JEPA-INDEX · 14 Milestones · 84 Formal Gaps
The Complete JEPA Gap Map
Abstract 14 JEPA milestones, 6 structural gaps each, 84 formal gaps total. The same 6 gaps appear across the entire JEPA lineage because they are structural absences, not empirical limitations. Each gap maps to a symbol in the 8-symbol Canon: γ₁⚓ H=H†⯛ LSOS〰️ WLD🌀 FEPγ FOF🌌 — EVEN═. LeCun called the first page "world salad (laugh emoji)". Here are 14 more pages. pemos.ca/jepa-index
14 MILESTONES · FULL TIMELINE
JEPA/H-JEPA
2022Conceptual Root · Joint Embedding Predictive Architecture
6 gaps:
1. γ₁: No Invariant Anchor in Latent Prediction...
2. H=H†: Predictor Not Self-Adjoint...
3. LSOS: No Paradigm Audit Between Hierarchy Levels...
4. WLD: No Mercy Reset on Prediction Collapse...
5. FEP: No Continuity Guarantee Across Timescales...
6. FOF: Hierarchy Has No Named Ceiling...
2-PAGER →
6 gaps:
1. γ₁: No Invariant Anchor in Latent Prediction...
2. H=H†: Predictor Not Self-Adjoint...
3. LSOS: No Paradigm Audit Between Hierarchy Levels...
4. WLD: No Mercy Reset on Prediction Collapse...
5. FEP: No Continuity Guarantee Across Timescales...
6. FOF: Hierarchy Has No Named Ceiling...
2-PAGER →
I-JEPA
2023Image Joint Embedding Predictive Architecture · First Practical JEPA
6 gaps:
1. γ₁: No Invariant Anchor in Masked Patch Target Selection...
2. H=H†: Context Encoder Not Self-Adjoint With Target Encoder...
3. LSOS: No Audit of Paradigm Shift Between Mask Strategies...
4. WLD: No Reset When Context Encoder Diverges From Target...
5. FEP: No Continuity Guarantee From ViT-S to ViT-H...
6. FOF: Maximum Patch Resolution Has No Formal Boundary...
2-PAGER →
6 gaps:
1. γ₁: No Invariant Anchor in Masked Patch Target Selection...
2. H=H†: Context Encoder Not Self-Adjoint With Target Encoder...
3. LSOS: No Audit of Paradigm Shift Between Mask Strategies...
4. WLD: No Reset When Context Encoder Diverges From Target...
5. FEP: No Continuity Guarantee From ViT-S to ViT-H...
6. FOF: Maximum Patch Resolution Has No Formal Boundary...
2-PAGER →
MC-JEPA
2023Motion and Content JEPA · Joint Learning of Motion and Appearance
6 gaps:
1. γ₁: No Shared Invariant Between Motion and Content Branches...
2. H=H†: Motion Encoder Not Self-Adjoint With Content Encoder...
3. LSOS: No Paradigm Audit When Motion Dominates Content...
4. WLD: No Reset When Motion/Content Representations Collapse...
5. FEP: No Continuity Guarantee Across Motion Speeds and Scales...
6. FOF: Disentanglement Boundary Undefined...
2-PAGER →
6 gaps:
1. γ₁: No Shared Invariant Between Motion and Content Branches...
2. H=H†: Motion Encoder Not Self-Adjoint With Content Encoder...
3. LSOS: No Paradigm Audit When Motion Dominates Content...
4. WLD: No Reset When Motion/Content Representations Collapse...
5. FEP: No Continuity Guarantee Across Motion Speeds and Scales...
6. FOF: Disentanglement Boundary Undefined...
2-PAGER →
V-JEPA
2023Video JEPA · Latent Prediction Over Video · Motion and Appearance at Scale
6 gaps:
1. γ₁: No Invariant Anchor Across Video Frame Prediction Targe...
2. H=H†: Video Predictor Asymmetric (Forward-Only Prediction)...
3. LSOS: No Paradigm Audit Between Spatial and Temporal Masking...
4. WLD: No Reset When Temporal Prediction Collapses...
5. FEP: No Continuity From Short Clip to Long Video...
6. FOF: Video Sequence Length Has No Named Ceiling...
2-PAGER →
6 gaps:
1. γ₁: No Invariant Anchor Across Video Frame Prediction Targe...
