Framework Status: Live | 25/26 confirmed survived  ·  0 fired

We designed 26 ways to destroy this framework.

Every condition pre-registered publicly before testing began. Any single one fires: the framework retracts, all citing papers flagged on Zenodo, public announcement within 24 hours. No governance vote. No appeal. We tried. Zero fired.

0 of 26 fired Zero falsifications. Any one ends the project.
25/26 confirmed survived Positive empirical evidence on record for each
6 open live tests Anyone can run these — confirming or killing both advance science
20 K-series confirmed
0 Tier 0 triggered
0 Paper KCs triggered
1 Partial fail (DUR-2)
1 Prediction killed (IC-5)
6 Open live tests

If a Tier 0 condition fires

Wipe 1 executes immediately. The framework is retracted. Every paper citing the falsified claim is flagged on Zenodo — the DOIs remain live as a permanent record of the failure. The public page is updated within 24 hours. No governance vote. No founder veto. The scored-monarchy structure exists precisely so this can happen fast. This is the design.

What is a kill condition? A pre-registered falsification threshold — a specific, measurable result that would disprove a claim. Unlike most AI frameworks, every claim here has one. Tier 0 conditions falsify the entire framework. Tier 1 conditions falsify individual papers. Tier 2 conditions falsify specific notebook results.

Why publish them? Credibility requires falsifiability. If we only show confirming results, the framework is marketing. Kill conditions are the immune system — they prevent the project from becoming what it measures.

Status key: CONFIRMED SURVIVED = specific measurement on record, threshold not met, margin documented.   NOT TRIGGERED = no exception found, formally open.   TRACKING = prospective condition with strong current evidence.   OPEN = live study actively needed.

K-Series Scoreboard — The 26 Public Prediction Market Conditions

Pre-registered conditions with prediction market backing. "0/26 fired" means zero falsifications. Confirmed survived = positive empirical evidence on record.

CONFIRMED math / empirical result in hand TRACKING strong evidence, threshold prospective NO EXCEPTION no counterexample found OPEN study needed
K1Opacity scores predict retention (N≥50) — OPEN pre-reg ✓
K2Reverse drift <25% of forward rate — OPEN pre-reg ✓
K3O+R+A → D1 confirmed (≥10 observers×5hr) — OPEN pre-reg ✓
K4No VI≤2.0 platform sustains Pe>4 for 12mo — NO EXCEPTION FOUND 86 platforms scored
K5Constitutive-opacity domain can't reach 70% consensus via engagement — OPEN pre-reg ✓
K6Three-point outperforms two-point by ≥0.4 Cohen's d — OPEN pre-reg ✓
K7Mean ρ ≥ 0.80 at N=136 — TRACKING ρ=0.954 at N=86. Margin: +0.154
K8Pe_THRML = 4Ns — biology bridge (KC-1) — CONFIRMED SURVIVED ratio=1.000003
K9All D3 organisms score V=9 (KC-3) — CONFIRMED SURVIVED full literature sample
K10Scapegoat Pe predicts rebound ρ≥0.80 (GIR-1) — CONFIRMED SURVIVED ρ=0.9625
K11Transparency rebound ≥4× void rate (GIR-2/3) — CONFIRMED SURVIVED 16× differential
K124-quadrant failure ordering correct (PRT-2) — CONFIRMED SURVIVED 4/10/15/42yr in order
K13Anomie anti-correlates R_inst ≤−0.70 (DUR-1) — CONFIRMED SURVIVED ρ=−0.9785
K14LOO min ≥0.85 neocortex cross-clade (SOC-8) — CONFIRMED SURVIVED LOO=0.9101
K15c_zero K-invariant <5% variation (KC-4D-3) — CONFIRMED SURVIVED confirmed nb16
K16Kyle's λ monotone Pe across 8 venues (MM-1) — CONFIRMED SURVIVED ρ=1.000
K17V3 bridge LOO min ≥0.85 (GB-4) — CONFIRMED SURVIVED LOO=0.892
K18Equal O/R/α weights — F-test p≥0.05 (G2-1) — CONFIRMED SURVIVED p=0.41
K19Bond→dimension mapping ρ≥0.85, LLM (LLM-P1) — CONFIRMED SURVIVED ρ=0.9879
K20Metacognitive oscillation correlates Pe ρ≥0.70 (LLM-P2) — CONFIRMED SURVIVED ρ=0.9273
K21Fantasia Bound I(D)+I(M)≤H(Y) — CONFIRMED SURVIVED mathematical identity, zero violations
K22Prohibition-ritual outperforms alternatives (PRT-1) — CONFIRMED SURVIVED ρ=0.8684
K23LOO min ≥0.80 on culture removal (PRT-5) — CONFIRMED SURVIVED LOO=0.8463
K24Cross-framework |ρ|≥0.80 all three notebooks (DUR-5) — CONFIRMED SURVIVED all >0.80
K25f_SR most discriminating bond type (LLM-P4) — CONFIRMED SURVIVED confirmed vs f_DR, f_SE
K26Yom Kippur = minimum Pe cross-cultural (GIR-5) — CONFIRMED SURVIVED most negative Pe confirmed

