[mabry1985]

Summary

E1 — limit shrank the statistics. ebay_price_check and amazon_price_check did listings = listings[:cap] before summarize() / results_found / _record(), so limit=1 reported results_found: 1 and a one-row median, contradicting "statistics cover every result found". Now the stats, results_found and the price-history record cover the full set (still bounded by max_results, as ebay_price_and_profit already was); limit only sizes the returned sample (default 8). Documented in both docstrings.

E2 — concurrent calls read each other's pages. All tools share one Browser / labelled tab, and the host runs sync tools on worker threads, so parallel ebay_price_check calls interleaved navigate → wait → read and reported another query's comps.

• _read_page now holds a per-session threading.Lock for the whole navigate/settle/retry sequence.
• A read that is still hijacked after the single retry now raises EbayError (the docstring promised "a second hijack is reported"; the code returned the foreign data). ebay_page_probe and ebay_session_status pass strict=False and keep reporting it their own way.
• _hijacked compares the search's filters too — LH_Sold/LH_Complete, LH_ItemCondition, _udlo/_udhi (Amazon: low-price/high-price) — not just the keywords. Sort (_sop / s) is judged only when both pages state one, so an implicit default sort is not a false hijack.

One existing fixture used a landed URL without the request's filters (?_nkw=x answering a sold search); it now lands on the real search_url(..., sold=True).

Tests

ruff check . && ruff format --check . clean; pytest -q → 189 passed (172 before + 17 new). The 12 new behavioural tests fail against the old tools.py:

• limit trims only the sample (eBay + Amazon, incl. history row count)
• second hijack → error, retried exactly once; probe still reports instead of failing
• filter-mismatch hijack matrix (sold vs active, condition, price band, sort, Amazon)
• 6 threads on a shared stub tab: max one navigate→read in flight, each result carries its own query

No version bump / release.

🤖 Generated with Claude Code

[protoreview]

QA panel review — PASS

code-review-structural · head dddfb1560c2a · formal

No findings were raised, so there was nothing to verify. The panel (all five finders plus the structural engine) reviewed the diff and found no defects. No disagreements, no coverage gaps, no prior requests to disposition.

No findings — the review came back clean.

findings JSON (machine-readable)

[]

[protoreview]

Promoting the PASS verdict for head dddfb1560c2a: all checks terminal-green, zero unresolved review threads. (approve-on-green)