2. H=H†: Video Predictor Asymmetric (Forward-Only Prediction)...
3. LSOS: No Paradigm Audit Between Spatial and Temporal Masking...
4. WLD: No Reset When Temporal Prediction Collapses...
5. FEP: No Continuity From Short Clip to Long Video...
6. FOF: Video Sequence Length Has No Named Ceiling...
2-PAGER →
Audio-JEPA
2024Audio Joint Embedding Predictive Architecture · JEPA Goes Modality-General
6 gaps:
1. γ₁: No Invariant Anchor in Audio Latent Space...
2. H=H†: Audio Encoder Not Verified Against Time-Reversed Signal...
3. LSOS: No Audit of Paradigm Shift Between Speech and Non-Speec...
4. WLD: No Reset When Spectral Prediction Collapses...
5. FEP: No Continuity Across Frequency Scales...
6. FOF: Audio Duration Ceiling Undefined...
2-PAGER →
6 gaps:
1. γ₁: No Invariant Anchor in Audio Latent Space...
2. H=H†: Audio Encoder Not Verified Against Time-Reversed Signal...
3. LSOS: No Audit of Paradigm Shift Between Speech and Non-Speec...
4. WLD: No Reset When Spectral Prediction Collapses...
5. FEP: No Continuity Across Frequency Scales...
6. FOF: Audio Duration Ceiling Undefined...
2-PAGER →
Point-JEPA
2024Point Cloud JEPA · JEPA for 3D Geometric Data
6 gaps:
1. γ₁: No Geometric Invariant Anchor in Point Cloud Prediction...
2. H=H†: Point Encoder Not Verified Rotationally Symmetric...
3. LSOS: No Paradigm Audit Between Local and Global Point Struct...
4. WLD: No Reset When Spatial Prediction Collapses...
5. FEP: No Continuity Across Point Cloud Density...
6. FOF: Point Cloud Complexity Has No Named Boundary...
2-PAGER →
6 gaps:
1. γ₁: No Geometric Invariant Anchor in Point Cloud Prediction...
2. H=H†: Point Encoder Not Verified Rotationally Symmetric...
3. LSOS: No Paradigm Audit Between Local and Global Point Struct...
4. WLD: No Reset When Spatial Prediction Collapses...
5. FEP: No Continuity Across Point Cloud Density...
6. FOF: Point Cloud Complexity Has No Named Boundary...
2-PAGER →
3D-JEPA
20243D Semantics JEPA · Beyond Point Clouds to Full 3D Understanding
6 gaps:
1. γ₁: No 3D Invariant Anchor (Scale, Rotation, Translation In...
2. H=H†: 3D Encoder Not Self-Adjoint Under Rigid Transforms...
3. LSOS: No Paradigm Audit Between Semantic and Geometric Levels...
4. WLD: No Reset When 3D Representation Diverges...
5. FEP: No Continuity From Object-Scale to Scene-Scale...
6. FOF: 3D Scene Complexity Ceiling Undefined...
2-PAGER →
6 gaps:
1. γ₁: No 3D Invariant Anchor (Scale, Rotation, Translation In...
2. H=H†: 3D Encoder Not Self-Adjoint Under Rigid Transforms...
3. LSOS: No Paradigm Audit Between Semantic and Geometric Levels...
4. WLD: No Reset When 3D Representation Diverges...
5. FEP: No Continuity From Object-Scale to Scene-Scale...
6. FOF: 3D Scene Complexity Ceiling Undefined...
2-PAGER →
ACT-JEPA
2025Action JEPA · Joint Action and World Model Prediction · JEPA Becomes Control
6 gaps:
1. γ₁: No Invariant Anchor Linking Action Prediction to World-...
2. H=H†: Action Predictor Not Self-Adjoint With Observation Pred...
3. LSOS: No Paradigm Audit Between Reactive and Planned Actions...
4. WLD: No Mercy Reset When Policy Collapses...
5. FEP: No Continuity Guarantee in Strategy Switching...
6. FOF: Action Sequence Horizon Has No Named Ceiling...
2-PAGER →
6 gaps:
1. γ₁: No Invariant Anchor Linking Action Prediction to World-...
