diff --git a/.github/workflows/misc_sync.yml b/.github/workflows/misc_sync.yml
new file mode 100644
index 0000000..78dd0cb
--- /dev/null
+++ b/.github/workflows/misc_sync.yml
@@ -0,0 +1,107 @@
+name: Misc Invention Web Sync
+
+on:
+ workflow_dispatch: # Allows you to run this manually from the Actions tab
+
+permissions:
+ contents: write # Grants permission to push files directly to your repo
+
+jobs:
+ sync-archives:
+ runs-on: ubuntu-latest
+ steps:
+ - name: Checkout Repository
+ uses: actions/checkout@v4
+
+ - name: Setup Python
+ uses: actions/setup-python@v4
+ with:
+ python-version: '3.10'
+
+ - name: Build Structure and Download Assets
+ run: |
+ cat << 'EOF' > web_sync.py
+ import os
+ import time
+ import urllib.request
+ import ssl
+ import json
+ import shutil
+
+ ssl._create_default_https_context = ssl._create_unverified_context
+ GITHUB_USER = "Devanik21"
+
+ INVENTIONS = {
+ "constructs/Misc/001_LIM": "Latent-Inference-Manifold",
+ "constructs/Misc/002_LCM": "Latent-Consensus-Manifold",
+ "constructs/Misc/003_XV": "xylia-vision",
+ "constructs/Misc/004_LB": "Life-Beyond",
+ "constructs/Misc/005_TM": "Thermodynamic-Mind",
+ "constructs/Misc/006_DTM": "Dark-Thermodynamic-Mind",
+ "constructs/Misc/007_AARGE": "AION-Algorithmic-Reversal-of-Genomic-Entropy",
+ "constructs/Misc/008_DU": "Deep-Universe"
+ }
+
+ PENDING = []
+
+ def make_perfect_structure(folder, repo_name=""):
+ snap_path = os.path.join(folder, "snapshot")
+ notes_path = os.path.join(folder, "notes")
+ versions_path = os.path.join(folder, "versions")
+
+ os.makedirs(snap_path, exist_ok=True)
+ os.makedirs(notes_path, exist_ok=True)
+ os.makedirs(versions_path, exist_ok=True)
+
+ # Generate perfectly formatted internal markdown files
+ with open(os.path.join(notes_path, "evolution.md"), "w") as f: f.write("# Evolution\n")
+ with open(os.path.join(notes_path, "thoughts.md"), "w") as f: f.write("# Thoughts\n")
+ with open(os.path.join(versions_path, "v1.md"), "w") as f: f.write("# Version 1.0\n")
+
+ # Generate standard metadata JSON
+ meta = {"name": repo_name if repo_name else folder.split("/")[-1], "status": "active"}
+ with open(os.path.join(folder, "metadata.json"), "w") as f: json.dump(meta, f, indent=4)
+
+ return snap_path
+
+ # 1. Process known repos
+ for folder, repo in INVENTIONS.items():
+ snap_path = make_perfect_structure(folder, repo)
+
+ for branch in ['main', 'master']:
+ try:
+ readme_url = f"https://raw.githubusercontent.com/{GITHUB_USER}/{repo}/{branch}/README.md"
+ zip_url = f"https://github.com/{GITHUB_USER}/{repo}/archive/refs/heads/{branch}.zip"
+
+ # A. Download README directly into the snapshot folder
+ snap_readme = os.path.join(snap_path, "README.md")
+ urllib.request.urlretrieve(readme_url, snap_readme)
+
+ # B. Copy that exact README into the main invention folder as requested
+ main_readme = os.path.join(folder, "README.md")
+ shutil.copy(snap_readme, main_readme)
+
+ # C. Download ZIP strictly into the snapshot folder
+ urllib.request.urlretrieve(zip_url, os.path.join(snap_path, f"{repo}.zip"))
+
+ print(f"✅ Synced {repo} ({branch})")
+ break
+ except Exception:
+ continue
+ time.sleep(1)
+
+ # 2. Build empty pending structures to maintain standard architecture
+ for folder in PENDING:
+ make_perfect_structure(folder)
+
+ print("All downloads and structural formatting complete.")
+ EOF
+ python web_sync.py
+
+ - name: Commit and Push to Web Repo
+ run: |
+ git config --global user.name "github-actions[bot]"
+ git config --global user.email "github-actions[bot]@users.noreply.github.com"
+ git add .
+ git commit -m "Flawless sync: READMEs in main & snapshot folders, ZIPs in snapshot, full directory structure built for Misc constructs" || echo "No changes to commit"
+ git push origin main
diff --git a/.github/workflows/perfect_sync.yml b/.github/workflows/perfect_sync.yml
index 4e4ebd1..0341bea 100644
--- a/.github/workflows/perfect_sync.yml
+++ b/.github/workflows/perfect_sync.yml
@@ -44,15 +44,7 @@ jobs:
"constructs/010_LDD": "Lucid-Dark-Dreamer",
"constructs/011_BA": "BSHDER-Architecture",
"constructs/012_GGEO": "GENEVO-GENetic-EVolutionary-Organoid",
- "constructs/013_HM": "HAG-MoE",
- "constructs/Misc/001_LIM": "Latent-Inference-Manifold",
- "constructs/Misc/002_LCM": "Latent-Consensus-Manifold",
- "constructs/Misc/003_XV": "xylia-vision",
- "constructs/Misc/004_LB": "Life-Beyond",
- "constructs/Misc/005_TM": "Thermodynamic-Mind",
- "constructs/Misc/006_DTM": "Dark-Thermodynamic-Mind",
- "constructs/Misc/007_AARGE": "AION-Algorithmic-Reversal-of-Genomic-Entropy",
- "constructs/Misc/008_DU": "Deep-Universe"
+ "constructs/013_HM": "HAG-MoE"
}
PENDING = []
diff --git a/Misc/001_LIM/README.md b/Misc/001_LIM/README.md
deleted file mode 100644
index 282040d..0000000
--- a/Misc/001_LIM/README.md
+++ /dev/null
@@ -1,28 +0,0 @@
-# Latent Inference Manifold
-
-    
-
-> Exploring the geometry of learned representations — manifold structure, intrinsic dimensionality, and latent space topology in deep neural networks.
-
----
-
-**Topics:** `arc-agi-2` · `inference-time` · `latent-space` · `machine-learning` · `meta-learning` · `multi-agent-reinforcement-learning` · `multi-agent-systems` · `neurons` · `program-synthesis`
-
-
-
-**Devanik Debnath**
-B.Tech, Electronics & Communication Engineering
-National Institute of Technology Agartala
-
-[](https://github.com/Devanik21)
-[](https://www.linkedin.com/in/devanik/)
-
----
-
-## License
-
-This project is open source and available under the [MIT License](LICENSE).
-
----
-
-*Built with curiosity, depth, and care — because good projects deserve good documentation.*
diff --git a/Misc/001_LIM/metadata.json b/Misc/001_LIM/metadata.json
deleted file mode 100644
index 366154f..0000000
--- a/Misc/001_LIM/metadata.json
+++ /dev/null
@@ -1,4 +0,0 @@
-{
- "name": "Latent-Inference-Manifold",
- "status": "active"
-}
\ No newline at end of file
diff --git a/Misc/001_LIM/notes/evolution.md b/Misc/001_LIM/notes/evolution.md
deleted file mode 100644
index 8073517..0000000
--- a/Misc/001_LIM/notes/evolution.md
+++ /dev/null
@@ -1 +0,0 @@
-# Evolution
diff --git a/Misc/001_LIM/notes/thoughts.md b/Misc/001_LIM/notes/thoughts.md
deleted file mode 100644
index b2afb31..0000000
--- a/Misc/001_LIM/notes/thoughts.md
+++ /dev/null
@@ -1 +0,0 @@
-# Thoughts
diff --git a/Misc/001_LIM/snapshot/Latent-Inference-Manifold.zip b/Misc/001_LIM/snapshot/Latent-Inference-Manifold.zip
deleted file mode 100644
index 3eb03b9..0000000
Binary files a/Misc/001_LIM/snapshot/Latent-Inference-Manifold.zip and /dev/null differ
diff --git a/Misc/001_LIM/snapshot/README.md b/Misc/001_LIM/snapshot/README.md
deleted file mode 100644
index 282040d..0000000
--- a/Misc/001_LIM/snapshot/README.md
+++ /dev/null
@@ -1,28 +0,0 @@
-# Latent Inference Manifold
-
-    
-
-> Exploring the geometry of learned representations — manifold structure, intrinsic dimensionality, and latent space topology in deep neural networks.
