diff options
Diffstat (limited to 'ORAL_B.md')
| -rw-r--r-- | ORAL_B.md | 10 |
1 files changed, 10 insertions, 0 deletions
@@ -12,6 +12,13 @@ No oral-B result may be generated before this protocol is committed. Native accept-bar baselines may continue in parallel, but the oral-B confirmation may not be inspected until C4 is closed. +**Development status: failed on 2026-07-22.** The complete 36-run frozen grid +was generated from clean revision `b4458cd` and no confirmation seed was +touched. All 12 candidates failed the sign-inversion requirement in all three +task environments. No candidate is eligible, so the confirmation specified +below is prohibited. Detailed negative results are retained in `RESULTS.md` +and `results/bci_dev_selection.json`. + ## What the biological experiment establishes The paper motivates four experimental conditions for a vectorized dendritic @@ -106,6 +113,9 @@ no run or seed may be removed. A result file must record the source revision, dirty state, task/model seed, full dynamics, phase masks, scalar reward observations, forward-equivalent work, and peak memory. +This section remains the preregistered confirmation that would have followed +an eligible development result; it was not run after the failed screen. + ### B1. Innovation identification and network predictability - Mean absolute per-cell `corr(r_i, h_i)` is at most 0.10. |