )
+    parser.add_argument(
+        "--input",
+        required=True,
+        help="Path to a NativePort evals.json snapshot.",
+    )
+    parser.add_argument(
+        "--output-dir",
+        default=str(default_output_dir()),
+        help="Directory to write benchmarks.csv, benchmarks.jsonl and summary.json into.",
+    )
+    args = parser.parse_args(argv)
+
+    try:
+        summary = build(args.input, args.output_dir)
+    except BuildError as exc:
+        print(f"build failed: {exc}", file=sys.stderr)
+        return 1
+
+    outputs = summary["outputs"]
+    print(f"source          {summary['source_url']}")
+    print(f"snapshot sha256 {summary['snapshot_sha256']}")
+    print(f"latest run      {summary['latest_run']}")
+    print(
+        "counts          providers={} capabilities={} evaluations={} metric_rows={}".format(
+            summary["provider_count_represented"],
+            summary["capability_count"],
+            summary["evaluation_count"],
+            summary["metric_row_count"],
+        )
+    )
+    for name in sorted(outputs):
+        entry = outputs[name]
+        size = entry.get("data_rows", entry.get("records"))
+        print(f"wrote           {name} ({size} rows, {entry['bytes']} bytes) {entry['sha256']}")
+    print("wrote           summary.json")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

tests/test_dataset.py

@@ -0,0 +1,931 @@
+#!/usr/bin/env python3
+"""Tests for the NativePort Web-Access API Benchmarks dataset package.
+
+Standard library only:
+
+    python3 -m unittest discover -s tests -v
+
+Tests that exercise the generator against the real snapshot look for it at
+/tmp/nativeport-evals-current.json, overridable with the NP_EVALS_JSON environment
+variable. Determinism and exact-number behaviour are additionally proven against an
+embedded fixture, so those properties are covered even without the snapshot at hand.
+"""
+
+from __future__ import annotations
+
+import csv
+import hashlib
+import importlib.util
+import json
+import os
+import re
+import tempfile
+import unicodedata
+import unittest
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parent.parent
+DATA = ROOT / "data"
+CSV_PATH = DATA / "benchmarks.csv"
+JSONL_PATH = DATA / "benchmarks.jsonl"
+SUMMARY_PATH = DATA / "summary.json"
+README_PATH = ROOT / "README.md"
+LICENSE_PATH = ROOT / "LICENSE"
+MAKEFILE_PATH = ROOT / "Makefile"
+
+SOURCE_PATH = Path(os.environ.get("NP_EVALS_JSON", "/tmp/nativeport-evals-current.json"))
+
+BACKLINK = (
+    "https://nativeport.ai/methodology/"
+    "?utm_source=huggingface&utm_medium=referral"
+    "&utm_campaign=backlink_hf_web_access_benchmarks_20260811"
+)
+
+# The licence approved for publication. These three values must agree with each other
+# and with the LICENSE file; nothing in the package may assert a different licence.
+LICENSE_HUB_ID = "cc-by-4.0"          # Hugging Face front-matter identifier
+LICENSE_SPDX_ID = "CC-BY-4.0"         # SPDX identifier
+LICENSE_URL = "https://creativecommons.org/licenses/by/4.0/"
+LICENSE_LEGALCODE_URL = "https://creativecommons.org/licenses/by/4.0/legalcode"
+
+# SHA-256 of the CC BY 4.0 legal code reduced to its alphanumeric words (see
+# normalise_legal_text). Pinning the normalised form makes the assertion insensitive to
+# line wrapping, indentation and quote style while still detecting any change of
+# wording, however small. The digest was taken from the legal code served by
+# https://creativecommons.org/licenses/by/4.0/legalcode.en and independently
+# cross-checked against the SPDX license-list text for CC-BY-4.0; the two agree word
+# for word apart from list enumerators the CC page renders with CSS counters.
+CC_BY_4_0_NORMALISED_SHA256 = (
+    "b94bf421a55fe7bc2d698d314ee46b5e22bfcc7afbf93301978a35fd8bddd9ba"
+)
+
+# Files that ship to the public Hugging Face repository.
+PUBLIC_UPLOAD_SET = (
+    "README.md",
+    "LICENSE",
+    "data/benchmarks.csv",
+    "data/benchmarks.jsonl",
+    "data/summary.json",
+    "scripts/build_dataset.py",
+    "tests/test_dataset.py",
+    "Makefile",
+)
+
+# Review and tooling documents that stay internal. They may be absent — this suite
+# ships with the public set and must pass without them.
+INTERNAL_ONLY = ("APPROVAL.md", "VALIDATION.md", "scripts/validate.py")
+
+
+def normalise_legal_text(text):
+    """Reduce legal text to lowercase alphanumeric words separated by single spaces."""