-
----
-
-**Topics:** `arc-agi-2` · `inference-time` · `latent-space` · `machine-learning` · `meta-learning` · `multi-agent-reinforcement-learning` · `multi-agent-systems` · `neurons` · `program-synthesis`
-
-
-
-**Devanik Debnath**
-B.Tech, Electronics & Communication Engineering
-National Institute of Technology Agartala
-
-[](https://github.com/Devanik21)
-[](https://www.linkedin.com/in/devanik/)
-
----
-
-## License
-
-This project is open source and available under the [MIT License](LICENSE).
-
----
-
-*Built with curiosity, depth, and care — because good projects deserve good documentation.*
diff --git a/Misc/001_LIM/versions/v1.md b/Misc/001_LIM/versions/v1.md
deleted file mode 100644
index af65bbe..0000000
--- a/Misc/001_LIM/versions/v1.md
+++ /dev/null
@@ -1 +0,0 @@
-# Version 1.0
diff --git a/Misc/002_LCM/README.md b/Misc/002_LCM/README.md
deleted file mode 100644
index 6457808..0000000
--- a/Misc/002_LCM/README.md
+++ /dev/null
@@ -1,902 +0,0 @@
-
-
-
-
-```
-██╗ █████╗ ████████╗███████╗███╗ ██╗████████╗
-██║ ██╔══██╗╚══██╔══╝██╔════╝████╗ ██║╚══██╔══╝
-██║ ███████║ ██║ █████╗ ██╔██╗ ██║ ██║
-██║ ██╔══██║ ██║ ██╔══╝ ██║╚██╗██║ ██║
-███████╗██║ ██║ ██║ ███████╗██║ ╚████║ ██║
-╚══════╝╚═╝ ╚═╝ ╚═╝ ╚══════╝╚═╝ ╚═══╝ ╚═╝
-```
-
-**Latent Consensus Manifold**
-*A 9-Agent Neuro-Symbolic Collective for Abstract Reasoning*
-
-
-
-[](https://python.org)
-[](https://streamlit.io)
-[](https://numpy.org)
-[](https://scipy.org)
-[](LICENSE)
-[-22c55e?style=for-the-badge)]()
-[]()
-
-
-
-> *"Rather than learning monolithic end-to-end mappings, LAteNT decomposes the reasoning process into nine specialized cognitive modules, each contributing distinct epistemic functions."*
-
-
-
-**Author:** [Devanik](https://github.com/Devanik21) • B.Tech ECE '26, NIT Agartala
-**Fellowship:** Samsung Convergence Software Fellowship (Grade I) • Indian Institute of Science
-
-
-
----
-
-## Table of Contents
-
-1. [Abstract](#-abstract)
-2. [Why This Exists](#-why-this-exists)
-3. [System Architecture](#-system-architecture)
- - [The Blackboard](#-the-blackboard-shared-free-energy-substrate)
- - [Domain-Specific Language (DSL)](#-domain-specific-language)
- - [The Nine Agents](#-the-nine-agent-council)
- - [Council Meeting Protocol](#-council-meeting-protocol)
-4. [The ARC-AGI-2 Universe](#-the-arc-agi-2-universe)
-5. [Live Dashboard](#-live-dashboard)
-6. [Experimental Results](#-experimental-results)
-7. [Complexity Analysis](#-complexity-analysis)
-8. [Codebase Tour](#-codebase-tour)
-9. [Installation & Usage](#-installation--usage)
-10. [Roadmap — True AGI Plan](#-roadmap--true-agi-plan)
-11. [Limitations & Honest Assessment](#-limitations--honest-assessment)
-12. [References](#-references)
-13. [Contact](#-contact)
-
----
-
-## 🧠 Abstract
-
-LAteNT implements a **multi-agent neuro-symbolic architecture** for solving abstract reasoning tasks of the ARC-AGI paradigm. Rather than training a monolithic neural network, the system deploys nine cognitively specialized agents over a **shared blackboard substrate**, executing a Socratic loop of iterative hypothesis generation, empirical falsification, and counterfactual causal validation.
-
-The core insight: reasoning under uncertainty is most robust when modeled as an **adversarial epistemic process** — where agents simultaneously propose, challenge, and refine each other's beliefs. This mirrors how scientific communities arrive at ground truth: not through individual brilliance, but through structured disagreement.
-
-**Key contributions:**
-- A 9-agent council with explicit role separation — each agent implements a distinct epistemological function
-- A zero-cheat procedural task generator grounded in Core Knowledge Priors from developmental psychology
-- A growing **Skill Library** that enables meta-learning across episodes via biased hypothesis generation
-- A **Surprise/Free Energy metric** (inspired by Friston's Active Inference) that detects epistemic plateaus and triggers exploration directives
-- A fully auditable blackboard: no agent maintains private state; all reasoning is observable
-- **100% solve rate across 18 procedurally-generated tasks** (avg. 15.9 rounds, 53% budget utilization)
-
----
-
-## 💡 Why This Exists
-
-The [ARC-AGI benchmark](https://arcprize.org/) is designed to resist pattern-matching. Every task requires identifying a transformation rule from 3 training examples and applying it to a novel test input. The transformations are never repeated. State-of-the-art neural models as of early 2026 (Gemini 3 Deep Think: 84.6%; Human baseline: ~80%) still struggle with this benchmark because they lack the ability to:
-
-1. **Discover discrete compositional rules** — not just interpolate between learned patterns
-2. **Falsify their own hypotheses** — confirming bias is a core failure mode of neural networks
-3. **Reason causally** — distinguishing correlation ("this output looks right") from causation ("this transformation law *generates* the output")
-4. **Transfer across episodes** — each task is a cold start for end-to-end models
-
-LAteNT directly addresses all four. It is not a SOTA-chasing architecture — it is a **transparency-first research system** designed to produce interpretable reasoning traces and quantify emergent properties of multi-agent symbolic deliberation.
-
----
-
-## 🏗 System Architecture
-
-The architecture consists of four interlocking subsystems:
-
-```
-┌─────────────────────────────────────────────────────────────────────┐
-│ UNIVERSE (universe.py) │
-│ Procedural ARC-AGI-2 task generator • Core Knowledge Priors │
-│ Zero-cheat fingerprinted task synthesis • L1–L5 difficulty │
-└──────────────────────────────┬──────────────────────────────────────┘
- │ ARCTask
-┌──────────────────────────────▼──────────────────────────────────────┐
-│ BLACKBOARD (memory.py) │
-│ Shared working memory • Hypothesis stack (max 50) • Surprise history│
-│ Contradiction log (max 100) • Agent call log • Meeting agenda │
-└──┬───────────────────────────────────────────────────────────────┬──┘
- │ reads/writes reads/writes
-┌──▼──────────────────────────────────────────────────────────────▼──┐
-│ COUNCIL (council.py) │
-│ │
-│ Perceiver → Dreamer → Scientist → Skeptic → Philosopher │
-│ CausalReasoner → CuriosityEngine → Metacognitor → Archivist │
-└──────────────────────────────┬──────────────────────────────────────┘
- │ snapshots
-┌──────────────────────────────▼──────────────────────────────────────┐
-│ DASHBOARD (LAteNT.py) │
-│ Streamlit live dashboard • 7 tabs • 40+ scientific visualizations │
-│ Real-time agent dialogue • Surprise metric • Skill library browser │
-└─────────────────────────────────────────────────────────────────────┘
-```
-
-### 🗄 The Blackboard: Shared Free Energy Substrate
-
-`memory.py` implements the central knowledge substrate that all agents read from and write to. This is not message-passing — it is a **structured shared memory model** (Newell's Blackboard Architecture, 1962), extended with typed constraints and convergence signals.
-
-**Core data structures:**
-
-```python
-class HypothesisStatus(Enum):
- PENDING → TESTING → FALSIFIED
- → CAUSAL_LAW → ACCEPTED
- → COINCIDENCE
-```
-
-| Structure | Description | Capacity |
-|-----------|-------------|----------|
-| `Hypothesis` | Candidate DSL program + predicted grid + confidence + MDL score + causal verdict | Stack of 50 |
-| `ContradictionEntry` | Falsification event: counter-example, failure mode, originating agent | Log of 100 |
-| `WorldState` | Perceiver output: segmented objects, bounding boxes, grid shape | Per-round |
-| `SurpriseTracker` | Rolling prediction error L2 distance; plateau detection | Real-time |
-| `EpisodeRecord` | Archived task: priors, winning program, rounds, verdict, dialogue log | 500 max |
-| `DSLSkillLibrary` | Indexed primitive library with usage counts and success rates | Unbounded |
-
-**Hypothesis ranking composite score:**
-```
-score(h) = confidence(h) × (1 − false_positive_rate(h)) × (1 + causal_bonus(h))
-where causal_bonus = 1.0 if h.status == CAUSAL_LAW else 0.0
-```
-
-**Blackboard invariants enforced at runtime:**
-- No agent reads from another agent's private state (none exists)
-- All mutations are timestamped and logged to `agent_call_log`
-- Budget is decremented by the blackboard, not by agents
-- `final_verdict` transitions are irreversible once set
-
----
-
-### 🔧 Domain-Specific Language
-
-The DSL (`council.py`, ~100 lines) defines 10 reversible, compositional transformation primitives over integer grids in the 10-color ARC color space:
-
-```python
-DSL.PRIMITIVES = {
- "rotate90": np.rot90(g, 1),
- "rotate180": np.rot90(g, 2),
- "rotate270": np.rot90(g, 3),
- "mirror_h": np.fliplr(g),
- "mirror_v": np.flipud(g),
- "gravity_down": # non-bg cells settle downward per column
- "gravity_up": # non-bg cells settle upward per column
- "majority_recolor": # all non-bg → most frequent color
- "sort_by_size": # objects reordered left→right by ascending cell count
- "identity": # no-op baseline
-}
-```
-
-Programs are sequences of primitive names serialized as `"prim_a → prim_b → prim_c"`. The **MDL score** is simply the program length — shortest programs that generalize win. This implements the **Minimum Description Length principle**: if two programs fit all training examples equally well, prefer the simpler one.
