fix(run): survive malformed links and explain unreachable companies

[anusbutt]

Found while running a 54-company sweep: one company crashed and 50 were misreported.

A single malformed href took down a whole company

dryerventheroes.com ships an unfilled WordPress placeholder in its nav:

<a href="http://[BookingLink]" target="_blank">Schedule Service</a>

Python's urlsplit (3.11.4+) reads the brackets as an IPv6 literal and calls ipaddress.ip_address("BookingLink"), which raises ValueError. discover_extra_pages did not guard the parse, so the exception escaped and the company failed outright:

error: dryer-vent-superheroes-of-dallas: 'BookingLink' does not appear to be an IPv4 or IPv6 address

Per-company isolation contained it — the batch finished — but the prospect was lost over someone else's broken link. The link is now skipped.

Verified against the live site: discover_extra_pages now returns [('contact', 'https://dryerventheroes.com/contact-us/')] instead of raising.

Every unreachable company reported "blank email"

The skip reason echoed the ingest-time bucket_reason, which describes the input row rather than what research found — so all 50 unreachable companies reported blank email, reading as though address recovery never ran.

The reason is now judged after recovery, separating the cases that call for different responses:

Condition	Reason
No website resolved	no website could be resolved
Website resolved, nothing readable	no page could be fetched from <url>
Pages read, no address published	no published address on any fetched page

The first and third are what the README already documented, so this makes the code match the docs rather than the other way round. On the sweep above, those 50 rows now read no published address on any fetched page — which says the keyword is fine and the vertical prefers contact forms.

Guarantees

• No company is silently dropped — strengthened. One company moves from failed to a named, explained skip, and the remaining skips gain a reason that distinguishes a sourcing problem from an unreachable vertical.
• No change to network behaviour, the Facebook host guard, citation validation, the locked fallback, or approval-gated sending. Recovery still issues no new requests.

Tests

• tests/unit/test_extract.py — a bracketed placeholder href alongside a good link; the good link still comes back
• tests/unit/test_pipeline_reasons.py (new) — all three reason branches
• tests/integration/test_email_recovery_batch.py — end-to-end skip reason describes the outcome, not the input row

Full suite: 612 passed.

🤖 Generated with Claude Code