Loading...
Paper ยท ADC Release v1 ยท 2026-05-26
Paper
All files below are frozen as of 2026-05-26. Claims, thresholds, and run IDs are immutable. The canonical figure and all data artifacts match the evidence reported in the journal draft.
โ Frozen export (sealed)as of 2026-05-26 00:00:00 UTCโ sealed snapshot; exact statistics are pinned to this run + time, not the live DB.Release v1 bundle โ the 8/8 result is the sealed preregistered classification; download the artifacts below to verify.
Sealed Thresholds (immutable)
alpha_floor = 0.01 ย ยทย signal_ratio_threshold = 0.08Frozen before any run data was observed. Applied to all 8 dataset/phase predictions.
โโ Download
topology_taxonomy.json~5 KB8/8 preregistered ADC predictions across 5 topology classes โ sealed before any run data was observed.
โฟโ Download
r12_ablation.json~2 KBR12 within-adapter ablation (Graph ON vs OFF). WITHDRAWN โ confounded, sign-unstable; retained for audit only. Causal evidence is the shuffled-DAG control (+0.053, p<0.001, K=1000).
โโ Download
governance_trace.json~80 KBEnd-to-end trace: learner ex_junyi_graph_135350, 100 interactions, full JT attribution per event.
โโ Download
baseline_comparison.json~3 KBMatched evaluation: BKT/DKT/SAKT/GKT/HCIE Phase 2 on same 10 held-out users, same AUC protocol.
โฌกโ Download
canonical_figure.png~120 KBOne-page causal chain figure: Topology โ ADC โ Activation โ Trace โ Ablation โ Delta (300 dpi PNG).
Canonical Chain Figure Preview
Figure generated by
research_validation/paper/figures/generate_figure_canonical_chain.py from sealed artifacts. Six steps: measurement chain (blue, Steps 1โ3) and causal confirmation chain (amber, Steps 4โ6).Permanent Run IDs โ hcie-final-postgres
Phase 1 (no graph injection)
run-217532ca-39e6-4859-a41f-88ed53e904a2Phase 2 (1,052 DAG edges injected)
run-94a3b8ba-015b-4d84-b288-004fe60bc282R12 (graph OFF ablation)
run-aecd9059-aac1-4800-b738-d508eef79608"We introduce the ADC, an instrument showing that explicit prerequisite topology โ and only explicit topology โ activates the transfer dimension of adaptive governance, and we causally confirm this by holding learner sequences constant while toggling graph injection."