-
-```python
-DSL.execute(grid, ["gravity_down", "mirror_h"]) # returns np.ndarray
-DSL.mdl_score(["gravity_down", "mirror_h"]) # returns 2.0
-```
-
-**Design rationale for primitives:**
-Each primitive preserves semantic content (colors, cell counts) while transforming spatial structure. This ensures that the program search space is tractable (10^6 programs of length ≤6), verifiable in O(H×W) per training pair, and interpretable to humans.
-
----
-
-### 🤖 The Nine-Agent Council
-
-Each agent implements a distinct epistemological function. No agent duplicates another's role.
-
----
-
-#### Agent 1 — Perceiver 👁️
-**Role:** Object segmentation → WorldState
-
-Performs connected-component analysis on raw integer grids to extract discrete objects. Each object is characterized by `{id, color, cells:[(r,c),...], bbox:(r0,c0,r1,c1), size}`. The Perceiver is the only agent with direct read access to the raw grid; all subsequent agents operate on the structured `WorldState`.
-
-**Invariant:** Invoked first in every episode; re-invoked after Philosopher reframing.
-
----
-
-#### Agent 2 — Dreamer 💭
-**Role:** Stochastic hypothesis generation
-
-Generates `K=8` candidate output grids per invocation by sampling DSL primitive compositions:
-
-1. Sample program length uniformly from `{1, 2, 3}`
-2. With 50% probability, prepend a primitive biased by prior art hints from the Archivist
-3. Fill remaining slots with uniform samples from `DSL.PRIMITIVES`
-4. Deduplicate consecutive identical primitives
-5. Execute on all training inputs; compute confidence as fraction of training pairs correctly produced
-6. Push to Blackboard if confidence ≥ 0.3
-
-**Exploration bias:** Primitives in the skill library receive 10× higher sampling weight.
-
----
-
-#### Agent 3 — Scientist 🔬
-**Role:** MDL-optimal program synthesis
-
-Runs `MCTS_ROLLOUTS=60` random program searches per invocation. For each candidate program, checks generalization across all training pairs. If a program perfectly generalizes, scores it with MDL. Attaches the shortest generalizing program to the top hypothesis and marks it `TESTING`.
-
-```python
-def _generalizes(program, task) -> bool:
- for inp, expected in task.train_pairs:
- produced = DSL.execute(inp, program)
- if produced.shape != expected.shape: return False
- if not np.array_equal(produced, expected): return False
- return True
-```
-
-**Key insight:** The Scientist inverts typical program synthesis. Rather than searching for outputs given a program, it searches for a program that explains the Dreamer's already-imagined output. This dramatically narrows the search space.
-
----
-
-#### Agent 4 — Skeptic 🔴
-**Role:** Adversarial falsification (Popperian)
-
-Applies `MUTATION_COUNT=12` structural mutations to training inputs and checks program stability. Mutations include: color swaps, noise injection, grid shifts. A single failing mutation falsifies the hypothesis and logs a `ContradictionEntry`.
-
-```python
-# If mutation of a well-defined input causes shape mismatch under a supposedly general program:
-entry = ContradictionEntry(
- hypothesis_id=top_h.id,
- failure_mode="shape_mismatch_under_mutation",
- agent="Skeptic"
-)
-```
-
-The Skeptic embodies Karl Popper's demarcation criterion: a hypothesis is scientific only if it is falsifiable and has survived falsification attempts.
-
----
-
-#### Agent 5 — Philosopher 🏛️
-**Role:** Ontological reframing
-
-When the Curiosity Engine detects persistent failure, the Philosopher challenges the Perceiver's fundamental object decomposition. It proposes two alternative segmentation schemes:
-
-- **Revision 0 → 1:** Include the background color as an explicit object (relevant when the task manipulates "holes" or "enclosed regions")
-- **Revision 1 → 2:** Merge all cells of the same color into a single object regardless of connectivity (relevant for color-based counting tasks)
-
-After reframing, the Perceiver is immediately re-invoked with the new `WorldState`. This allows the council to discover that the task requires treating "colors as objects" rather than "shapes as objects."
-
----
-
-#### Agent 6 — CausalReasoner 🕸️
-**Role:** Counterfactual causal verification
-
-Tests whether the Scientist's program is a **causal law** or a **spurious coincidence** via `COUNTERFACTUAL_COUNT=8` single-variable interventions:
-
-```python
-def _intervene(grid) -> np.ndarray:
- # Single-variable counterfactual: change one cell's color
- out[r, c] = random_color_not_equal_to(out[r, c])
- return out
-```
-
-**Verdict logic:**
-- If `program(intervened_input) == program(original_input)` despite `intervened_input ≠ original_input`, the program is **insensitive** to its inputs → `COINCIDENCE`
-- Verdict `CAUSAL_LAW` requires sensitivity in fewer than half of counterfactual tests
-
-This prevents the classic failure mode where a program accidentally satisfies all training examples by returning a constant or near-constant output.
-
----
-
-#### Agent 7 — CuriosityEngine ⚡
-**Role:** Active Inference / surprise monitoring
-
-Computes prediction error as normalized L2 distance between the top hypothesis's predicted grid and the ground-truth test output. Tracks error over time via `SurpriseTracker`:
-
-```python
-error_t = ‖predicted - actual‖₂ / (H × W × 9) # normalized to [0, 1]
-```
-
-**Plateau detection:** If `|error[t] - error[t-2]| < ε` for two consecutive observations, the tracker declares `is_plateauing=True`.
-
-**Directives issued on plateau:**
-
-| Condition | Directive |
-|-----------|-----------|
-| ≥3 falsified hypotheses | `PHILOSOPHER_REFRAME` |
-| Odd intervention count | `DREAMER_EXPLORE_LOW_CONFIDENCE` |
-| Even intervention count | `SCIENTIST_EXTEND_SEARCH` |
-
-This implements **active inference** in the sense of Friston: the agent drives itself toward states of minimal free energy (prediction error), not by passively waiting for better hypotheses, but by actively redirecting the council's exploration.
-
----
-
-#### Agent 8 — Metacognitor 🧭
-**Role:** Meeting chair, agenda setter, convergence arbiter
-
-The Metacognitor has no domain knowledge. Its sole function is **meta-cognitive orchestration**:
-
-1. Reads `curiosity_directive` from the CuriosityEngine
-2. Reads hypothesis stack state from the Blackboard
-3. Constructs an ordered `agenda` of agents to invoke this round
-4. When budget is critical, triggers the convergence vote:
-
-```python
-# Convergence vote mechanism
-winner = max(candidates,
- key=lambda h: h.confidence * (1.0 if h.causal_verdict == "CAUSAL_LAW" else 0.5))
-
-if winner.confidence >= 0.30:
- bb.declare_answer(winner.grid, "solved", "Council")
-```
-
-**Agenda examples:**
-
-| State | Agenda |
-|-------|--------|
-| No hypotheses | `[Dreamer, Scientist, Skeptic, CausalReasoner]` |
-| Top hypothesis PENDING | `[Scientist, Skeptic, CausalReasoner]` |
-| Top hypothesis FALSIFIED | `[Dreamer, Scientist, Skeptic, CausalReasoner]` |
-| Curiosity says PHILOSOPHER_REFRAME | `[Philosopher, Perceiver, Dreamer, Scientist, Skeptic, CausalReasoner]` |
-
----
-
-#### Agent 9 — Archivist 📚
-**Role:** Episodic memory, skill extraction, prior art injection
-
-The Archivist bridges episodes. At episode start, it retrieves `k=3` most similar past episodes (matched by prior overlap) and injects their winning programs as hints for the Dreamer. At episode end, it archives the full episode record and extracts skill primitives:
-
-```python
-for prim_name in winning_program.split(" → "):
- skill_lib.add_skill(SkillPrimitive(
- name=prim_name,
- origin_task_id=task.task_id,
- description=f"Used to solve {task.transformation_description}"
- ))
-```
-
-The skill library implements a form of **Bayesian program learning**: primitives that appear in successful programs are biased toward in future hypothesis generation, producing a soft meta-learning signal without gradient descent.
