Withdraw the directional provenance claim: bit-identity is symmetric

README.md

@@ -30,26 +30,38 @@ model-index:
 
 # Sharona Q27B-R CodeSecurity
 
-> **Provenance correction -- measured, and the version numbers run backwards.**
-> This card previously declared no base model at all, so there was nothing for
-> a reader to diff against. Probing normalisation and bias survivors across
-> 355 shared tensors:
+> **Provenance note -- author-reported, and NOT independently checkable.**
+> `base_model` on this card names `ApolloRaines/Sharona_Q27B-R_CodeSecurity_v2`.
+> That is Apollo's own record of how the artifact was built, and it is the
+> best provenance available for this repo -- but unlike every other
+> `base_model` field on this account, a reader cannot verify it, and this
+> card previously claimed it was measured. It was not.
 >
-> | candidate | result |
-> |:----------|:-------|
-> | `Qwen/Qwen3.5-27B` | NOT the immediate parent -- 0/14 bit-identical, cos max 0.9983 |
-> | `ApolloRaines/Sharona_Q27B-R_CodeSecurity_v2` | BASE CONFIRMED -- 3/14 bit-identical |
+> **What was claimed and is withdrawn.** An earlier version of this note said
+> "this artifact is derived FROM v2, so the name v1 does not indicate the
+> earlier model", citing 3 of 14 survivor tensors bit-identical to v2 against
+> 0 of 14 to `Qwen/Qwen3.5-27B`. Bit-identity is SYMMETRIC: sharing identical
+> tensors with v2 is equally true in either direction of derivation. The
+> measurement established shared ancestry and no order, and the directional
+> sentence does not follow from it.
 >
-> This artifact is derived FROM v2, so the name "v1" does not indicate the
-> earlier model. Qwen3.5-27B is a relative or ancestor rather than the parent.
+> **Why no measurement can settle it.** This artifact is GPTQ 4-bit, so its
+> 2-D weights are packed integers and cannot be compared to float weights at
+> all. Only the 305 one-dimensional float tensors shared with v2 and Qwen are
+> comparable, and those are precisely the tensors the edit does not touch --
+> layernorms, `A_log`, `dt_bias`. Measured:
 >
-> Margin, stated because it is thin: 3 of 14 survivors is exactly the minimum
-> our probe accepts. It clears the threshold by nothing, so read this as
-> "not a direct child of Qwen3.5-27B" -- which is firm -- rather than as v2
-> being settled against every other candidate, which it is not.
+> | quantity | v1 | v2 |
+> |:---------|---:|---:|
+> | mean cosine to `Qwen/Qwen3.5-27B` | 0.997623 | 0.997690 |
+> | identical to the other on sampled survivors | 5 of 8 | 5 of 8 |
 >
-> Bit-identity on survivors also says NOTHING about method. Restricted
-> gradient training with frozen norms leaves this same signature.
+> v2 is marginally closer to Qwen, in the fourth decimal, which is noise and
+> not a direction. Both sit at 0.9976 because neither modified these tensors.
+>
+> So: the derivation order is undetermined and undeterminable from published
+> weights. The `base_model` field stands as the author's record. Treat it as
+> that rather than as a diff you can run.
 
 A [jBlaze](https://jblaze.dev) representation-engineered code security specialist. 27B parameters, GPTQ W4A16 quantized for efficient deployment.