+    return " ".join(re.findall(r"[a-z0-9]+", unicodedata.normalize("NFKC", text).lower()))
+
+
+def load_builder():
+    spec = importlib.util.spec_from_file_location(
+        "build_dataset", ROOT / "scripts" / "build_dataset.py"
+    )
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+build_dataset = load_builder()
+
+
+def read_csv_rows():
+    with CSV_PATH.open(encoding="utf-8", newline="") as handle:
+        return list(csv.DictReader(handle))
+
+
+def read_jsonl_records():
+    with JSONL_PATH.open(encoding="utf-8") as handle:
+        return [json.loads(line) for line in handle]
+
+
+def read_summary():
+    return json.loads(SUMMARY_PATH.read_text(encoding="utf-8"))
+
+
+def sha256_of(path):
+    return hashlib.sha256(path.read_bytes()).hexdigest()
+
+
+# A miniature snapshot exercising the shapes that matter: a trailing-zero float, an
+# integer-valued float, a small decimal, a zero, an empty note and a unicode label.
+FIXTURE = {
+    "schema_version": 1,
+    "source": "https://nativeport.ai/evals.json",
+    "latest_run": "2026-08-05",
+    "verbs": {
+        "search": {"label": "Search", "description": "A query in, ranked links out"},
+        "extract_ai": {"label": "Extract · AI/schema", "description": "Fields from a page"},
+    },
+    "providers": {
+        "beta": {
+            "name": "Beta",
+            "group": "Search",
+            "category": "Search API",
+            "page": "https://nativeport.ai/providers/beta/",
+            "evals": [
+                {
+                    "verb": "search",
+                    "label": "Search",
+                    "composite": 7.5,
+                    "rank": 2,
+                    "of": 2,
+                    "top": False,
+                    "measured": "2026-08-05",
+                    "note": "",
+                    "metrics": [
+                        {"key": "recall_at_10", "label": "Recall@10", "raw": 0.588, "value": "0.59"},
+                        {"key": "latency_p50_ms", "label": "Latency p50", "raw": 886.0, "value": "886 ms"},
+                        {"key": "error_rate_pct", "label": "Errors", "raw": 0, "value": "0%"},
+                    ],
+                }
+            ],
+        },
+        "alpha": {
+            "name": "Alpha",
+            "group": "Search",
+            "category": "Search API",
+            "page": "https://nativeport.ai/providers/alpha/",
+            "evals": [
+                {
+                    "verb": "search",
+                    "label": "Search",
+                    "composite": 9.13,
+                    "rank": 1,
+                    "of": 2,
+                    "top": True,
+                    "measured": "2026-08-04",
+                    "note": "Comma, and \"quotes\" in the note.",
+                    "metrics": [
+                        {"key": "recall_at_10", "label": "Recall@10", "raw": 0.62, "value": "0.62"},
+                        {"key": "latency_p50_ms", "label": "Latency p50", "raw": 815.5, "value": "816 ms"},
+                        {
+                            "key": "cost_per_useful_usd",
+                            "label": "Cost",
+                            "raw": 0.000484,
+                            "value": "$0.00048 / useful result",
+                        },
+                        {"key": "error_rate_pct", "label": "Errors", "raw": 0.0, "value": "0%"},
+                    ],
+                },
+                {
+                    "verb": "extract_ai",
+                    "label": "Extract · AI/schema",
+                    "composite": 0.0,
+                    "rank": 1,
+                    "of": 1,
+                    "top": True,
+                    "measured": "2026-08-05",
+                    "note": "Every extraction failed.",
+                    "metrics": [
+                        {"key": "field_accuracy", "label": "Field accuracy", "raw": 0.0, "value": "0.00"},
+                        {"key": "latency_p50_ms", "label": "Latency p50", "raw": 0.0, "value": "0 ms"},
+                        {"key": "error_rate_pct", "label": "Errors", "raw": 100.0, "value": "100%"},
+                    ],
+                },
+            ],
+        },
+        "unscored": {
+            "name": "Unscored",
+            "group": "Search",
+            "category": "Search API",
+            "page": "https://nativeport.ai/providers/unscored/",
+            "evals": [],
+        },
+    },
+    "run": {
+        "name": "2026-08-05",
+        "scored_providers": 2,
+        "catalog_providers": 3,
+        "scorecards": 3,
+        "capabilities": 2,
+        "unscored_pairs": [],
+    },
+}
+
+
+def write_fixture(directory, document=None):
+    path = Path(directory) / "fixture-evals.json"
+    path.write_text(
+        json.dumps(document if document is not None else FIXTURE, ensure_ascii=False),
+        encoding="utf-8",
+    )
+    return path
+
+
+class GeneratorDeterminismTest(unittest.TestCase):
+    """The generator must be a pure function of its input bytes."""
+
+    def test_two_builds_from_the_fixture_are_byte_identical(self):
+        with tempfile.TemporaryDirectory() as workspace:
+            source = write_fixture(workspace)
+            digests = []
+            for run in ("first", "second"):
+                out = Path(workspace) / run
+                build_dataset.build(source, out)
+                digests.append(
+                    tuple(
+                        sha256_of(out / name)
+                        for name in ("benchmarks.csv", "benchmarks.jsonl", "summary.json")
+                    )
+                )
+            self.assertEqual(digests[0], digests[1])
+
+    def test_outputs_carry_no_wall_clock_timestamp(self):
+        with tempfile.TemporaryDirectory() as workspace:
+            source = write_fixture(workspace)
+            out = Path(workspace) / "out"
+            summary = build_dataset.build(source, out)
+            blob = json.dumps(summary)
+            for field in ("generated_at", "timestamp", "build_time", "created_at"):
+                self.assertNotIn(field, blob)
+
+    def test_ordering_is_stable_and_independent_of_source_order(self):
+        """Records sort by (capability, rank, provider), not by dict insertion order."""