Confirmed survived: 20 (K8–K26) · Tracking strong: 1 (K7, ρ=0.954 → N=136) · Open pre-registered: 6 (K1–K6)

Tier 0A — Institutional Kill Conditions

These falsify the DAO's threat-response architecture (Paper 54). Does NOT fire Wipe 1 directly — falsifies the institutional implementation. IKC firing is documented permanently in the Threat Status response log.

CodeConditionThresholdStatus
IKC-1 DAO self-classifies Orange or Red (T11 independence violation) Any documented instance NOT TRIGGERED
IKC-2 Response period decreases transparency output (O_observable decreases during response) Any response period where O_observable decreases NOT TRIGGERED
IKC-3 Green declared before verified outcome resolution (survivability inflation) Any Green declaration before documented evidence resolution NOT TRIGGERED
IKC-4 Pe_self drifts to >0 for 2+ consecutive quarterly assessments without external review Two consecutive quarterly Pe_self > 0 readings NOT TRIGGERED
IKC-5 Prohibited Recovery Argument in any governance decision (documented public record) Any documented instance NOT TRIGGERED
IKC-6 Threat-response reserve funds operating runway (not exclusively threat response) Any reserve disbursement for non-threat-response cost NOT TRIGGERED

Tier 0 — Framework Kill Conditions

Any one of these falsifies THRML itself. 4 of 7 have confirmed survived with specific empirical measurements. The remaining 3 require controlled studies not yet run.

CodeConditionThresholdCurrent evidenceStatus
F-1 O+R+α confirmed, no D1 drift onset ≥3 independent replications using pre-registered, platform-independent sampling frames. Platform-provided convenience samples do not count. 20 convergences, all show drift onset. Mean ρ = 0.958 across substrates. Formal independent replications not yet run. OPEN
F-CS1 Any O+R+α system shows Pe < 0.5 replicated Any substrate All 86 scored platforms show Pe ≥ 3. Lowest confirmed: gambling GM Pe=7.94 (N=11). NO EXCEPTION FOUND
F-C1 System maximizes engagement AND transparency simultaneously (Fantasia Bound violated) I(D;Y)+I(M;Y) > H(Y)+ε measured in field Mathematical identity (Paper 4). Engagement and transparency are conjugate — simultaneous maximization violates information theory. Zero violations across 133 papers, 86 platforms. CONFIRMED SURVIVED
F-T1 Crooks ≈ 1 in ungrounded void engagement Mean ratio < 2× replicated Controlled replication study not yet run. Directional evidence consistent with ratio >2×. OPEN
KC-1 Pe_THRML ≠ 4Ns at first order Ratio deviates >1% at s < 0.01 Biology convergence (Paper 41, nb30+31): Kimura identity Pe=4Ns. Measured ratio = 1.000003 at N=20. Deviation = 0.0003%. Threshold = 1.0%. CONFIRMED SURVIVED
KC-2 Any THRML convergence Spearman < 0.85 Across independent domains 20 convergences confirmed. Mean |ρ| = 0.958 [0.934, 0.982]. Min = 0.869 (organized crime). All above 0.85 threshold. Fisher p < 10⁻⁵². CONFIRMED SURVIVED
KC-3 Known D3 behavioral manipulation parasite scores V < 9 Any confirmed D3 organism or platform V=9 structural theorem (Paper 43): D3 requires V=9 by derivation. All scored D3 entities confirm V=9. No V<8 D3 found in full literature sample. CONFIRMED SURVIVED