-
----
-
-### 🔄 Council Meeting Protocol
-
-The main loop (`council.py → Council.solve()`) is a **streaming generator** — every agent action yields a `Blackboard.snapshot()` for real-time dashboard rendering:
-
-```
-PHASE 0: ORIENTATION
- └─ Perceiver.perceive(test_input)
- └─ Archivist.inject_hints()
-
-PHASE 1: FIRST IMAGINATION
- └─ Dreamer.imagine(K=8 hypotheses)
-
-PHASE 2: MAIN DEBATE LOOP [repeat until solved or MAX_ROUNDS=30]
- └─ Metacognitor.arbitrate() → sets agenda
- └─ For each agent in agenda:
- ├─ Scientist.synthesize() → attaches program to top hypothesis
- ├─ Skeptic.challenge() → falsifies or passes
- │ └─ if fail: CuriosityEngine.observe() → sets directive → break round
- ├─ CausalReasoner.verify() → CAUSAL_LAW or COINCIDENCE
- │ └─ if CAUSAL_LAW: declare_answer("solved") → exit
- ├─ Dreamer.imagine() → if directed
- └─ Philosopher.reframe() → if directed → re-invoke Perceiver
- └─ CuriosityEngine.observe() [end-of-round]
-
-PHASE 3: ARCHIVAL
- └─ Archivist.archive()
-```
-
-**Budget accounting:** Each round costs 1 budget unit. Budget runs from 0 to 100; `budget_critical` triggers at 85. This forces convergence under uncertainty.
-
----
-
-## 🌌 The ARC-AGI-2 Universe
-
-`universe.py` implements a **zero-cheat procedural task generator**. "Zero-cheat" means no task is ever repeated — each task receives a unique fingerprint (`SHA-256(transformation_composition)`), and the Blackboard enforces that the same fingerprint cannot appear twice in a session.
-
-### Core Knowledge Priors
-
-The task generator is grounded in **Spelke's Core Knowledge Theory** — the set of innate concepts human infants possess by ~6 months of age:
-
-| Prior | Enum | Description |
-|-------|------|-------------|
-| OBJECTNESS | `Prior.OBJECTNESS` | Discrete objects persist through transformation |
-| NUMEROSITY | `Prior.NUMEROSITY` | Quantities are conserved or predictably altered |
-| SYMMETRY | `Prior.SYMMETRY` | Spatial invariances constrain valid transformations |
-| CAUSALITY | `Prior.CAUSALITY` | Transformations exhibit consistent causal structure |
-| CONTAINMENT | `Prior.CONTAINMENT` | Objects may contain other objects |
-| GRAVITY | `Prior.GRAVITY` | Non-background elements settle toward edges |
-| GOAL_DIRECTEDNESS | `Prior.GOAL` | Transformations optimize toward target configurations |
-
-These priors are combined compositionally at difficulty level `L` by selecting `L` priors and chaining their corresponding transformation primitives.
-
-### Difficulty Levels
-
-| Level | Priors | Task Example | Complexity |
-|-------|--------|-------------|------------|
-| L1 | 1 | `mirror_v` | Trivially generalizable |
-| L2 | 2 | `gravity_down → mirror_h` | Requires composition |
-| L3 | 3 | `rotate90 → sort_by_size → majority_recolor` | Multi-step reasoning |
-| L4 | 4 | 4-primitive chain | Expert-level ARC difficulty |
-| L5 | 4+ | Chained dependencies | Frontier-level |
-
-### Task Synthesis Pipeline
-
-```python
-1. Sample priors(difficulty_level) # e.g., [GRAVITY, SYMMETRY]
-2. Compose transform_fn = T₁ ∘ T₂ ∘ ... Tₙ
-3. Generate n_train=3 random input grids # 5×5 to 30×30, 1–15 objects
-4. Produce outputs = [transform_fn(inp) for inp in inputs]
-5. Generate test_input (same prior structure, different grid)
-6. Compute test_output = transform_fn(test_input) # ground truth, hidden from agents
-7. Fingerprint = SHA-256(repr(transform_fn))
-```
-
-Grid specifications:
-- **Dimensions:** 5×5 to 30×30 (procedurally bounded by prior requirements)
-- **Colors:** 10-color ARC standard palette (`{0:black, 1:blue, 2:red, ..., 9:purple}`)
-- **Objects per grid:** 1 to 15
-- **Training pairs:** 3 examples per task
-
----
-
-## 📊 Live Dashboard
-
-The Streamlit dashboard (`LAteNT.py`, ~1880 lines) provides a real-time scientific interface with **7 tabs** and **40+ visualizations** across **10 observatory sections**:
-
-| Tab | Contents |
-|-----|----------|
-| 🏛️ **Council Chamber** | Live agent dialogue stream • Answer comparison grid • Cell accuracy badge • Hypothesis breakdown |
-| ⚡ **Surprise Metric** | Prediction error curve • Resolution stats • Convergence status |
-| 🔬 **Program Inspector** | Discovered DSL rule • MDL score • Step-by-step execution trace • Applied to training example |
-| 🔴 **Skeptic's Dossier** | Contradiction log • All hypotheses table • Falsification breakdown |
-| 📉 **Generalization Curve** | Rounds-to-solve over time • Rolling mean • Solve rate trend |
-| 📚 **Skill Library** | Usage bar chart (builtin vs discovered) • Full skill dataframe with pseudocode |
-| 🔭 **Observatory** | 10 sections, 40+ charts (full breakdown below) |
-
-### Observatory Sections
-
-```
-A — Hypothesis Manifold
- A1 Confidence Cascade (horizontal bar, plasma colormap)
- A2 Status Mosaic (donut chart)
- A3 MDL Score Waterfall (cool colormap)
- A4 Contradiction Pressure (polar bar chart)
- A5 Confidence × Age Heatmap (inferno scatter)
-
-B — Free Energy & Surprise
- B1 Free Energy Landscape (segmented gradient fill)
- B2 Surprise Gradient dE/dt (green/red bars)
- B3 Active Inference Phase Space (E[t-1] vs E[t])
- B4 Resolution Speedometer (polar gauge)
- B5 Entropy Reduction Timeline (cumulative area)
-
-C — Agent Council Activity
- C1 Agent Brain Heatmap (9 agents × N rounds)
- C2 Council Speaking Clock (polar bar)
- C3 Agent Activation Gantt (horizontal timeline)
- C4 Dialogue Density Wave (stacked area)
- C5 Agent Co-activation Matrix (viridis)
-
-D — Skill Meme Grid
- D1 Skill Meme Grid (HSV pixel mosaic, brightness ∝ usage)
- D2 Skill Usage Heatmap (YlOrRd)
- D3 Success Rate Radar (top 8 skills)
- D4 Discovery Timeline (builtin vs emergent)
- D5 Skill Gravity Well (bubble chart)
-
-E — Program Structure Analysis
- E1 Program Length Distribution
- E2 MDL vs Confidence Scatter (magma)
- E3 Primitive Co-occurrence Matrix (RdYlGn)
- E4 Winning Program Spotlight (colorized pipeline)
- E5 Confidence Distribution (histogram)
-
-F — Causal Reasoning Engine
- F1 Causal Law vs Coincidence bar chart
- F2 Falsification Heatmap (Agent × failure_mode)
- F3 Causal Confidence Scatter
- F4 Skeptic Contradiction Spiral (polar)
- F5 Causal Law Rate Over Hypotheses (cumulative %)
-
-G — World State & Perception
- G1 Object Color Distribution (polar)
- G2 Object Size Histogram
- G3 Color Transition Matrix Input→Output (plasma, annotated)
- G4 Philosopher Revision Depth Gauge
- G5 Object Bounding Box Map
-
-H — Multi-Episode Meta-Learning
- H1 Rounds to Solve Learning Curve
- H2 Cumulative Solve Rate
- H3 Difficulty vs Rounds Scatter
- H4 Final Surprise per Episode
- H5 Budget Efficiency per Episode
-
-I — Curiosity Engine Deep Dive
- I1 Directive Frequency
- I2 Plateau Detection on Surprise
- I3 Surprise Spectrum (1D heartbeat colorbar)
- I4 Curiosity Engine Stats box
- I5 Free Energy Convergence (log scale)
-
-J — Emergent Intelligence Metrics
- J1 GI Progress Multi-Ring Gauge (5 metrics)
- J2 Intelligence Fingerprint Radar (8 dimensions)
- J3 Metacognitor Activity Heatmap
- J4 Council Consensus Heat
- J5 System Complexity Score Timeline + Composite GI Score
-```
-
-**Composite General Intelligence Score** (Section J5) is a session-level aggregate:
-```
-GI_score = mean([solve_rate, skill_reuse, surprise_decay,
- causal_law_rate, budget_efficiency, round_efficiency])
-```
-
----
-
-## 📈 Experimental Results
-
-All results reproducible with `seed=11290`.