+
+        def records_without_snapshot_hash(path):
+            # Reordering the source changes its bytes, so the snapshot hash legitimately
+            # differs; everything else must be identical.
+            out = []
+            for line in path.read_text(encoding="utf-8").splitlines():
+                record = json.loads(line)
+                record.pop("snapshot_sha256")
+                out.append(record)
+            return out
+
+        with tempfile.TemporaryDirectory() as workspace:
+            out = Path(workspace) / "out"
+            build_dataset.build(write_fixture(workspace), out)
+            first = records_without_snapshot_hash(out / "benchmarks.jsonl")
+
+            shuffled = json.loads(json.dumps(FIXTURE))
+            shuffled["providers"] = {
+                key: shuffled["providers"][key] for key in ("unscored", "alpha", "beta")
+            }
+            other = Path(workspace) / "shuffled"
+            build_dataset.build(write_fixture(other.parent, shuffled), other)
+            second = records_without_snapshot_hash(other / "benchmarks.jsonl")
+
+            self.assertEqual(first, second)
+            self.assertEqual(
+                [record["evaluation_id"] for record in first],
+                ["alpha:extract_ai", "alpha:search", "beta:search"],
+            )
+
+    @unittest.skipUnless(SOURCE_PATH.exists(), f"snapshot not available at {SOURCE_PATH}")
+    def test_committed_artifacts_match_a_fresh_rebuild(self):
+        with tempfile.TemporaryDirectory() as workspace:
+            out = Path(workspace) / "out"
+            build_dataset.build(SOURCE_PATH, out)
+            for name in ("benchmarks.csv", "benchmarks.jsonl", "summary.json"):
+                self.assertEqual(
+                    sha256_of(out / name),
+                    sha256_of(DATA / name),
+                    f"{name} differs from a rebuild of the recorded snapshot",
+                )
+
+
+class ExactNumericPreservationTest(unittest.TestCase):
+    """Measured values must survive the pipeline character for character."""
+
+    def test_refuses_a_token_it_cannot_reproduce_exactly(self):
+        document = json.loads(json.dumps(FIXTURE))
+        document["providers"]["alpha"]["evals"][0]["metrics"][0]["raw"] = 1.10
+        with tempfile.TemporaryDirectory() as workspace:
+            # 1.10 must reach the parser as text, so patch the serialised form directly.
+            path = Path(workspace) / "fixture-evals.json"
+            blob = json.dumps(document, ensure_ascii=False).replace('"raw": 1.1,', '"raw": 1.10,', 1)
+            self.assertIn('"raw": 1.10,', blob)
+            path.write_text(blob, encoding="utf-8")
+            with self.assertRaises(build_dataset.BuildError) as caught:
+                build_dataset.build(path, Path(workspace) / "out")
+            self.assertIn("round-trip", str(caught.exception))
+
+    def test_exact_number_keeps_the_source_token(self):
+        for token in ("0.938", "886.0", "0.0003", "0.000484", "0", "100.0", "9.13"):
+            text, value = build_dataset.exact_number(token, "$.test")
+            self.assertEqual(text, token)
+            self.assertEqual(json.dumps(value), token)
+
+    @unittest.skipUnless(SOURCE_PATH.exists(), f"snapshot not available at {SOURCE_PATH}")
+    def test_every_published_number_equals_its_source_token(self):
+        document = json.loads(SOURCE_PATH.read_text(encoding="utf-8"), parse_float=str, parse_int=str)
+        expected = {}
+        for provider_id, provider in document["providers"].items():
+            for entry in provider.get("evals", []):
+                key = f"{provider_id}:{entry['verb']}"
+                expected[key] = {
+                    "composite": entry["composite"],
+                    "rank": entry["rank"],
+                    "of": entry["of"],
+                    "metrics": {m["key"]: m["raw"] for m in entry["metrics"]},
+                }
+        self.assertTrue(expected)
+
+        seen = set()
+        for row in read_csv_rows():
+            reference = expected[row["evaluation_id"]]
+            self.assertEqual(row["composite_score"], reference["composite"])
+            self.assertEqual(row["rank"], reference["rank"])
+            self.assertEqual(row["rank_of"], reference["of"])
+            self.assertEqual(row["metric_raw"], reference["metrics"][row["metric_key"]])
+            seen.add((row["evaluation_id"], row["metric_key"]))
+        self.assertEqual(
+            len(seen), sum(len(value["metrics"]) for value in expected.values())
+        )
+
+        for record in read_jsonl_records():
+            reference = expected[record["evaluation_id"]]
+            self.assertEqual(json.dumps(record["composite_score"]), reference["composite"])
+            self.assertEqual(json.dumps(record["rank"]), reference["rank"])
+            self.assertEqual(json.dumps(record["rank_of"]), reference["of"])
