# New analysis pass: what survived and what failed

The complete source proxy was submitted again in `09-meta-function-native`, with an
explicit instruction to produce a product-agnostic function and unrelated product calls.
The provider returned useful pointers but failed its own transfer test.

- **Rejected function:** requiring a vulnerable proxy, a secret elder and cultural
  validation still encodes this particular plot. A bookkeeping “Japanese ledger secret”
  and a water sensor described as a “pipe-sealing secret” are not valid product swaps.
  The authored v2 function instead transfers the causal belief progression and requires
  real product-specific demonstrations.
- **Rejected end anchor:** the provider calls a 4:27 event the end of a 5:25.68 film.
  Do not use its clock as the authoritative edit map. Source annotations and bounded
  excerpts retain their known offsets.
- **Retained as interpretation:** question-led discovery, delayed product relevance,
  skeptical explanation, inspection of the result, and a callback before the close.
- **Retained limitation:** intermediate activity scenes and spoken week labels are not
  direct proof of the invisible medical mechanism. They must not become cosmetic evidence.
- **Localization correction:** keep saturated color and appealing environments; local
  life does not require national motifs. The original moodboard is used as color/set
  conditioning with explicit “not identity / not a grid” instructions.

This critique is separate from the immutable model output. It demonstrates why a second
automated “looks good” opinion cannot substitute for evidence and a hostile swap test.