2. H=H†: Action Predictor Not Self-Adjoint With Observation Pred...
3. LSOS: No Paradigm Audit Between Reactive and Planned Actions...
4. WLD: No Mercy Reset When Policy Collapses...
5. FEP: No Continuity Guarantee in Strategy Switching...
6. FOF: Action Sequence Horizon Has No Named Ceiling...
2-PAGER →
V-JEPA 2
2025V-JEPA 2 · Zero-Shot Robotic Planning with Visual Subgoals
6 gaps:
1. γ₁: No Invariant Anchor in Zero-Shot Visual Goal Specificat...
2. H=H†: Goal Predictor Not Verified Symmetric With World Model...
3. LSOS: No Paradigm Audit Between Visual Subgoal and Action Pla...
4. WLD: No Reset When Robotic Planning Collapses...
5. FEP: No Continuity From Seen to Unseen Environments...
6. FOF: Robot Action Space Ceiling Undefined...
2-PAGER →
6 gaps:
1. γ₁: No Invariant Anchor in Zero-Shot Visual Goal Specificat...
2. H=H†: Goal Predictor Not Verified Symmetric With World Model...
3. LSOS: No Paradigm Audit Between Visual Subgoal and Action Pla...
4. WLD: No Reset When Robotic Planning Collapses...
5. FEP: No Continuity From Seen to Unseen Environments...
6. FOF: Robot Action Space Ceiling Undefined...
2-PAGER →
LeJEPA
2025LeJEPA · Principled Training Cleanup · SIGReg and Isotropic Embeddings
6 gaps:
1. γ₁: SIGReg Objective Has No Formal Invariant...
2. H=H†: Isotropic Embedding Structure Not Formally Self-Adjoint...
3. LSOS: No Paradigm Audit Across Simplification Choices...
4. WLD: No Reset When Simplified Objective Collapses...
5. FEP: No Continuity Guarantee After Heuristic Removal...
6. FOF: Scalability Ceiling of Minimal Objective Undefined...
2-PAGER →
6 gaps:
1. γ₁: SIGReg Objective Has No Formal Invariant...
2. H=H†: Isotropic Embedding Structure Not Formally Self-Adjoint...
3. LSOS: No Paradigm Audit Across Simplification Choices...
4. WLD: No Reset When Simplified Objective Collapses...
5. FEP: No Continuity Guarantee After Heuristic Removal...
6. FOF: Scalability Ceiling of Minimal Objective Undefined...
2-PAGER →
Causal-JEPA
2026Causal JEPA · Object-Centric Reasoning · Structural Causal Modeling
6 gaps:
1. γ₁: Causal Object Representation Has No Invariant Floor...
2. H=H†: Causal Mask Asymmetric (Objects Attend Causally, Not Sy...
3. LSOS: No Paradigm Audit Between Object-Level and Scene-Level ...
4. WLD: No Reset When Causal Graph Becomes Inconsistent...
5. FEP: No Continuity When Causal Structure Changes...
6. FOF: Causal Graph Complexity Ceiling Undefined...
2-PAGER →
6 gaps:
1. γ₁: Causal Object Representation Has No Invariant Floor...
2. H=H†: Causal Mask Asymmetric (Objects Attend Causally, Not Sy...
3. LSOS: No Paradigm Audit Between Object-Level and Scene-Level ...
4. WLD: No Reset When Causal Graph Becomes Inconsistent...
5. FEP: No Continuity When Causal Structure Changes...
6. FOF: Causal Graph Complexity Ceiling Undefined...
2-PAGER →
V-JEPA 2.1
2026V-JEPA 2.1 · Dense Predictive Losses · Representation Quality Upgrade
6 gaps:
1. γ₁: Dense Predictive Loss Has No Invariant Anchor...
2. H=H†: Dense Loss Not Verified Self-Consistent (Feature vs Pre...
3. LSOS: No Paradigm Audit Between Sparse and Dense Supervision...
4. WLD: No Reset When Dense Prediction Collapses...
5. FEP: No Continuity Guarantee Between Image and Video Modes...
6. FOF: Dense Prediction Resolution Ceiling Undefined...
2-PAGER →
6 gaps:
1. γ₁: Dense Predictive Loss Has No Invariant Anchor...