+            for metric in record["metrics"]:
+                self.assertEqual(
+                    json.dumps(metric["raw_value"]),
+                    reference["metrics"][metric["metric_key"]],
+                    f"{record['evaluation_id']}/{metric['metric_key']}",
+                )
+
+    def test_display_values_are_not_substituted_for_measurements(self):
+        for row in read_csv_rows():
+            self.assertNotEqual(row["metric_raw"], "")
+            float(row["metric_raw"])  # parses as a number, unlike "816 ms" or "$0.0003 / call"
+
+
+class SchemaAndCountConsistencyTest(unittest.TestCase):
+    def setUp(self):
+        self.summary = read_summary()
+        self.rows = read_csv_rows()
+        self.records = read_jsonl_records()
+
+    def test_csv_header_matches_the_generator_contract(self):
+        with CSV_PATH.open(encoding="utf-8", newline="") as handle:
+            header = next(csv.reader(handle))
+        self.assertEqual(header, build_dataset.CSV_COLUMNS)
+
+    def test_row_counts_agree_across_artifacts(self):
+        self.assertEqual(len(self.rows), self.summary["metric_row_count"])
+        self.assertEqual(len(self.records), self.summary["evaluation_count"])
+        self.assertEqual(
+            sum(len(record["metrics"]) for record in self.records),
+            self.summary["metric_row_count"],
+        )
+        self.assertEqual(
+            len({row["evaluation_id"] for row in self.rows}),
+            self.summary["evaluation_count"],
+        )
+
+    def test_counts_are_computed_not_asserted(self):
+        summary = self.summary
+        self.assertEqual(
+            summary["provider_count_represented"],
+            len({row["provider_id"] for row in self.rows}),
+        )
+        self.assertEqual(
+            summary["capability_count"], len({row["capability_id"] for row in self.rows})
+        )
+        self.assertEqual(
+            summary["provider_count_in_catalog_without_evaluations"],
+            summary["provider_count_in_source_catalog"] - summary["provider_count_represented"],
+        )
+        self.assertTrue(all(summary["cross_check"].values()), summary["cross_check"])
+
+    def test_coverage_is_not_overstated(self):
+        """The catalog is larger than the scored set; the dataset must say so."""
+        summary = self.summary
+        self.assertLess(
+            summary["provider_count_represented"], summary["provider_count_in_source_catalog"]
+        )
+        self.assertGreater(summary["provider_count_in_catalog_without_evaluations"], 0)
+
+    def test_measured_date_range_is_derived_from_the_rows(self):
+        dates = sorted({row["measured_date"] for row in self.rows})
+        self.assertEqual(self.summary["measured_date_min"], dates[0])
+        self.assertEqual(self.summary["measured_date_max"], dates[-1])
+        self.assertEqual(self.summary["measured_dates"], dates)
+        for value in dates:
+            self.assertRegex(value, r"^\d{4}-\d{2}-\d{2}$")
+
+    def test_summary_records_matching_output_hashes(self):
+        for name, entry in self.summary["outputs"].items():
+            self.assertEqual(entry["sha256"], sha256_of(DATA / name), name)
+            self.assertEqual(entry["bytes"], (DATA / name).stat().st_size, name)
+
+    def test_every_csv_row_maps_onto_a_jsonl_record(self):
+        nested = {
+            record["evaluation_id"]: {m["metric_key"]: m for m in record["metrics"]}
+            for record in self.records
+        }
+        for row in self.rows:
+            metric = nested[row["evaluation_id"]][row["metric_key"]]
+            self.assertEqual(row["metric_label"], metric["metric_label"])
+            self.assertEqual(row["metric_display"], metric["display_value"])
+            self.assertEqual(int(row["metric_index"]), metric["metric_index"])
+
+    def test_capability_metric_keys_are_documented_in_the_summary(self):
+        for capability in self.summary["capabilities"]:
+            observed = {
+                row["metric_key"]
+                for row in self.rows
+                if row["capability_id"] == capability["capability_id"]
+            }
+            self.assertEqual(observed, set(capability["metric_keys"]))
+            self.assertEqual(
+                capability["evaluation_count"],
+                len(
+                    {
+                        row["evaluation_id"]
+                        for row in self.rows
+                        if row["capability_id"] == capability["capability_id"]
+                    }
+                ),
+            )
+
+    def test_excluded_commercial_fields_did_not_leak(self):
+        blob = CSV_PATH.read_text(encoding="utf-8") + JSONL_PATH.read_text(encoding="utf-8")
+        blob += SUMMARY_PATH.read_text(encoding="utf-8")
+        for field in (
+            "pricing_entry",
+            "choose_if",
+            "avoid_if",
+            "latency_note",
+            "call_body",