-
-### Overall Performance
-
-| Metric | Value |
-|--------|-------|
-| **Solve Rate** | **18/18 (100%)** |
-| Avg. Rounds to Solve | 15.9 ± 3.2 |
-| Budget Utilization | 53% (15.9 / 30) |
-| Worst Case | 23 rounds (T0013, L3) |
-| Best Case | 14 rounds (T0001, T0005, T0006, T0007) |
-
-### Per-Task Results
-
-```
-Task ID Difficulty Rounds Verdict
-───────────────────────────────────────────────────────────
-T0000_7046e3eef9c38598 L1 15 ✅ SOLVED
-T0001_f0ff7e211c60a023 L1 14 ✅ SOLVED
-T0002_1e94c74b1c4cd52a L2 15 ✅ SOLVED
-T0003_44f0bbbd3ae17296 L2 15 ✅ SOLVED
-T0004_a633ef1a1bbaadbb L2 15 ✅ SOLVED
-T0005_a31a626e619c8024 L2 14 ✅ SOLVED
-T0006_9224a01a0b0e5d79 L2 14 ✅ SOLVED
-T0007_3eb91af5049b67fe L2 14 ✅ SOLVED
-T0008_efb451d8fa4c9405 L3 15 ✅ SOLVED
-T0009_5b46400abc669d5a L3 15 ✅ SOLVED
-T0010_d4dad90e496df51f L3 18 ✅ SOLVED
-T0011_4689eee23a368e7b L3 17 ✅ SOLVED
-T0012_63a7c7a103865c22 L3 15 ✅ SOLVED
-T0013_4fba135a50799a8e L3 23 ✅ SOLVED ← hardest
-T0014_35ae0bce04df14d6 L3 19 ✅ SOLVED
-T0015_ca91975b5d86d463 L3 18 ✅ SOLVED
-T0016_7bde9a0ea5fce54e L3 16 ✅ SOLVED
-T0017_a8005ea73b232c2d L3 15 ✅ SOLVED
-```
-
-### Skill Library Growth
-
-| Metric | Value |
-|--------|-------|
-| Initial builtin primitives | 10 |
-| Total skills after 18 episodes | 17 |
-| Emergent (discovered) skills | 7 |
-| Transfer rate (emergent → reused) | 5/7 (71%) |
-
-**Top 5 skills by usage:**
-1. `majority_recolor` — 6 uses, **100% success rate**
-2. `gravity_down` — 4 uses, **100% success rate**
-3. `gravity_up` — 4 uses, **100% success rate**
-4. `mirror_v` — 3 uses, **100% success rate**
-5. `sort_by_size` — 3 uses, **100% success rate**
-
-### Agent Contribution (avg invocations/task)
-
-| Agent | Avg. Invocations | Function |
-|-------|-----------------|----------|
-| Dreamer | 6.2 | Hypothesis generation |
-| Metacognitor | 6.2 | Agenda + convergence |
-| Scientist | 6.1 | Program synthesis |
-| Skeptic | 6.0 | Falsification |
-| CuriosityEngine | 6.0 | Surprise tracking |
-| CausalReasoner | 5.8 | Causal validation |
-| Philosopher | 1.8 | Ontological reframing |
-| Perceiver | 1.4 | Segmentation |
-| Archivist | 1.0 | End-of-task archival |
-
----
-
-## ⚙️ Complexity Analysis
-
-| Component | Time Complexity | Space Complexity |
-|-----------|----------------|-----------------|
-| Perceiver (segmentation) | O(H × W) | O(n_objects) |
-| Dreamer (hypothesis gen) | O(K × n_train × L) | O(K) |
-| Scientist (MDL search) | O(\|DSL\|^L × n_train) | O(\|DSL\|^L) |
-| Skeptic (falsification) | O(n_hyp × n_train) | O(1) streaming |
-| CausalReasoner | O(C × L) | O(C) |
-| **Session total** | O(MAX_ROUNDS × \|DSL\|^L × n_train) | O(MAX_HYPO + EPISODE_CAP) |
-
-For typical parameters (`MAX_ROUNDS=30, |DSL|=10, L=6, n_train=3`):
-~100M primitive operations per task, all in NumPy vectorized ops.
-
----
-
-## 📁 Codebase Tour
-
-```
-Latent-Consensus-Manifold/
-├── LAteNT.py # Main Streamlit dashboard (~1880 lines)
-│ ├── ARC_CMAP # ARC 10-color palette → ListedColormap
-│ ├── _render_grid() # Dark-themed ARC grid renderer
-│ ├── _grid_fig() # Multi-panel figure builder
-│ ├── _agent_html() # Colored agent dialogue HTML
-│ ├── _verdict_badge() # Status badge renderer
-│ ├── _winning_program() # Extracts accepted program from snapshot
-│ ├── _answer_grid() # Re-executes program on test input
-│ ├── Sidebar # Difficulty selector + session stats + export
-│ ├── Tab 1–6 # Core analysis tabs
-│ └── Tab 7 # Observatory (sections A–J)
-│
-├── council.py # The 9-agent council (~962 lines)
-│ ├── DSL # Primitive interpreter + program execution
-│ ├── AgentResult # Typed return value dataclass
-│ ├── Perceiver # Connected-component segmentation
-│ ├── Dreamer # Stochastic hypothesis generator
-│ ├── Scientist # MDL program synthesizer (MCTS-style)
-│ ├── Skeptic # Adversarial falsifier (Popperian)
-│ ├── Philosopher # Ontological reframer
-│ ├── CausalReasoner # Counterfactual verifier
-│ ├── CuriosityEngine # Active inference / surprise tracker
-│ ├── Metacognitor # Meeting chair + convergence vote
-│ ├── Archivist # Episode memory + skill extraction
-│ └── Council # Main orchestration loop (streaming generator)
-│
-├── memory.py # Shared state substrate (~565 lines)
-│ ├── HypothesisStatus # Enum: PENDING→TESTING→{FALSIFIED,CAUSAL_LAW,ACCEPTED}
-│ ├── Hypothesis # Candidate program + grid + metadata
-│ ├── ContradictionEntry # Falsification event record
-│ ├── WorldState # Perceiver output: segmented objects
-│ ├── Blackboard # Central mutable state (all agent I/O)
-│ ├── EpisodeMemory # Persistent episode archive (FIFO, 500 cap)
-│ ├── DSLSkillLibrary # Primitive library with usage tracking
-│ └── SurpriseTracker # Rolling L2 error + plateau detector
-│
-├── universe.py # Procedural task generator (~503 lines)
-│ ├── Prior # Enum: 7 core knowledge priors
-│ ├── DifficultyLevel # Enum: L1–L5
-│ ├── GridObject # Segmented object dataclass
-│ ├── ARCTask # Task specification + train/test pairs
-│ ├── GridTransforms # All atomic transformation implementations
-│ └── Universe # Zero-cheat procedural task factory
-│
-├── meta_learner.py # (in development)
-├── latent_dictionary.py # (in development)
-│
-├── requirements.txt
-├── plan.md # True AGI research roadmap
-├── LICENSE # Apache 2.0
-└── Results Archive/
- ├── agi_session_11290.json # 18-task session (seed 11290)
- ├── agi_session_11290_latest.json
- ├── agi_session_70290.json # 18-task session (seed 70290)
- ├── general_intelligence_session_6262.json
- └── readme.md # Full technical paper
-```
-
----
-
-## 🚀 Installation & Usage
-
-### Requirements
-
-```
-Python >= 3.10
-numpy >= 1.24.0
-scipy >= 1.11.0
-streamlit >= 1.32.0
-matplotlib >= 3.7.0
-pandas >= 2.0.0
-```
-
-### Install
-
-```bash
-git clone https://github.com/Devanik21/Latent-Consensus-Manifold.git
-cd Latent-Consensus-Manifold
-pip install -r requirements.txt
-```
-
-### Launch Dashboard
-
-```bash
-streamlit run LAteNT.py
-```
-
-Open `http://localhost:8501`. Select a difficulty level in the sidebar and click **⚡ Run Council**.
