feat(site): compare against Jamie, and drop Fireflies (#213)

Jamie (meetjamie.ai) joins the comparison and Fireflies leaves it, so
the page still runs three competitors rather than four.

Fireflies was the one to go because it is the least like Nojoin on the
axes this page argues: bot-by-default capture, a US cloud, and no
self-hosting position to concede anything against. Jamie is the opposite
on all three, which makes it the harder comparison and the more useful
one. A fifth column was the alternative and it is worse: the detailed
table is already the widest thing on the site, so another column either
squeezes every cell or pushes one somewhere most readers never scroll.

Jamie sits second, directly after Nojoin, because column order decides
which competitor actually gets read and Jamie is the closest product to
Nojoin's own pitch.

Twenty-three claims, each read from Jamie's own current documentation
and carrying the URL and the date. All ten Jamie URLs and all thirty in
the file were re-verified to return 200. The dates are per-claim rather
than per-table: Jamie's rows were researched a day after the others and
say 3 Aug 2026, so a footnote reports when that page was actually read
instead of when the file was last touched.

Jamie does not lose this table, which is the point of running it. It
takes an outright yes on capturing without a bot and on remembering
speakers between meetings, and a partial on assistant write access --
the row Nojoin leads with. Its thirteen MCP tools do write, just only to
tags, and the cell says so rather than rounding it down. Where Jamie is
behind, the reason is structural: a proprietary licence, an EU cloud
that is still somebody else's, vendor-chosen models, and a credit a
meeting that locks the notes you already have when it runs out.

The one row Jamie is alone on is live in-meeting guidance: its widget is
a manual scratchpad and everything else happens after recording stops.
That stays out of the summary table, because Otter and Granola both have
some form of it, so it is not a line between Nojoin and the field.

Two table fixes came with it. The row label was white-space: nowrap, so
"What an assistant can do with your meetings" claimed about 385px of a
table whose whole job is the prose beside it; a min-width does not fix
that, because auto table layout sizes a column from content and happily
overshoots a minimum, so the label column now takes an explicit 12rem
and wraps. Product columns gain about 40px each as a result -- wider
than before this change, at the same column count. And .glance-table
carried a dead min-width: 32rem that never applied, since the element
also carries .compare-table and that rule is declared later with equal
specificity; the comment above it claimed a narrower glance table that
has never existed. Removed rather than left as intent that does nothing.

Verified: all 30 source URLs return 200, no horizontal overflow across 4
pages x 2 widths x 4 theme combinations, the detailed table needs no
sideways scroll above 1024px, skim layer holds at 456 words, and the
contrast gate, docs validator, 1333 backend tests, scripts/check.py and
git diff --check all pass.

Author: Valtora

docs/SITE.md

@@ -211,6 +211,21 @@ credibility of every checkable one.
 Every claim on the site is traceable to `README.md` or a file in `docs/`; nothing is
 invented for marketing effect.
 
+**Three competitors, and which three is a decision.** The page compares Nojoin against
+Jamie, Otter and Granola. Fireflies was dropped when Jamie was added rather than running a
+fifth column: the detailed table is already the widest thing on the site, and another column
+either squeezes every cell or pushes one somewhere nobody scrolls to. Fireflies went because
+it is the least like Nojoin on the axes this page argues — bot-by-default capture, a US
+cloud, no self-hosting to concede against — and Jamie is the opposite on all three, which
+makes it the harder comparison and the more useful one. Jamie sits second, directly after
+Nojoin, because column order decides which competitor actually gets read.
+
+**Dropping a competitor is not a licence to pick easy ones.** The set has to stay honest, and
+the test is whether the table still contains rows a competitor wins. Jamie takes an outright
+yes on no-bot capture and on remembering speakers, and a partial on assistant write access —
+the row Nojoin leads with. If a future edit leaves Nojoin sweeping every row, the problem is
+the competitor set, not the product.
+
 The comparison page holds itself to the standard it would want applied to it:
 
 - Every competitor claim is verified against **that vendor's own current documentation**,
@@ -262,8 +277,9 @@ rather than the technology.
   length, no history that expires while the customer is still paying. There is no meter in
   the software to hit. Every competitor has a tier where something runs out, which the
   comparison page now shows in a sourced row rather than asserting — Otter meters minutes
-  below its upper plans, Fireflies caps storage and AI credits, Granola limits history on
-  its entry plan. Fireflies transcribes without limit on every plan, and the row says so.
+  below its upper plans, Granola limits history on its entry plan, and Jamie charges a
+  credit a meeting and locks the notes you already have when they run out. Where a
+  competitor's upper plans lift a limit, the row says so.
 - **Nothing on the page is a promise that breaks on a bad week.** No uptime figure, no SLA,
   no capacity claim. Same day if the instance is down, next working day otherwise, UK hours,
   and the limit named rather than implied. Monitoring is listed as part of the fee because