+            "call_method",
+            "Authorization",
+            "sign_up",
+        ):
+            self.assertNotIn(field, blob, f"{field} leaked into the published data")
+
+
+class RankingIntegrityTest(unittest.TestCase):
+    def setUp(self):
+        self.records = read_jsonl_records()
+
+    def test_rank_is_within_bounds(self):
+        for record in self.records:
+            self.assertGreaterEqual(record["rank"], 1, record["evaluation_id"])
+            self.assertLessEqual(record["rank"], record["rank_of"], record["evaluation_id"])
+
+    def test_top_flag_agrees_with_rank(self):
+        for record in self.records:
+            self.assertEqual(
+                record["is_capability_top"], record["rank"] == 1, record["evaluation_id"]
+            )
+
+    def test_composite_is_inside_the_declared_scale(self):
+        for record in self.records:
+            self.assertEqual(record["composite_scale_max"], 10)
+            self.assertGreaterEqual(record["composite_score"], 0, record["evaluation_id"])
+            self.assertLessEqual(record["composite_score"], 10, record["evaluation_id"])
+
+    def test_ranks_are_unique_and_complete_within_each_capability(self):
+        by_capability = {}
+        for record in self.records:
+            by_capability.setdefault(record["capability_id"], []).append(record)
+        for capability_id, group in by_capability.items():
+            ranks = sorted(record["rank"] for record in group)
+            self.assertEqual(ranks, list(range(1, len(group) + 1)), capability_id)
+            self.assertEqual(
+                {record["rank_of"] for record in group}, {len(group)}, capability_id
+            )
+
+    def test_no_duplicate_evaluation_keys(self):
+        keys = [record["evaluation_id"] for record in self.records]
+        self.assertEqual(len(keys), len(set(keys)))
+        pairs = [(record["provider_id"], record["capability_id"]) for record in self.records]
+        self.assertEqual(len(pairs), len(set(pairs)))
+        for record in self.records:
+            self.assertEqual(
+                record["evaluation_id"],
+                f"{record['provider_id']}:{record['capability_id']}",
+            )
+
+    def test_metric_keys_are_unique_within_an_evaluation(self):
+        for record in self.records:
+            keys = [metric["metric_key"] for metric in record["metrics"]]
+            self.assertEqual(len(keys), len(set(keys)), record["evaluation_id"])
+            self.assertEqual(
+                [metric["metric_index"] for metric in record["metrics"]],
+                list(range(len(keys))),
+                record["evaluation_id"],
+            )
+
+
+class UrlPolicyTest(unittest.TestCase):
+    URL_FIELDS = ("provider_page_url", "source_url")
+
+    def test_data_urls_are_https(self):
+        for row in read_csv_rows():
+            for field in self.URL_FIELDS:
+                self.assertTrue(row[field].startswith("https://"), f"{field}={row[field]}")
+        for record in read_jsonl_records():
+            for field in self.URL_FIELDS:
+                self.assertTrue(record[field].startswith("https://"), f"{field}={record[field]}")
+
+    def test_summary_urls_are_https(self):
+        summary = read_summary()
+        for field in ("source_url", "methodology_url", "leaderboards_url"):
+            self.assertTrue(summary[field].startswith("https://"), field)
+
+    def test_no_plain_http_anywhere_in_the_package(self):
+        offenders = []
+        for path in sorted(ROOT.rglob("*")):
+            if not path.is_file() or "__pycache__" in path.parts:
+                continue
+            try:
+                text = path.read_text(encoding="utf-8")
+            except (UnicodeDecodeError, OSError):
+                continue
+            # self-avoiding pattern: matches "http://" but not this line
+            offenders.extend(
+                f"{path.relative_to(ROOT)}: {hit}"
+                for hit in re.findall(r"htt[p]://[^\s\"'<>)\]]+", text)
+            )
+        self.assertEqual(offenders, [])
+
+    def test_data_files_carry_no_campaign_tagged_urls(self):
+        for path in (CSV_PATH, JSONL_PATH, SUMMARY_PATH):
+            self.assertNotIn("utm_", path.read_text(encoding="utf-8"), path.name)
+
+
+class ReadmeCardTest(unittest.TestCase):
+    def setUp(self):
+        self.text = README_PATH.read_text(encoding="utf-8")
+
+    def front_matter(self):
+        return self.text[4 : self.text.find("\n---\n", 4)]
+
+    def section(self, heading):
+        """Body of one `## ` section, up to the next one."""
+        start = self.text.index(heading) + len(heading)
+        rest = self.text[start:]
+        end = rest.find("\n## ")
+        return rest if end == -1 else rest[:end]
+
+    def flat(self, heading):
+        """Section body with all runs of whitespace collapsed, so prose assertions do
+        not depend on where the source happens to wrap."""
