S16 AK_165 x AK_550 EXTERNAL TRANSFERABILITY TEST — V1
Date: 2026-08-22

PRIMARY RESULT
- 6, 8, 10, 12 km: zero mechanically eligible center-radius combinations in both surveys. This is expected from the very coarse AK_550 line spacing and is reported as missing, not substituted.
- 16 km: 18 eligible centers. Topology agrees at 3/18 (16.7%); AK_165 is negative-definite at all 18 while AK_550 is saddle-like at 15/18. No center is saddle-like in both, so the Uruana-style signed saddle frame cannot be retained here.
- 20 km: 358 eligible centers. Topology agrees at 185/358 (51.7%). The match expected from the two marginal topology frequencies is 54.5%, so there is no descriptive excess of topology agreement.
- Among 154 centers saddle-like in BOTH surveys at 20 km, signed negative-curvature axis mismatch has median 62.80 deg. Only 41/154 are within 10 deg; 27/154 are >=80 deg, showing frequent branch-sign swapping rather than stable signed-frame retention.
- The frozen strong transfer claim (signed frame + topology persistence) is therefore NOT SUPPORTED in this first selected external pair. Under the frozen rule, this counts against transferability.

NON-PROMOTED POST-OUTCOME OBSERVATION
- Ignoring eigenvalue sign assignment and comparing the geometric eigenframe modulo 90 deg, the same 154 double-saddle centers have median frame mismatch 12.90 deg; 68/154 are within 10 deg and 106/154 within 20 deg.
- This suggests a possible post-outcome hypothesis: an unsigned geometric frame may persist more readily than the signed curvature branches. This is NOT a rescue of S16 and must be labelled as a new hypothesis if pursued.

OTHER DESCRIPTORS AT 20 KM
- Median local gradient-direction difference: 13.680 deg.
- Median gradient-magnitude ratio AK_550/AK_165: 0.576.
- Fitted-center amplitude Spearman correlation: 0.611; median absolute amplitude difference: 298.225 nT.
These descriptors are not interpreted as independent probabilities and do not override the signed-frame/topology endpoint.
