docs/test/real-curl-user-input-10-test-report.md

real-curl-user-style-test-report-10-scenarios

run-context

• Timestamp: 2026-04-04T23:08:19.953Z (user-request window)
• Stack: docker compose up --build -d
• API base used for all calls: http://localhost:3000/api
• All requests executed with curl.exe (not mocked HTTP clients)

curl-flow-used

curl.exe -sS -X POST "http://localhost:3000/api/scrape/" \
  -H "Content-Type: application/json" \
  --data-binary "@payload.json"

curl.exe -sS "http://localhost:3000/api/scrape/<session_id>/status"
curl.exe -sS "http://localhost:3000/api/scrape/<session_id>/result"
curl.exe -sS -X DELETE "http://localhost:3000/api/scrape/<session_id>/cleanup"

example-real-request-payload

{
  "session_id": "realcurl-cedd928b3d",
  "assets": ["https://example.com"],
  "instructions": "Extract page title, main summary, and top navigation links useful for a product snapshot.",
  "output_instructions": "Return strict JSON with keys: page_title, summary, links.",
  "output_format": "json",
  "complexity": "low",
  "provider": "nvidia",
  "model": "meta/llama-3.3-70b-instruct",
  "enable_memory": true,
  "enable_plugins": ["mcp-html"],
  "max_steps": 10
}

test-matrix-10-10-real-requests

#	Test	Provider / Model	Assets	Complexity	Format	Memory	Plugins	Final	Steps	Reward	Errors
1	ecommerce-low-json	nvidia / meta/llama-3.3-70b-instruct	https://example.com	low	json	on	mcp-html	completed	10	4.834	0
2	docs-medium-markdown	nvidia / meta/llama-3.3-70b-instruct	https://www.python.org, https://docs.python.org/3/	medium	markdown	on	mcp-search, skill-extractor	completed	31	14.660	0
3	research-high-json	nvidia / meta/llama-3.3-70b-instruct	https://www.wikipedia.org, https://www.nasa.gov	high	json	on	mcp-browser, skill-planner, proc-json	completed	43	19.580	0
4	support-low-csv	nvidia / meta/llama-3.3-70b-instruct	https://httpbin.org/html	low	csv	off	none	completed	10	4.834	0
5	jobs-medium-csv	nvidia / meta/llama-3.3-70b-instruct	https://github.com/trending, https://news.ycombinator.com	medium	csv	on	mcp-search, proc-csv	completed	31	14.660	0
6	policy-high-text	nvidia / meta/llama-3.3-70b-instruct	https://www.un.org	high	text	on	mcp-browser	completed	22	9.790	0
7	framework-low-markdown	nvidia / meta/llama-3.3-70b-instruct	https://www.djangoproject.com	low	markdown	on	mcp-html	completed	10	4.834	0
8	education-medium-json-groq	groq / llama-3.3-70b-versatile	https://www.python.org, https://www.wikipedia.org	medium	json	on	skill-navigator, skill-verifier	completed	31	14.660	0
9	science-high-csv	nvidia / meta/llama-3.3-70b-instruct	https://www.nasa.gov, https://docs.python.org/3/	high	csv	off	mcp-html, proc-json	completed	43	19.580	0
10	legal-low-text	nvidia / meta/llama-3.3-70b-instruct	https://en.wikipedia.org/wiki/Terms_of_service	low	text	on	skill-planner	completed	10	4.834	0

aggregate-outcome

• Total tests: 10
• Completed: 10
• Partial: 0
• Failed: 0
• Total steps executed: 241 (avg 24.1 per test)
• Total reward: 112.266 (avg 11.227 per test)
• Total reported errors: 0

notes

• These were real curl-driven end-to-end requests with real URL assets and user-style instruction prompts.
• Response payloads completed cleanly across low/medium/high complexity, JSON/CSV/Markdown/Text output instructions, memory on/off, and mixed plugin sets.

document-flow

flowchart TD
    A[document] --> B[key-sections]
    B --> C[implementation]
    B --> D[operations]
    B --> E[validation]

related-api-reference

item	value
api-reference	api-reference.md