+        return " ".join(self.section(heading).split())
+
+    def test_front_matter_is_well_formed(self):
+        self.assertTrue(self.text.startswith("---\n"))
+        end = self.text.find("\n---\n", 4)
+        self.assertNotEqual(end, -1, "front matter is not closed")
+        front = self.text[4:end]
+        self.assertNotIn("\t", front)
+        top_keys = [
+            line.split(":", 1)[0]
+            for line in front.splitlines()
+            if line and not line.startswith((" ", "-", "#"))
+        ]
+        self.assertEqual(
+            top_keys,
+            ["license", "pretty_name", "language", "tags", "size_categories", "configs"],
+        )
+        for line in front.splitlines():
+            if line and not line.startswith((" ", "-")):
+                self.assertIn(":", line, line)
+
+    def test_title_and_pretty_name_match(self):
+        self.assertIn("pretty_name: NativePort Web-Access API Benchmarks", self.text)
+        self.assertIn("\n# NativePort Web-Access API Benchmarks\n", self.text)
+
+    def test_front_matter_declares_the_approved_licence(self):
+        """Exactly one licence key, carrying exactly the approved identifier."""
+        front = self.front_matter()
+        license_lines = [line for line in front.splitlines() if line.startswith("license")]
+        self.assertEqual(license_lines, [f"license: {LICENSE_HUB_ID}"])
+        # license_name / license_link belong to `license: other`; they must not appear.
+        for key in ("license_name", "license_link", "license_details"):
+            self.assertNotIn(key, front, key)
+        for competing in (
+            "cc-by-nc",
+            "cc-by-sa",
+            "cc-by-nd",
+            "cc0-1.0",
+            "odc-by",
+            "odbl",
+            "apache-2.0",
+            "unlicense",
+            "all-rights-reserved",
+        ):
+            self.assertNotIn(competing, front, competing)
+
+    def test_licensing_section_names_cc_by_4_0_and_the_canonical_urls(self):
+        section = self.flat("## Licensing")
+        self.assertIn("Creative Commons Attribution 4.0 International", section)
+        self.assertIn("CC BY 4.0", section)
+        self.assertIn(LICENSE_URL, section)
+        self.assertIn(LICENSE_LEGALCODE_URL, section)
+        self.assertIn(LICENSE_SPDX_ID, section)
+        self.assertIn("[`LICENSE`](LICENSE)", section)
+        self.assertIn(f"license: {LICENSE_HUB_ID}", section)
+
+    def test_licensing_section_no_longer_defers_the_decision(self):
+        section = self.flat("## Licensing")
+        for stale in (
+            "not been finalised",
+            "not finalised",
+            "all-rights-reserved",
+            "no open-data licence is asserted",
+            "seek permission before redistribution",
+            "Until a licence is published",
+        ):
+            self.assertNotIn(stale, section, stale)
+
+    def test_licensing_section_scopes_the_grant_to_what_nativeport_can_license(self):
+        section = self.flat("## Licensing")
+        for phrase in (
+            "NativePort licenses what it is in a position to license",
+            "this dataset as a compilation",
+            "the benchmark measurements NativePort itself produced",
+            "The grant extends only to those rights and only to the extent NativePort holds them.",
+        ):
+            self.assertIn(phrase, section, phrase)
+
+    def test_licensing_section_does_not_overstate_trademark_rights(self):
+        """The licence covers NativePort's compilation, never third-party marks."""
+        section = self.flat("## Licensing")
+        lowered = section.lower()
+        for overstatement in (
+            "trademarks are licensed",
+            "trademark rights are licensed",
+            "licenses the trademarks",
+            "licences the trademarks",
+            "grants you the trademarks",
+            "grant of trademark",
+            "including any trademarks",
+            "trademark licence is granted",
+            "trademark license is granted",
+        ):
+            self.assertNotIn(overstatement, lowered, overstatement)
+
+        # Every sentence that touches marks must withhold rather than grant: it carries
+        # a negation, or it attributes the marks to their owners.
+        denials = ("not ", "no ", "nor ", "never ", "remain the property", "outside what")
+        sentences = [
+            sentence
+            for sentence in re.split(r"(?<=[.;])\s+", lowered)
+            if "trademark" in sentence or " marks" in sentence
+        ]
+        self.assertTrue(sentences, "the licensing section must address trademarks")
+        for sentence in sentences:
+            self.assertTrue(
+                any(denial in sentence for denial in denials),
+                f"trademark sentence does not withhold a grant: {sentence}",
+            )
+
+    def test_licensing_section_states_attribution_and_warranty_terms(self):
+        section = self.flat("## Licensing")
+        for phrase in (
+            "Credit *NativePort*",
+            "state that the material is under CC BY 4.0 with a link to the licence",
+            "indicate whether you modified it",
+            "as-is and as-available, without warranties or conditions of any kind",
+        ):
+            self.assertIn(phrase, section, phrase)
+
+    def test_trademark_notice_denies_any_licence_grant_in_marks(self):
+        section = self.flat("## Trademark notice")
+        self.assertIn(
+            "The CC BY 4.0 licence described under [Licensing](#licensing) grants "
+            "**no** rights in any of these marks",
+            section,
+        )
+
+    def test_declared_data_files_exist(self):
+        front = self.front_matter()
+        for relative in re.findall(r"path:\s*(\S+)", front):
+            self.assertTrue((ROOT / relative).is_file(), relative)
+
+    def test_backlink_is_exact_and_appears_once(self):
+        self.assertEqual(self.text.count(BACKLINK), 1)
+        self.assertEqual(len(re.findall(r"utm_source=", self.text)), 1)
+        self.assertEqual(len(re.findall(r"utm_campaign=", self.text)), 1)
+        self.assertIn(
+            f"[How we measure]({BACKLINK})",
+            self.text.replace("\n", " ").replace("  ", " "),
+        )
+
+    def test_backlink_is_contextual_not_the_point_of_the_page(self):
+        """The card must stand on its own: the link sits inside the methodology prose."""
