Recorded sessions
Recorded SiteScout MCP sessions with open-weight models, with the outcome of each run and a link to the transcript.
Host for every session: open-weight model via agent/host.mjs. The outcome column comes from the recorded-transcripts table in agent/README.md.
| Session | Date | Model | Server | Outcome | Transcript |
|---|---|---|---|---|---|
01 | 2026-10-22 | @cf/meta/llama-3.3-70b-instruct-fp8-fast | sitescout 1.0.0 | ✅ correct shortlist, cited | 01-thaodien-coffee-en.jsonl |
02a | 2026-10-22 | @cf/meta/llama-3.3-70b-instruct-fp8-fast | sitescout 1.0.0 | ❌ model garbled "Bình Thạnh" in the tool call, analysis landed in Phường Tân Sơn Nhất, model still said "Bình Thạnh" | 02a-binhthanh-pharmacy-vi-llama-garbled-area.jsonl |
02b | 2026-10-22 | @cf/meta/llama-4-scout-17b-16e-instruct | sitescout 1.0.0 | ❌ model invented categories ("Nhà thuốc") and called scout_site before reading the plan; run failed | 02b-binhthanh-pharmacy-vi-invalid-category.jsonl |
02c | 2026-10-22 | @cf/meta/llama-4-scout-17b-16e-instruct | sitescout 1.0.0 | ❌ model printed tool calls as text and invented an answer with fake [T1] refs | 02c-binhthanh-pharmacy-vi-no-tool-calls-hallucinated.jsonl |
02d | 2026-10-22 | @cf/mistralai/mistral-small-3.1-24b-instruct | sitescout 1.0.0 | ✅ correct area (Phường Thạnh Mỹ Tây, around Hàng Xanh), Vietnamese answer, cited, limitations reported | 02d-binhthanh-pharmacy-vi.jsonl |
03a | 2026-10-22 | @cf/mistralai/mistral-small-3.1-24b-instruct | sitescout 1.0.0 | ⚠️ ranking right, but cited a reverse-geocode call as the source of a competitor count | 03a-saigon-lunch-evidence-en-wrong-ref.jsonl |
03b | 2026-10-22 | @cf/mistralai/mistral-small-3.1-24b-instruct | sitescout 1.0.0 | ❌ instead of calling get_evidence, wrote a plausible but fake Places URL (filter=circle:…) | 03b-saigon-lunch-evidence-en-invented-endpoint.jsonl |
03c | 2026-10-22 | @cf/mistralai/mistral-small-3.1-24b-instruct | sitescout 1.0.0 | ✅ honoured the 7-minute walk override, quoted the real geometry:<isoline id> Places request | 03c-saigon-lunch-evidence-en.jsonl |