-
-### Scripted Usage
-
-```python
-from universe import Universe, DifficultyLevel
-from council import Council
-
-# Reproducible session
-universe = Universe(seed=11290)
-council = Council(seed=11290)
-
-# Generate and solve a task
-task = universe.generate_task(DifficultyLevel.L2)
-print(f"Task: {task.task_id}")
-print(f"Rule (hidden from agents): {task.transformation_description}")
-
-# council.solve() is a streaming generator
-final_snapshot = None
-for snapshot in council.solve(task):
- final_snapshot = snapshot
- print(f"[Round {snapshot['round']}] {snapshot['final_verdict']}")
-
-print(f"\nVerdict : {final_snapshot['final_verdict']}")
-print(f"Rounds : {final_snapshot['round']}")
-print(f"Budget : {final_snapshot['budget_used']}/100")
-
-# Session-level meta-learning stats
-stats = council.stats()
-print(f"Skills discovered: {stats['skill_library_size']}")
-print(f"Avg rounds: {stats['avg_rounds']}")
-```
-
-### Export Session Data
-
-The dashboard provides a **💾 Download Session Data** button in the sidebar after ≥1 task run. The exported JSON contains:
-
-```json
-{
- "seed": 11290,
- "tasks_run": 18,
- "solved": 18,
- "avg_rounds": 15.9,
- "skills": [...],
- "generalization": [...],
- "cumulative_dialogue_logs": [...]
-}
-```
-
-### Self-Test
-
-```bash
-python council.py
-# Runs 3 tasks (L1, L2, L1) and prints per-task results
-```
-
----
-
-## 🗺 Roadmap — True AGI Plan
-
-The current system is an explicitly symbolic baseline. The research roadmap targets a system where **nothing is defined — agents must discover everything from data alone**.
-
-### Phase 1 — Latent Transformation Learning *(3–4 weeks)*
-Replace the handcrafted DSL with **learned transformation embeddings**:
-- Collect 1000+ `(input→output)` pairs across diverse domains
-- Train a transformation autoencoder: each transformation → point in continuous latent space
-- Agents sample from this space rather than from a fixed primitive list
-- Result: agents discover "rotation-like" operations without being told what rotation is
-
-### Phase 2 — Abstraction Discovery Through Pure Induction *(4–5 weeks)*
-Remove the DSL interpreter entirely. Replace program synthesis with **latent space search**:
-- The Scientist searches the learned transformation latent space (no discrete primitives)
-- The Dreamer samples latent transformations and imagines hypothetical outputs
-- Discovery of novel compositions through interpolation in latent space
-- No `IF-THEN`. No hardcoded logic. Pure causal inference in continuous space.
-
-### Phase 3 — Cross-Domain Generalization Test *(3 weeks)*
-The real measure of AGI:
-- Train on Task Domain A (30 tasks)
-- Validate on Domain B (30 tasks, different visual properties)
-- Test on Domain C (30 tasks, completely unseen transformation types)
-- Target: **50%+ on Domain C** without having seen that domain's pattern types
-
-### Phase 4 — Meta-Learning (Learning to Learn Faster) *(3–4 weeks)*
-- Track episode-level discovery speed
-- Train a meta-learner to predict which hypothesis directions are most promising
-- The meta-learner shapes the Dreamer and Scientist purely from episode history
-- Result: system improves its own discovery process through meta-reasoning, not programmer-defined improvements
-
-### Phase 5 — Scientific Validation & Emergence Analysis *(2 weeks)*
-- Does the system discover novel transformation types not in training data?
-- Do agents develop implicit strategies without explicit programming?
-- Can the learned transformation latent space be visualized? Do similar operations cluster?
-- Publish findings: *"Learning Transformation Abstractions Without Explicit Rules"*
-
-**Timeline:** 15–17 weeks to top-0.1% territory.
-**Success threshold:** 50%+ cross-domain transfer + demonstrated meta-learning improvement + published emergence analysis.
-
-**Ultimate goal:** A true AGI capable of solving any problem a human can solve — with infinite tools, skills, knowledge, and memories discovered autonomously.
-
----
-
-## ⚠️ Limitations & Honest Assessment
-
-This system achieves 100% solve rate on a task distribution it was designed for. That is not the same as general reasoning capability.
-
-**1. DSL Expressiveness Ceiling**
-The 10 builtin primitives cover a restricted subset of abstract reasoning. Tasks requiring conditional logic, counting, connectivity-based transformations, or arbitrary color mappings cannot be solved by construction. The 100% solve rate reflects task-distribution alignment.
-
-**2. Program Search Tractability**
-Scientist search is capped at programs of length ≤6. Solutions requiring longer compositions will not be found within budget.
-
-**3. Hypothesis Stack Saturation**
-Under sustained Dreamer output (8 hypotheses × 20 rounds = 160 generated vs. stack cap of 50), older hypotheses with initially low confidence are discarded. This may cause premature elimination of correct-but-initially-uncertain hypotheses.
-
-**4. Zero-Shot Regime**
-Without training examples, confidence computation is undefined. The system would degrade to random hypothesis sampling.
-
-**5. Transfer Scope**
-The skill library enables transfer within the same task distribution. Generalization to fundamentally different task types (e.g., trained on rotation-type tasks, tested on connectivity-based tasks) is constrained by DSL expressiveness.
-
-**6. Causal Reasoning Depth**
-Single-variable counterfactual testing (changing one cell) is a weak causal intervention. Real causal graph discovery requires structured interventions at the object level, which the current CausalReasoner does not implement.
-
----
-
-## 📖 References
-
-**Program Synthesis**
-- Gulwani, S. (2015). Dimensions in Program Synthesis. *PPLJ*.
-- Solar-Lezama, A. (2008). Program Synthesis by Sketching. *PhD dissertation, UC Berkeley*.
-
-**Multi-Agent Systems**
-- Stone, P., & Veloso, M. (2000). Multiagent systems: A survey from an AI perspective. *Autonomous Robots, 8(3)*, 345–383.
-- Newell, A. (1962). Some problems of basic organization in problem-solving programs. *Self-Organizing Systems*.
-
-**Active Inference & Free Energy**
-- Friston, K., et al. (2017). Active inference and learning. *Neuroscience & Biobehavioral Reviews*.
-- Friston, K. (2010). The free-energy principle: a unified brain theory? *Nature Reviews Neuroscience, 11(2)*, 127–138.
-
-**Causal Inference**
-- Pearl, J. (2009). Causality: Models, Reasoning, and Inference. *Cambridge University Press*.
-- Peters, J., Janzing, D., & Schölkopf, B. (2017). Elements of Causal Inference. *MIT Press*.
-
-**Minimum Description Length**
-- Rissanen, J. (1978). Modeling by shortest data description. *Automatica, 14(5)*, 465–471.
-- Grünwald, P. (2007). The Minimum Description Length Principle. *MIT Press*.
-
-**ARC Benchmark**
-- Chollet, F. (2019). On the Measure of Intelligence. *arXiv:1911.01547*.
-
-**Core Knowledge Theory**
-- Spelke, E. S., & Kinzler, K. D. (2007). Core knowledge. *Developmental Science, 10(1)*, 89–96.
-
-**Falsificationism**
-- Popper, K. (1959). The Logic of Scientific Discovery. *Hutchinson & Co*.