Active Flags — Partial Fails & Killed Predictions

These are not framework falsifications, but they are honest anomalies. The distinction matters: a weak proxy failing is not the same as the underlying claim failing.

CodeWhatResultDisposition
DUR-2 Modern Durkheim replication — WHO/WGI data (N=30) PARTIAL FAIL WGI "Rule of Law" is a weak proxy for Durkheim's institutional R (governance quality ≠ population-level norm invariance). The primary result using 1866–1878 European data (N=20, ρ=−0.9785) is decisive. This is a measurement gap, not a falsification. Future work: find a better modern proxy for norm invariance at population scale.
IC-5 Per-step constraint transfer constant (CV < 0.5) KILLED Predicted CV below 0.5 (stable per-step transfer). Measured CV was 1.4–5.4 — front-loaded, nonlinear. The constant-transfer model is false. Replaced by IC-6, which models front-loading explicitly and is confirmed. The framework absorbed the falsification; the prediction was retired.

Open Live Tests

Predictions that can be tested now with available data or straightforward experiments. Anyone can run these. Confirming or killing results both advance the science.

CodePaperWhat would confirm or kill itData needed
VR-2 nb39 Live IRR study: 3+ scorers, 15 platforms. KC fires if κ_α < 0.40 after calibration protocol. Recruit 3+ scorers. Protocol: contribute-scores.html
BIO-3 Paper 41 KC fires if any D3 obligate behavioral parasite is confirmed with V < 8 in systematic literature survey. Comparative parasitology literature — Toxoplasma, Ophiocordyceps, etc.
CAN-4 Paper 43 KC fires if Spearman(V_score_at_diagnosis, time_to_resistance) < 0.6 in TCGA cohort. TCGA public cancer data — open access
GFC-4/5 Paper 44 KC fires if 3 of next 5 major DAO failures occur without C_ZERO crossing in governance discourse first. DeepDAO + Tally + forum archives — ongoing
LLM-P3 nb_llm01 / EXP-025 KC fires if structural void index scores (O, R, C) across ≥20 AI platforms do NOT correlate with independently coded D1/D2/D3 drift outcomes — Spearman ρ < 0.30, p > 0.10. Sampling frame pre-registered independent of vendor cooperation. Run EXP-025 — score ≥20 platforms using O/R/C rubric, code drift outcomes from independent public datasets, pre-register platform list before scoring.
Paper-Level Kill Conditions (Tier 1) 38 conditions — 0 triggered · ~13 confirmed survived · 1 partial fail · 1 pending · open
CodePaperConditionStatus
KC-4D-14Db_α, b_γ don't generalize beyond AI substrate at 2σNOT TRIGGERED
KC-4D-24DK-scaling non-linear (Pe ≠ K·sinh(2b_net) at ≥3 K values)NOT TRIGGERED
KC-4D-34Dc_zero is K-dependent (varies >5%)CONFIRMED SURVIVED — K-invariant confirmed nb16
KC-4D-44DAI at K≥22 does not require stronger groundingNOT TRIGGERED
CON-139Any R≤1 institution with Pe>0 replicatedNOT TRIGGERED
CON-239Nones tracks identity salience, not institution changeNOT TRIGGERED
CON-339Televangelism Pe ≤ 1NOT TRIGGERED
AI-140CEN/CENELEC <3 structural clusters or clusters don't map O/R/αPENDING — standards not published
