Record bounded wrapped-label trials and independent conservation review

2026-10-01 19:59:24 -07:00
parent d0896ced75
commit 70b2561fdc
91 changed files with 31343 additions and 0 deletions
@@ -1,5 +1,28 @@
 # Architectural and product decisions

+## Measured label wrapping remains experimental — 2026-10-01
+
+Preserve the isolated E/F prototypes and their
+[validation, locked readings and conservation oracles](experiments/wrapped-label-20261001/validation.md)
+without promoting either placement rule. E centers two measured lines on the
+own stroke; F offsets three lines beside it. Both reduce backing width using
+actual Iced shaping and conserve canonical model data and route geometry. F's
+first reader still misidentified two exchanges; both readers retained a plausible
+wrong conversation ownership. The experiments therefore do not meet dependable
+association at Fit, despite correct E endpoint guesses. Separate readers and a
+single fixture do not establish causal superiority or general usability.
+
+Keep raw routing distinct from derived presentation during gesture previews:
+the adversarial different-edge test exposed stale wrapped bounds after guard
+refusal, fixed by retaining the original core RoutedView. This is reusable
+architecture evidence, not acceptance of private fixture-specific layout code.
+Saved fraction/offset survive under F, but ordinary reopening derives one line;
+external profiles cannot provide a production formatting contract. A general
+format/layout proposal requires explicit persistence semantics, shared consumer
+bounds, native representative workflows and glyph/performance evidence. Further
+work should address view scale and ambiguous stroke ownership rather than tune
+this same label scene until it looks successful.
+
 ## Private source composition decision — 2026-09-23

 Adopt the opaque validated original-span rename batch as the next private