The Luck sweep's non-empty-label assertions could not fail: _bind_bay and
_bind_callings always wrap prose in BBCode italics ("[i]%s[/i]"), and
_origin_prose() always emits the literal connective " Now you carry a %s's
work — " even when every authored fragment/blurb is blank. So
"[i][/i]".strip_edges().is_empty() reads false regardless of what ContentDB
served — the 28-combination sweep was proving the wrapper is non-empty, not
that §7's own guard against a blank ContentDB actually holds. Replaced the
emptiness checks with assertions that the DM panel and detail panel render
the WORLD's own fragment/blurb strings, and that those strings are
themselves non-blank, for every race x calling combination.
The M3-a collapse guard (`assert_gt(s.size.x, 0.0, ...)`) also could not
fail: the .tscn root is authored at exactly 1920x1080, matching the
project's viewport size, so the assertion passed on the authored offsets
alone with _fit_to_viewport() and its size_changed connection deleted
entirely. Replaced it with an actual resize proof: setting
Window.content_scale_size (the property this project's canvas_items stretch
mode actually keys Control.get_viewport_rect() off of — get_viewport().size
does not move it headless) to a non-1920x1080 value and asserting the root
followed, then restoring it.
Also covers three previously-unasserted binder behaviours: a race-granted
skill chip going inert (and a chip going inert once the pool's picks are
spent), the bonus-skill row's visibility gate per race, the ability card's
primary-stat Tag/theme_type_variation, and _on_enter_pressed's early-out on
an illegal draft.
Every new/changed guard was proven red against the bug it names (blanked a
calling's blurb; hardcoded a placeholder origin string; deleted
_fit_to_viewport's wiring; dropped the chip.disabled line; forced the bonus
row visible) and reverted. Suite: 298 -> 302, all green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Phillip Tarrant
4494856a44