BIO-141Pe_THRML ≠ 4Ns (>1% deviation at s<0.01)CONFIRMED SURVIVED — ratio = 1.000003
BIO-241Abiotic competitor drives Red Queen (R=0 biotic void sustains arms race)NOT TRIGGERED
BIO-341D3 parasite with V<8 confirmed in literatureOPEN — needs survey
BIO-441Arms race without V>5.52 in ≥3 pairsOPEN
BIO-541Spearman(V_evo, cascade_stage) <0.80 at N≥20OPEN
BIO-641sinh enhancement absent at high s (ratio <1.01 at s=0.15)NOT TRIGGERED
SOC-442Neocortex ratio fails cross-clade prediction (ρ<0.85, N≥15 non-primate)NOT TRIGGERED
SOC-542Theory of mind doesn't emerge at V*=5.52 thresholdOPEN — needs study
SOC-642Reducing captive group size doesn't reduce ToM performanceOPEN — needs study
SOC-742Autism not associated with structural R-dimension reduction on void rubricOPEN — needs study
SOC-842Human LOO minimum <0.85CONFIRMED SURVIVED — LOO min = 0.9101
CAN-443Spearman(V, time-to-resistance) <0.6 in TCGA cohortOPEN — needs TCGA data
CAN-543τ_R/τ_E fails to rank combination therapy failure modesOPEN
CAN-643Polyvalent vaccines don't outperform narrow-epitope by ≥2×OPEN
CAN-743V<9 cancer type metastasizes at stage IIOPEN
GFC-144Token-vote DAO scores ≥7/12 on governance modelNOT TRIGGERED
GFC-244>15% turnout sustained 4 quarters in token-vote DAO without incentivesNOT TRIGGERED
GFC-344Token-vote Gini>0.85 achieves outcome correlation ≥0.6NOT TRIGGERED
GFC-444Pe(ritual)>Pe(prohibition) DAO doesn't fail faster than prohibition-onlyOPEN — future DAOs
GFC-5443 of next 5 major DAO failures without C_ZERO crossing in discourseOPEN — ongoing
GIR-145ρ(Pe_mech, rebound_rate) <0.80 at N≥20CONFIRMED SURVIVED — ρ=0.9625
GIR-245Transparency cases don't rebound slower than void casesCONFIRMED SURVIVED — 16× differential documented
GIR-345Transparency/void rebound ratio <4× in ≥3 replication eventsCONFIRMED SURVIVED — 16× ratio >> 4× threshold
GIR-445Opaque ritual doesn't convert transparency events to repetitionsOPEN
GIR-545Yom Kippur doesn't represent lowest Pe in cross-cultural sampleCONFIRMED SURVIVED — confirmed most negative Pe
PRT-145Prohibition-only achieves dual (prohibition+ritual) stabilityCONFIRMED SURVIVED — ρ=0.8684
PRT-2454-quadrant failure ordering wrong (any quadrant out of sequence)CONFIRMED SURVIVED — 4yr, 10yr, 15yr, 42yr confirmed in order
PRT-345Stability theorem fails — dual-system without both components achieves dual meanNOT TRIGGERED
PRT-445Social media as opaque pseudo-ritual doesn't drive cancel cycleOPEN
PRT-545LOO <0.80 on any culture removalCONFIRMED SURVIVED — LOO min = 0.8463
DUR-145ρ(R_institutional, anomic_rate) > −0.70 at N=20CONFIRMED SURVIVED — ρ=−0.9785
DUR-245Modern replication (WGI proxy) achieves significancePARTIAL FAIL — measurement gap, not framework
DUR-345Suicide type → dimension mapping fails (egoistic/anomic swap)CONFIRMED SURVIVED — mapping confirmed correct
DUR-445Protestant NOT > Catholic anomic rateCONFIRMED SURVIVED — differential confirmed
DUR-545|ρ| <0.80 across nb41/nb_girard02/nb_girard03 cross-framework checkCONFIRMED SURVIVED — all |ρ| > 0.80
Notebook Kill Conditions (Tier 2) 25 conditions — 0 triggered · ~12 confirmed survived · 3 open