+        index = self.text.index(BACKLINK)
+        self.assertGreater(index, len(self.text) // 3, "backlink appears too early to be contextual")
+        self.assertIn("## Methodology", self.text[:index])
+
+    def test_required_sections_exist(self):
+        for heading in (
+            "## Disclosure",
+            "## Dataset structure",
+            "## Usage",
+            "## Methodology",
+            "## Provenance and reproducibility",
+            "## Limitations and scope",
+            "## Update policy",
+            "## Licensing",
+            "## Trademark notice",
+            "## Citation",
+        ):
+            self.assertIn(heading, self.text, heading)
+
+    def test_first_party_disclosure_is_explicit(self):
+        for phrase in (
+            "produced by **NativePort**",
+            "own first-party benchmark",
+            "not an independent third-party evaluation",
+        ):
+            self.assertIn(phrase, self.text, phrase)
+
+    def test_limitations_cover_the_known_risks(self):
+        section = self.text[self.text.index("## Limitations and scope") :]
+        section = section[: section.index("## Update policy")]
+        for phrase in (
+            "Coverage is partial",
+            "Composites are capability-local",
+            "Single point in time",
+            "First-party measurement",
+            "uptime",
+        ):
+            self.assertIn(phrase, section, phrase)
+
+    def test_card_does_not_claim_the_whole_catalog_is_benchmarked(self):
+        summary = read_summary()
+        represented = summary["provider_count_represented"]
+        catalog = summary["provider_count_in_source_catalog"]
+        self.assertIn(f"**{represented} commercial web-access", self.text)
+        self.assertIn(f"{represented} of the {catalog} providers in the source catalog", self.text)
+
+    def test_headline_counts_match_the_generated_summary(self):
+        summary = read_summary()
+        self.assertIn(f"**{summary['evaluation_count']} provider × capability", self.text)
+        self.assertIn(f"**{summary['capability_count']} capabilities**", self.text)
+        self.assertIn(f"**{summary['metric_row_count']} metric rows**", self.text)
+        self.assertIn(summary["snapshot_sha256"], self.text)
+        self.assertIn(summary["latest_run"], self.text)
+
+    def test_capability_comparison_caveat_is_present(self):
+        self.assertIn("Latency and cost are not comparable across capabilities", self.text)
+        self.assertIn("averaging a provider's composites across capabilities", self.text)
+
+    def test_trademark_notice_disclaims_endorsement(self):
+        section = self.text[self.text.index("## Trademark notice") :]
+        for phrase in ("trademarks of their respective owners", "does not\nimply any affiliation"):
+            self.assertIn(phrase, section, phrase)
+
+
+class LicenseFileTest(unittest.TestCase):
+    """LICENSE must be the complete, unmodified CC BY 4.0 legal code."""
+
+    def setUp(self):
+        self.assertTrue(LICENSE_PATH.is_file(), "LICENSE is missing from the package")
+        self.raw = LICENSE_PATH.read_bytes()
+        self.text = self.raw.decode("utf-8")
+        self.normalised = normalise_legal_text(self.text)
+
+    def test_license_file_is_plain_unix_utf8_text(self):
+        self.assertNotIn(b"\r", self.raw, "LICENSE must use unix line endings")
+        self.assertTrue(self.raw.endswith(b"\n"), "LICENSE must end with a newline")
+        self.assertNotIn("\t", self.text, "LICENSE must not contain tab characters")
+        self.assertGreater(
+            len(self.raw), 16000, "LICENSE is too short to be the full legal code"
+        )
+
+    def test_license_text_is_the_canonical_cc_by_4_0_legal_code(self):
+        digest = hashlib.sha256(self.normalised.encode("utf-8")).hexdigest()
+        self.assertEqual(
+            digest,
+            CC_BY_4_0_NORMALISED_SHA256,
+            "LICENSE wording differs from the canonical CC BY 4.0 legal code",
+        )
+
+    def test_license_carries_every_section_of_the_legal_code(self):
+        self.assertIn(
+            "creative commons attribution 4 0 international public license", self.normalised
+        )
+        for number, title in (
+            (1, "definitions"),
+            (2, "scope"),
+            (3, "license conditions"),
+            (4, "sui generis database rights"),
+            (5, "disclaimer of warranties and limitation of liability"),
+            (6, "term and termination"),
+            (7, "other terms and conditions"),
+            (8, "interpretation"),
+        ):
+            self.assertIn(f"section {number} {title}", self.normalised, title)
+        self.assertTrue(
+            self.text.rstrip().endswith("Creative Commons may be contacted at creativecommons.org."),
+            "LICENSE is truncated before the closing paragraph",
+        )
+
+    def test_license_keeps_the_clauses_the_card_relies_on(self):
+        for clause in (
+            "worldwide royalty free non sublicensable non exclusive irrevocable license",
+            "reproduce and share the licensed material in whole or in part",
+            "produce reproduce and share adapted material",
+            "patent and trademark rights are not licensed under this public license",
+            "extract reuse reproduce and share all or a substantial portion of the "
+            "contents of the database",
+            "as is and as available",
+        ):
+            self.assertIn(clause, self.normalised, clause)
+
+    def test_license_is_not_a_different_creative_commons_flavour(self):
+        for wrong in (
+            "noncommercial",
+            "non commercial",
+            "sharealike",
+            "share alike",
+            "noderivatives",
+            "no derivatives",
+            "cc0",
+        ):
+            self.assertNotIn(wrong, self.normalised, wrong)
+
+    def test_license_agrees_with_the_identifier_declared_in_the_card(self):
+        readme = README_PATH.read_text(encoding="utf-8")
+        front = readme[4 : readme.find("\n---\n", 4)]
+        self.assertIn(f"license: {LICENSE_HUB_ID}", front)
+        self.assertIn("Creative Commons Attribution 4.0 International", self.text)
+
+
+class PublicUploadSetTest(unittest.TestCase):
+    """The approved upload set must be complete and self-sufficient.
