feat(seo): 6 stron bledow staje sie 14, pod skok ruchu z konkretna data - #228

…ate on it

The reach idea, and it is the only one on the table that depends on nobody's
approval.

December 2026 is a scheduled event. Microsoft switches Basic auth off for SMTP
AUTH and a large number of people start pasting refusals into search boxes on a
known date. Microsoft documents 76 of these codes on one page. Thirteen concern
authentication or relay; eight of those were not covered here.

One page per string is already this project's position, and the reason is in
the generator: a page covering many strings loses each of them to a page
covering one. Microsoft's is a page covering seventy-six.

Sourced rather than invented, and the difference is written on every new page.
The generator has a line saying errors from real hardware are worth more than
anything transcribed from documentation, which these are - so entries now carry
a source, it renders as a paragraph naming where the entry came from, and the
reader can weigh it against what their own machine printed. The six existing
entries have no source either and that is now visible rather than assumed.

Two things were checked rather than assumed. Microsoft publishes most of these
without an SMTP reply code - "5.7.134", not "550 5.7.134" - and inventing one
would have been a fabrication on every page. Only 5.7.1 carries a documented
reply code, 530, and only that entry claims one. That in turn broke
build-cli-errors.py, which derived the enhanced code by splitting on a space
and would have raised on the first entry without a reply code; it matches the
pattern now.

And the reversibility text for wrong-sender said the From address was the part
being refused, which is right for 5.7.60 and wrong for 5.7.64, where the
message could not be attributed to a connector permitted to relay it. Caught by
reading the tool's own output against an entry I had just written. It covers
both now.

The GitHub search numbers that started this were not used. A quoted phrase
search behaves - a nonsense control returns zero - but "535 5.7.3" returns
135,361, which is not a real count of anything. The instrument is unreliable
for numeric strings and the decision was made on Microsoft's documented list
instead.