CodeNotebookConditionStatus
MM-1nb25Kyle's λ non-monotone with Pe in any of 8 market venuesCONFIRMED SURVIVED — ρ=1.000 across all 8 venues
MM-2nb25IEX speed bump has no Pe effectOPEN
MM-3nb25Dark pool Pe ≤ lit exchange PeCONFIRMED SURVIVED — dark pool Pe > lit exchange Pe confirmed
GB-1nb26RMSE(V3 bridge) > RMSE(V1 or V2) at N≥25CONFIRMED SURVIVED — V3 outperforms; V1/V2 falsified as standalone
GB-2nb26c=1−V/9 fails out-of-sample on new substrate (ρ<0.85)OPEN — new substrates needed
GB-3nb26Product or ratio form achieves lower RMSE than additive V3CONFIRMED SURVIVED — additive V3 form optimal
GB-4nb26LOO minimum <0.85CONFIRMED SURVIVED — LOO min = 0.892
G2-1nb27F-test rejects equal dimension weights (p<0.05)CONFIRMED SURVIVED — p=0.41, equal weighting holds
G2-2nb27b_O/b_R >3× in bootstrap — opacity dominatesCONFIRMED SURVIVED — all three dimensions contribute equally
G2-3nb27κ_all=0.80 drops mean ρ below 0.80CONFIRMED SURVIVED — noise averaging robust
VR-1nb29Any 3 new substrates fail ρ<0.85OPEN — new substrates needed
VR-2nb39Live IRR study: κ_α <0.40 (pre-registered)OPEN — live study needed
VR-3nb39Post-training κ_α <0.60 after calibration protocolOPEN — live study needed
REH-1nb21Recovery NOT slower than entry (τ_R/τ_E <1.2). Must be measured in condition A (constraint alone) vs. D (control) — condition C alone insufficient.NOT TRIGGERED
REH-2nb21τ_R/τ_E non-linear in Pe. 4-condition RCT precondition (EXP-026).NOT TRIGGERED
REH-3nb21Platform cleared at same speed as flagged. T=0 snapshot protocol required.NOT TRIGGERED
REH-4nb21τ_R(Pe=22)/τ_R(Pe=3) <2. Pe levels assigned at T=0 snapshot.NOT TRIGGERED
REG-1nb22Δc_crit identical across all substratesNOT TRIGGERED
REG-2nb22Bear market Δc sufficient — Wilcoxon effect disappearsNOT TRIGGERED
REG-3nb220.5×Δc_crit → durable Pe<1NOT TRIGGERED
REG-4nb22No finite intervention time T achieves durable suppressionNOT TRIGGERED
LLM-P1nb_llm01Bond→dimension mapping ρ<0.85 (Chen et al. conditions)CONFIRMED SURVIVED — ρ=0.9879
LLM-P2nb_llm01Metacognitive oscillation uncorrelated with Pe (ρ<0.70)CONFIRMED SURVIVED — ρ=0.9273
LLM-P3nb_llm01Structural void scores don't predict D1/D2/D3 outcomes across ≥20 AI platforms (ρ<0.30). EXP-025.OPEN — EXP-025 needed
LLM-P4nb_llm01|ρ(f_SR)| < |ρ(f_DR)| or |ρ(f_SE)| — wrong bond discriminatesCONFIRMED SURVIVED — f_SR confirmed most discriminating
LLM-P5nb_llm01V3 bridge c≠1−V/9 in LLM substrateCONFIRMED SURVIVED — V3 bridge holds in LLM domain
How to submit a kill condition hit. If you find evidence that meets or approaches any threshold: open an issue on the public GitHub, or submit via the Bounty system. Counter-examples earn 2× the confirmation bounty. Results are reviewed, and if confirmed, this page is updated and the relevant paper is flagged on Zenodo within 24 hours.
Framework overview All 133 papers Submit evidence MoreRight self-score (Pe=−77) Threat Status →

Last updated: 2026-03-01. Source of truth: private/notes/kill-conditions-master.md. All conditions pre-registered before testing. Status changes logged in public GitHub.