+
+    APPROVAL.md, VALIDATION.md and scripts/validate.py are internal review and tooling
+    documents that do not ship, so nothing in the public set may depend on them.
+    """
+
+    def test_every_public_file_is_present(self):
+        for relative in PUBLIC_UPLOAD_SET:
+            self.assertTrue((ROOT / relative).is_file(), f"{relative} is missing")
+
+    def test_card_does_not_point_at_internal_documents(self):
+        readme = README_PATH.read_text(encoding="utf-8")
+        for relative in INTERNAL_ONLY:
+            self.assertNotIn(Path(relative).name, readme, relative)
+
+    def test_makefile_tolerates_the_internal_validator_being_absent(self):
+        makefile = MAKEFILE_PATH.read_text(encoding="utf-8")
+        self.assertIn("scripts/build_dataset.py", makefile)
+        self.assertIn(
+            "if [ -f scripts/validate.py ]",
+            makefile,
+            "make validate must degrade gracefully when the internal validator does not ship",
+        )
+        self.assertIn("LICENSE", makefile, "make hashes must cover LICENSE")
+
+
+class NoSecretsOrStandInsTest(unittest.TestCase):
+    """Patterns are written self-avoidingly so this file can be scanned by its own rules."""
+
+    PATTERNS = [
+        ("openai-style key", r"sk-[A-Za-z0-9]{16,}"),
+        ("aws access key id", r"AKIA[0-9A-Z]{16}"),
+        ("github token", r"gh[pousr]_[A-Za-z0-9]{20,}"),
+        ("hugging face token", r"hf_[A-Za-z0-9]{20,}"),
+        ("private key block", r"BEGIN [A-Z ]*PRIVATE KEY"),
+        ("bearer credential", r"(?i)bearer\s+[A-Za-z0-9._\-]{16,}"),
+        (
+            "assigned credential",
+            r"(?i)(api[_-]?key|secret|passwd|password|access[_-]?token)"
+            r"\s*[:=]\s*[\"']?[A-Za-z0-9._\-]{12,}",
+        ),
+        ("unfinished-work marker", r"\bT[O]DO\b"),
+        ("unfinished-work marker", r"\bFIX[M]E\b"),
+        ("unfinished-work marker", r"\bX[X]X\b"),
+        ("unresolved stand-in", r"\bT[B]D\b"),
+        ("unresolved stand-in", r"CHANGE[M]E"),
+        ("unresolved stand-in", r"(?i)\bplace[h]older\b"),
+        ("template slot", r"<YOUR[_ ]"),
+    ]
+
+    def test_package_is_free_of_credentials_and_stand_ins(self):
+        hits = []
+        scanned = 0
+        for path in sorted(ROOT.rglob("*")):
+            if not path.is_file() or "__pycache__" in path.parts:
+                continue
+            try:
+                text = path.read_text(encoding="utf-8")
+            except (UnicodeDecodeError, OSError):
+                continue
+            scanned += 1
+            for label, pattern in self.PATTERNS:
+                for match in re.findall(pattern, text):
+                    hits.append(f"{path.relative_to(ROOT)}: {label}: {str(match)[:40]}")
+        self.assertGreater(scanned, 0)
+        self.assertEqual(hits, [])
+
+    def test_no_environment_or_key_material_is_referenced_in_the_data(self):
+        blob = CSV_PATH.read_text(encoding="utf-8") + JSONL_PATH.read_text(encoding="utf-8")
+        for needle in ("Bearer", "api.nativeport.ai", "accounts.nativeport.ai", "panel."):
+            self.assertNotIn(needle, blob, needle)
+
+
+if __name__ == "__main__":
+    unittest.main(verbosity=2)