-
----
-
-## 🔗 Contact
-
-
-
-**Devanik**
-B.Tech ECE '26 • National Institute of Technology Agartala
-Samsung Convergence Software Fellowship (Grade I) • Indian Institute of Science
-
-
-
-[](https://github.com/Devanik21)
-[](https://twitter.com/devanik2005)
-[](mailto:devanik@iisertirupati.ac.in)
-
-
-
-*This work represents independent research conducted during the Samsung Convergence Software Fellowship at the Indian Institute of Science. All code, experimental data, and analysis are made available for academic and research purposes.*
-
-
-
----
-
-
-
-**License:** [Apache 2.0](LICENSE) • **Last Updated:** March 2026
-
-*Built with deliberate constraints — not to beat the leaderboard, but to understand reasoning itself.*
-
-
diff --git a/Misc/002_LCM/metadata.json b/Misc/002_LCM/metadata.json
deleted file mode 100644
index 1b67c96..0000000
--- a/Misc/002_LCM/metadata.json
+++ /dev/null
@@ -1,4 +0,0 @@
-{
- "name": "Latent-Consensus-Manifold",
- "status": "active"
-}
\ No newline at end of file
diff --git a/Misc/002_LCM/notes/evolution.md b/Misc/002_LCM/notes/evolution.md
deleted file mode 100644
index 8073517..0000000
--- a/Misc/002_LCM/notes/evolution.md
+++ /dev/null
@@ -1 +0,0 @@
-# Evolution
diff --git a/Misc/002_LCM/notes/thoughts.md b/Misc/002_LCM/notes/thoughts.md
deleted file mode 100644
index b2afb31..0000000
--- a/Misc/002_LCM/notes/thoughts.md
+++ /dev/null
@@ -1 +0,0 @@
-# Thoughts
diff --git a/Misc/002_LCM/snapshot/Latent-Consensus-Manifold.zip b/Misc/002_LCM/snapshot/Latent-Consensus-Manifold.zip
deleted file mode 100644
index f28492c..0000000
Binary files a/Misc/002_LCM/snapshot/Latent-Consensus-Manifold.zip and /dev/null differ
diff --git a/Misc/002_LCM/snapshot/README.md b/Misc/002_LCM/snapshot/README.md
deleted file mode 100644
index 6457808..0000000
--- a/Misc/002_LCM/snapshot/README.md
+++ /dev/null
@@ -1,902 +0,0 @@
-
-
-
-
-```
-██╗ █████╗ ████████╗███████╗███╗ ██╗████████╗
-██║ ██╔══██╗╚══██╔══╝██╔════╝████╗ ██║╚══██╔══╝
-██║ ███████║ ██║ █████╗ ██╔██╗ ██║ ██║
-██║ ██╔══██║ ██║ ██╔══╝ ██║╚██╗██║ ██║
-███████╗██║ ██║ ██║ ███████╗██║ ╚████║ ██║
-╚══════╝╚═╝ ╚═╝ ╚═╝ ╚══════╝╚═╝ ╚═══╝ ╚═╝
-```
-
-**Latent Consensus Manifold**
-*A 9-Agent Neuro-Symbolic Collective for Abstract Reasoning*
-
-
-
-[](https://python.org)
-[](https://streamlit.io)
-[](https://numpy.org)
-[](https://scipy.org)
-[](LICENSE)
-[-22c55e?style=for-the-badge)]()
-[]()
-
-
-
-> *"Rather than learning monolithic end-to-end mappings, LAteNT decomposes the reasoning process into nine specialized cognitive modules, each contributing distinct epistemic functions."*
-
-
-
-**Author:** [Devanik](https://github.com/Devanik21) • B.Tech ECE '26, NIT Agartala
-**Fellowship:** Samsung Convergence Software Fellowship (Grade I) • Indian Institute of Science
-
-
-
----
-
-## Table of Contents
-
-1. [Abstract](#-abstract)
-2. [Why This Exists](#-why-this-exists)
-3. [System Architecture](#-system-architecture)
- - [The Blackboard](#-the-blackboard-shared-free-energy-substrate)
- - [Domain-Specific Language (DSL)](#-domain-specific-language)
- - [The Nine Agents](#-the-nine-agent-council)
- - [Council Meeting Protocol](#-council-meeting-protocol)
-4. [The ARC-AGI-2 Universe](#-the-arc-agi-2-universe)
-5. [Live Dashboard](#-live-dashboard)
-6. [Experimental Results](#-experimental-results)
-7. [Complexity Analysis](#-complexity-analysis)
-8. [Codebase Tour](#-codebase-tour)
-9. [Installation & Usage](#-installation--usage)
-10. [Roadmap — True AGI Plan](#-roadmap--true-agi-plan)
-11. [Limitations & Honest Assessment](#-limitations--honest-assessment)
-12. [References](#-references)
-13. [Contact](#-contact)
-
----
-
-## 🧠 Abstract
-
-LAteNT implements a **multi-agent neuro-symbolic architecture** for solving abstract reasoning tasks of the ARC-AGI paradigm. Rather than training a monolithic neural network, the system deploys nine cognitively specialized agents over a **shared blackboard substrate**, executing a Socratic loop of iterative hypothesis generation, empirical falsification, and counterfactual causal validation.
-
-The core insight: reasoning under uncertainty is most robust when modeled as an **adversarial epistemic process** — where agents simultaneously propose, challenge, and refine each other's beliefs. This mirrors how scientific communities arrive at ground truth: not through individual brilliance, but through structured disagreement.
-
-**Key contributions:**
-- A 9-agent council with explicit role separation — each agent implements a distinct epistemological function
-- A zero-cheat procedural task generator grounded in Core Knowledge Priors from developmental psychology
-- A growing **Skill Library** that enables meta-learning across episodes via biased hypothesis generation
-- A **Surprise/Free Energy metric** (inspired by Friston's Active Inference) that detects epistemic plateaus and triggers exploration directives
-- A fully auditable blackboard: no agent maintains private state; all reasoning is observable
-- **100% solve rate across 18 procedurally-generated tasks** (avg. 15.9 rounds, 53% budget utilization)
-
----
-
-## 💡 Why This Exists
-
-The [ARC-AGI benchmark](https://arcprize.org/) is designed to resist pattern-matching. Every task requires identifying a transformation rule from 3 training examples and applying it to a novel test input. The transformations are never repeated. State-of-the-art neural models as of early 2026 (Gemini 3 Deep Think: 84.6%; Human baseline: ~80%) still struggle with this benchmark because they lack the ability to:
-
-1. **Discover discrete compositional rules** — not just interpolate between learned patterns
-2. **Falsify their own hypotheses** — confirming bias is a core failure mode of neural networks
-3. **Reason causally** — distinguishing correlation ("this output looks right") from causation ("this transformation law *generates* the output")
-4. **Transfer across episodes** — each task is a cold start for end-to-end models
-
-LAteNT directly addresses all four. It is not a SOTA-chasing architecture — it is a **transparency-first research system** designed to produce interpretable reasoning traces and quantify emergent properties of multi-agent symbolic deliberation.
-
----
-
-## 🏗 System Architecture
-
-The architecture consists of four interlocking subsystems:
-
-```
-┌─────────────────────────────────────────────────────────────────────┐
-│ UNIVERSE (universe.py) │
-│ Procedural ARC-AGI-2 task generator • Core Knowledge Priors │
-│ Zero-cheat fingerprinted task synthesis • L1–L5 difficulty │
-└──────────────────────────────┬──────────────────────────────────────┘
- │ ARCTask
-┌──────────────────────────────▼──────────────────────────────────────┐
-│ BLACKBOARD (memory.py) │
-│ Shared working memory • Hypothesis stack (max 50) • Surprise history│
-│ Contradiction log (max 100) • Agent call log • Meeting agenda │
-└──┬───────────────────────────────────────────────────────────────┬──┘
- │ reads/writes reads/writes
-┌──▼──────────────────────────────────────────────────────────────▼──┐
-│ COUNCIL (council.py) │
-│ │
-│ Perceiver → Dreamer → Scientist → Skeptic → Philosopher │
-│ CausalReasoner → CuriosityEngine → Metacognitor → Archivist │
-└──────────────────────────────┬──────────────────────────────────────┘
- │ snapshots
-┌──────────────────────────────▼──────────────────────────────────────┐
-│ DASHBOARD (LAteNT.py) │
-│ Streamlit live dashboard • 7 tabs • 40+ scientific visualizations │
-│ Real-time agent dialogue • Surprise metric • Skill library browser │
-└─────────────────────────────────────────────────────────────────────┘
-```
-
-### 🗄 The Blackboard: Shared Free Energy Substrate
-
-`memory.py` implements the central knowledge substrate that all agents read from and write to. This is not message-passing — it is a **structured shared memory model** (Newell's Blackboard Architecture, 1962), extended with typed constraints and convergence signals.
-
-**Core data structures:**
-
-```python
-class HypothesisStatus(Enum):
- PENDING → TESTING → FALSIFIED
- → CAUSAL_LAW → ACCEPTED
- → COINCIDENCE
-```
-
-| Structure | Description | Capacity |
-|-----------|-------------|----------|
-| `Hypothesis` | Candidate DSL program + predicted grid + confidence + MDL score + causal verdict | Stack of 50 |
-| `ContradictionEntry` | Falsification event: counter-example, failure mode, originating agent | Log of 100 |
-| `WorldState` | Perceiver output: segmented objects, bounding boxes, grid shape | Per-round |
-| `SurpriseTracker` | Rolling prediction error L2 distance; plateau detection | Real-time |
-| `EpisodeRecord` | Archived task: priors, winning program, rounds, verdict, dialogue log | 500 max |
-| `DSLSkillLibrary` | Indexed primitive library with usage counts and success rates | Unbounded |
-
-**Hypothesis ranking composite score:**
-```
-score(h) = confidence(h) × (1 − false_positive_rate(h)) × (1 + causal_bonus(h))
-where causal_bonus = 1.0 if h.status == CAUSAL_LAW else 0.0
-```
-
-**Blackboard invariants enforced at runtime:**
-- No agent reads from another agent's private state (none exists)
-- All mutations are timestamped and logged to `agent_call_log`
-- Budget is decremented by the blackboard, not by agents
-- `final_verdict` transitions are irreversible once set
-
----
-
-### 🔧 Domain-Specific Language
-
-The DSL (`council.py`, ~100 lines) defines 10 reversible, compositional transformation primitives over integer grids in the 10-color ARC color space:
-
-```python
-DSL.PRIMITIVES = {
- "rotate90": np.rot90(g, 1),
- "rotate180": np.rot90(g, 2),
- "rotate270": np.rot90(g, 3),
- "mirror_h": np.fliplr(g),
- "mirror_v": np.flipud(g),
- "gravity_down": # non-bg cells settle downward per column
- "gravity_up": # non-bg cells settle upward per column
- "majority_recolor": # all non-bg → most frequent color
- "sort_by_size": # objects reordered left→right by ascending cell count
- "identity": # no-op baseline
-}
-```
-
-Programs are sequences of primitive names serialized as `"prim_a → prim_b → prim_c"`. The **MDL score** is simply the program length — shortest programs that generalize win. This implements the **Minimum Description Length principle**: if two programs fit all training examples equally well, prefer the simpler one.