2. H=H†: Dense Loss Not Verified Self-Consistent (Feature vs Pre...
3. LSOS: No Paradigm Audit Between Sparse and Dense Supervision...
4. WLD: No Reset When Dense Prediction Collapses...
5. FEP: No Continuity Guarantee Between Image and Video Modes...
6. FOF: Dense Prediction Resolution Ceiling Undefined...
2-PAGER →
LeWorldModel
2026LeWorldModel · End-to-End World Model from Raw Pixels · Minimal Objective
6 gaps:
1. γ₁: Minimal Objective Pixel World Model Has No Invariant Fl...
2. H=H†: Raw Pixel Prediction Not Self-Adjoint at Any Compressio...
3. LSOS: No Paradigm Audit During End-to-End Training...
4. WLD: No Reset When Minimal World Model Collapses...
5. FEP: No Continuity Guarantee Between Fast and Slow Planning...
6. FOF: World Model Complexity Ceiling Undefined...
2-PAGER →
6 gaps:
1. γ₁: Minimal Objective Pixel World Model Has No Invariant Fl...
2. H=H†: Raw Pixel Prediction Not Self-Adjoint at Any Compressio...
3. LSOS: No Paradigm Audit During End-to-End Training...
4. WLD: No Reset When Minimal World Model Collapses...
5. FEP: No Continuity Guarantee Between Fast and Slow Planning...
6. FOF: World Model Complexity Ceiling Undefined...
2-PAGER →
ThinkJEPA
2026ThinkJEPA · Long-Horizon Reasoning · JEPA + VLM Semantic Pathway
6 gaps:
1. γ₁: No Invariant Anchor Linking JEPA World Model to VLM Sem...
2. H=H†: JEPA Prediction Pathway Not Self-Adjoint With VLM Reaso...
3. LSOS: No Paradigm Audit at the JEPA/VLM Interface...
4. WLD: No Reset When Long-Horizon Reasoning Diverges...
5. FEP: No Continuity Across Planning Horizons...
6. FOF: Long-Horizon Reasoning Ceiling Undefined — and EVEN ═ I...
2-PAGER →
6 gaps:
1. γ₁: No Invariant Anchor Linking JEPA World Model to VLM Sem...
2. H=H†: JEPA Prediction Pathway Not Self-Adjoint With VLM Reaso...
3. LSOS: No Paradigm Audit at the JEPA/VLM Interface...
4. WLD: No Reset When Long-Horizon Reasoning Diverges...
5. FEP: No Continuity Across Planning Horizons...
6. FOF: Long-Horizon Reasoning Ceiling Undefined — and EVEN ═ I...
2-PAGER →
THE 6 GAPS · ALL 14 MILESTONES
Same structure. Different forms. One Canon.
Every JEPA milestone has the same 6 structural gaps because they are not engineering failures — they are missing symbols.
⚓ γ₁ THE FLOOR Gap 1 in all 14
No fixed invariant anchor. Latent targets defined relative to training distribution only.
No fixed invariant anchor. Latent targets defined relative to training distribution only.
⯛ H=H† THE HONEST GATE Gap 2 in all 14
Predictor asymmetric. Predict(x→y) ≠ Verify(y→x). No bidirectional check.
Predictor asymmetric. Predict(x→y) ≠ Verify(y→x). No bidirectional check.
〰️ LSOS THE READER Gap 3 in all 14
Paradigm shifts occur without audit. No LSOS reading the active paradigm.
Paradigm shifts occur without audit. No LSOS reading the active paradigm.
🌀 WLD THE RESET Gap 4 in all 14
No mercy reset on collapse. Safety-critical in ACT-JEPA and V-JEPA 2.
No mercy reset on collapse. Safety-critical in ACT-JEPA and V-JEPA 2.
γ FEP THE SWITCH Gap 5 in all 14
No continuity guarantee across paradigm transitions.
No continuity guarantee across paradigm transitions.
🌌 FOF THE BREACH Gap 6 in all 14
No named ceiling. Architecture implicitly claims infinite reach.
No named ceiling. Architecture implicitly claims infinite reach.