Review and fix Luban one-by-one phone runner

2026-05-12 00:31:48 +08:00
parent c29567eda6
commit 0370366000
12 changed files with 1230 additions and 4 deletions
@@ -242,6 +242,13 @@ Integration/verification policy:
 - Provider abstraction lessons: `references/sms-provider-abstraction-and-luban-2026-05-11.md`.
 - Luban UK live test 2026-05-11: with `chickliu2031som6d8@2925.com` email-verified and on OpenAI `add-phone`, 10 United Kingdom/England OpenAI orders using service_id `537935` were bought and held for ~190s each before release. Attempts 1-9 reached `phone-verification` and were not immediately rejected by OpenAI, but Luban `getSms` stayed `msg=wait`; attempt 10 stayed on `add-phone` after submit and should be classified as `submit_no_sms_page`, not a confirmed SMS page. Summary: bought=10, SMS received=0, all orders released successfully. Reference: `references/luban-uk-openai-10-attempts-no-sms-2026-05-11.md`.
 - Luban Chile live test 2026-05-11: service_id `508058` resolved for Chile/OpenAI at cost `0.05`. First number passed the `CL` / `智利` country gate and was bought successfully, but OpenAI stayed on `add-phone` and was classified as immediate reject; the order was held for the 3-minute minimum window and released successfully. Summary: bought=1, rejected_by_openai=1, SMS received=0. Reference: `references/luban-chile-openai-immediate-reject-2026-05-11.md`.
+- OpenAI phone-code submit audit 2026-05-11: current phone-verification DOM uses a single input `name=code`, `autocomplete=one-time-code`, `maxLength=6`. Reuse the proven 5sim Vietnam submit method: native value setter + `InputEvent('input', {inputType:'insertText', data: code})` + `change` + exact `继续` click. The Luban free-country runner's generic plain `Event('input')` submit may not update React state; classify received-but-not-accepted codes as `sms_submit_failed`, not success. Reference/helper: `references/openai-phone-code-submit-method-audit-2026-05-11.md`, `scripts/submit_openai_phone_code_corrected.py`.
+- Luban one-by-one runner review/fix 2026-05-11: after reviewing the 5sim Vietnam success and 335-activation Indonesia success records, the previous ad-hoc Luban runner had defects: continued after OpenAI `糟糕，出错了！`/Bad Request pages, shallow Luban release retries, possible non-React-safe code submit, and under-logged WhatsApp/SMS channel ambiguity. Use `scripts/luban_one_by_one_phone_verify.py` instead of `/tmp/luban_one_by_one_stop_on_sms.py`. It validates `add-phone` readiness before buying, stops on unrecoverable bad pages, loops Luban `reject` until success/already-released/wrong_status, submits codes with the proven `InputEvent` method, and exits while keeping page/order if `sms_submit_failed_hold_for_retry` occurs. Reference: `references/luban-one-by-one-runner-review-and-fixes-2026-05-11.md`.
+- For broad Luban testing, always log page channel text (`WhatsApp` vs SMS). If page says WhatsApp but provider returns SMS-style code, treat submit failures as ambiguous until provider raw fields and OpenAI error text are inspected.
+- Luban unrestricted live test 2026-05-11: 10 actual bought numbers across Brazil, Philippines, Argentina, Indonesia, Vietnam, Netherlands, Malaysia, Mexico, France, and Brazil again; max 2 per country, held/released per 3-minute policy. Results: 7 OpenAI immediate rejects, 2 SMS timeouts after reaching `phone-verification`, and 1 France SMS delivery. France `service_id=735664` delivered in ~11s, but code submission left OpenAI on `phone-verification`; classify as `sms_submit_failed`, not success. Do not set `success_request_id` unless OpenAI leaves `phone-verification`; see `references/luban-unrestricted-openai-10-buys-france-sms-submit-failed-2026-05-11.md`.
+- Deprecated Luban one-by-one runner note 2026-05-11: `/tmp/luban_one_by_one_stop_on_sms.py` / `scripts/luban_one_by_one_stop_on_sms.py` was an earlier ad-hoc version. It proved the stop-on-SMS idea, but it was superseded by `scripts/luban_one_by_one_phone_verify.py` because the old version could continue after OpenAI bad-page states and had shallow release retry. Do not use the old script for new runs; use the optimized script above.
+- Before continuing after a Luban phone run, confirm OpenAI `add-phone` is not in `糟糕，出错了！` state and that the country selector is present; `countryValue:null` / `countryText:null` is a page-state failure, not evidence that a country is unsupported.
+- Session review 2026-05-11: class-level lessons from the 2925 + Luban OpenAI onboarding run are summarized in `references/session-review-2925-luban-openai-onboarding-2026-05-11.md`; use it to avoid route-based 2925 false negatives, distinguish OpenAI phone acceptance from SMS delivery, and preserve Luban 3-minute release timing.

 If using the long-running runner: