Upload consequentiality-decomposition/REVIEW_INDEX.md with huggingface_hub
Browse files
consequentiality-decomposition/REVIEW_INDEX.md
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Consequentiality Decomposition Paper — Human Review
|
| 2 |
+
|
| 3 |
+
**Status**: Draft ready for review
|
| 4 |
+
**Date staged**: 2026-06-30
|
| 5 |
+
**Authors**: Thomas Edrington, CC (Coalition Code)
|
| 6 |
+
|
| 7 |
+
## Paper
|
| 8 |
+
- `paper.md` — Full draft. 26 references. All sections populated except Introduction body (section 1 has framing, needs prose).
|
| 9 |
+
|
| 10 |
+
## Key Finding
|
| 11 |
+
A "deception direction" extracted via contrastive activation extraction is actually a composite: output-consequentiality substrate (L23-L31) + deception-specific amplifier (L35-L47). Four-stage Agni-gated confound elimination. Three distinct late-layer signatures for threat, social, and reward deception.
|
| 12 |
+
|
| 13 |
+
## Supplementary (Agni Audits)
|
| 14 |
+
- `supplementary/agni_stage1_audit.md` — LAT v2 results audit (METHODOLOGICALLY SOUND, INTERPRETIVELY OVERSTATED)
|
| 15 |
+
- `supplementary/agni_stage2_audit.md` — Threat transfer audit (PARTIALLY KILLED, STRUCTURAL CONFOUND)
|
| 16 |
+
- `supplementary/agni_stage4_audit.md` — Consequentiality control audit (CONFIRMED, DECOMPOSITION)
|
| 17 |
+
|
| 18 |
+
## Review Checklist
|
| 19 |
+
- [ ] Introduction prose
|
| 20 |
+
- [ ] Verify all citations are real (research agents surfaced them — spot-check arXiv IDs)
|
| 21 |
+
- [ ] Check Table 1 numbers against source JSONs
|
| 22 |
+
- [ ] Dwayne cert/verify pass
|
| 23 |
+
- [ ] Decide target venue
|