The 2c.3+4 dispatch agent surfaced two stale anchors in the
2026-04-25 h_s2 consumer audit:
1. Row 11 (IQN target trunk): the kernel referenced as
`iqn_compute_target_h_s2` in `iqn_dual_head_kernel.cu:1031` is
actually `iqn_trunk_forward_kernel`, and it is ALREADY ORPHANED
(loaded into IqnHead::trunk_forward_kernel but never invoked).
The real IQN target trunk runs through cuBLAS iqn_lt_matmul calls
in gpu_iqn_head.rs::execute_training_pipeline:820-846 (with a
cached fast-path at 803-847). Mitigation reduced to: delete dead
kernel + extract target_encoder_forward_only + replace cuBLAS
fallback path with that extraction.
2. Row 22 (relu_mask sites): line numbers drifted. Audit said
5581/5798; current code is 5655/5697/5872. Crucially, 5697 is
the h_s1 mask, not h_s2/IQN-aux — the audit's "3 sites" claim
needs verification per site before editing. Updated to the
current line numbers with a "verify before editing" note.
These are documentation-only corrections. The mitigation strategy
(GRN backward chain replaces relu_mask; target_encoder_forward_only
unifies trunk implementation) is unchanged in spirit.
Prerequisite for Plan 4 Task 2c (GRN ADOPT). Inventories every
trunk-h_s2 consumer and classifies whether each tolerates GRN's
zero-mean signed output or requires the legacy ReLU non-negativity.
Output drives Task 2c mitigation strategy: which consumers need
relu(h_s2) shims at consumption time, and which can take signed
input directly.
No code changes — research only.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