-
-```python
-DSL.execute(grid, ["gravity_down", "mirror_h"]) # returns np.ndarray
-DSL.mdl_score(["gravity_down", "mirror_h"]) # returns 2.0
-```
-
-**Design rationale for primitives:**
-Each primitive preserves semantic content (colors, cell counts) while transforming spatial structure. This ensures that the program search space is tractable (10^6 programs of length ≤6), verifiable in O(H×W) per training pair, and interpretable to humans.
-
----
-
-### 🤖 The Nine-Agent Council
-
-Each agent implements a distinct epistemological function. No agent duplicates another's role.
-
----
-
-#### Agent 1 — Perceiver 👁️
-**Role:** Object segmentation → WorldState
-
-Performs connected-component analysis on raw integer grids to extract discrete objects. Each object is characterized by `{id, color, cells:[(r,c),...], bbox:(r0,c0,r1,c1), size}`. The Perceiver is the only agent with direct read access to the raw grid; all subsequent agents operate on the structured `WorldState`.
-
-**Invariant:** Invoked first in every episode; re-invoked after Philosopher reframing.
-
----
-
-#### Agent 2 — Dreamer 💭
-**Role:** Stochastic hypothesis generation
-
-Generates `K=8` candidate output grids per invocation by sampling DSL primitive compositions:
-
-1. Sample program length uniformly from `{1, 2, 3}`
-2. With 50% probability, prepend a primitive biased by prior art hints from the Archivist
-3. Fill remaining slots with uniform samples from `DSL.PRIMITIVES`
-4. Deduplicate consecutive identical primitives
-5. Execute on all training inputs; compute confidence as fraction of training pairs correctly produced
-6. Push to Blackboard if confidence ≥ 0.3
-
-**Exploration bias:** Primitives in the skill library receive 10× higher sampling weight.
-
----
-
-#### Agent 3 — Scientist 🔬
-**Role:** MDL-optimal program synthesis
-
-Runs `MCTS_ROLLOUTS=60` random program searches per invocation. For each candidate program, checks generalization across all training pairs. If a program perfectly generalizes, scores it with MDL. Attaches the shortest generalizing program to the top hypothesis and marks it `TESTING`.
-
-```python
-def _generalizes(program, task) -> bool:
- for inp, expected in task.train_pairs:
- produced = DSL.execute(inp, program)
- if produced.shape != expected.shape: return False
- if not np.array_equal(produced, expected): return False
- return True
-```
-
-**Key insight:** The Scientist inverts typical program synthesis. Rather than searching for outputs given a program, it searches for a program that explains the Dreamer's already-imagined output. This dramatically narrows the search space.
-
----
-
-#### Agent 4 — Skeptic 🔴
-**Role:** Adversarial falsification (Popperian)
-
-Applies `MUTATION_COUNT=12` structural mutations to training inputs and checks program stability. Mutations include: color swaps, noise injection, grid shifts. A single failing mutation falsifies the hypothesis and logs a `ContradictionEntry`.
-
-```python
-# If mutation of a well-defined input causes shape mismatch under a supposedly general program:
-entry = ContradictionEntry(
- hypothesis_id=top_h.id,
- failure_mode="shape_mismatch_under_mutation",
- agent="Skeptic"
-)
-```
-
-The Skeptic embodies Karl Popper's demarcation criterion: a hypothesis is scientific only if it is falsifiable and has survived falsification attempts.
-
----
-
-#### Agent 5 — Philosopher 🏛️
-**Role:** Ontological reframing
-
-When the Curiosity Engine detects persistent failure, the Philosopher challenges the Perceiver's fundamental object decomposition. It proposes two alternative segmentation schemes:
-
-- **Revision 0 → 1:** Include the background color as an explicit object (relevant when the task manipulates "holes" or "enclosed regions")
-- **Revision 1 → 2:** Merge all cells of the same color into a single object regardless of connectivity (relevant for color-based counting tasks)
-
-After reframing, the Perceiver is immediately re-invoked with the new `WorldState`. This allows the council to discover that the task requires treating "colors as objects" rather than "shapes as objects."
-
----
-
-#### Agent 6 — CausalReasoner 🕸️
-**Role:** Counterfactual causal verification
-
-Tests whether the Scientist's program is a **causal law** or a **spurious coincidence** via `COUNTERFACTUAL_COUNT=8` single-variable interventions:
-
-```python
-def _intervene(grid) -> np.ndarray:
- # Single-variable counterfactual: change one cell's color
- out[r, c] = random_color_not_equal_to(out[r, c])
- return out
-```
-
-**Verdict logic:**
-- If `program(intervened_input) == program(original_input)` despite `intervened_input ≠ original_input`, the program is **insensitive** to its inputs → `COINCIDENCE`
-- Verdict `CAUSAL_LAW` requires sensitivity in fewer than half of counterfactual tests
-
-This prevents the classic failure mode where a program accidentally satisfies all training examples by returning a constant or near-constant output.
-
----
-
-#### Agent 7 — CuriosityEngine ⚡
-**Role:** Active Inference / surprise monitoring
-
-Computes prediction error as normalized L2 distance between the top hypothesis's predicted grid and the ground-truth test output. Tracks error over time via `SurpriseTracker`:
-
-```python
-error_t = ‖predicted - actual‖₂ / (H × W × 9) # normalized to [0, 1]
-```
-
-**Plateau detection:** If `|error[t] - error[t-2]| < ε` for two consecutive observations, the tracker declares `is_plateauing=True`.
-
-**Directives issued on plateau:**
-
-| Condition | Directive |
-|-----------|-----------|
-| ≥3 falsified hypotheses | `PHILOSOPHER_REFRAME` |
-| Odd intervention count | `DREAMER_EXPLORE_LOW_CONFIDENCE` |
-| Even intervention count | `SCIENTIST_EXTEND_SEARCH` |
-
-This implements **active inference** in the sense of Friston: the agent drives itself toward states of minimal free energy (prediction error), not by passively waiting for better hypotheses, but by actively redirecting the council's exploration.
-
----
-
-#### Agent 8 — Metacognitor 🧭
-**Role:** Meeting chair, agenda setter, convergence arbiter
-
-The Metacognitor has no domain knowledge. Its sole function is **meta-cognitive orchestration**:
-
-1. Reads `curiosity_directive` from the CuriosityEngine
-2. Reads hypothesis stack state from the Blackboard
-3. Constructs an ordered `agenda` of agents to invoke this round
-4. When budget is critical, triggers the convergence vote:
-
-```python
-# Convergence vote mechanism
-winner = max(candidates,
- key=lambda h: h.confidence * (1.0 if h.causal_verdict == "CAUSAL_LAW" else 0.5))
-
-if winner.confidence >= 0.30:
- bb.declare_answer(winner.grid, "solved", "Council")
-```
-
-**Agenda examples:**
-
-| State | Agenda |
-|-------|--------|
-| No hypotheses | `[Dreamer, Scientist, Skeptic, CausalReasoner]` |
-| Top hypothesis PENDING | `[Scientist, Skeptic, CausalReasoner]` |
-| Top hypothesis FALSIFIED | `[Dreamer, Scientist, Skeptic, CausalReasoner]` |
-| Curiosity says PHILOSOPHER_REFRAME | `[Philosopher, Perceiver, Dreamer, Scientist, Skeptic, CausalReasoner]` |
-
----
-
-#### Agent 9 — Archivist 📚
-**Role:** Episodic memory, skill extraction, prior art injection
-
-The Archivist bridges episodes. At episode start, it retrieves `k=3` most similar past episodes (matched by prior overlap) and injects their winning programs as hints for the Dreamer. At episode end, it archives the full episode record and extracts skill primitives:
-
-```python
-for prim_name in winning_program.split(" → "):
- skill_lib.add_skill(SkillPrimitive(
- name=prim_name,
- origin_task_id=task.task_id,
- description=f"Used to solve {task.transformation_description}"
- ))
-```
-
-The skill library implements a form of **Bayesian program learning**: primitives that appear in successful programs are biased toward in future hypothesis generation, producing a soft meta-learning signal without gradient descent.
-
----
-
-### 🔄 Council Meeting Protocol
-
-The main loop (`council.py → Council.solve()`) is a **streaming generator** — every agent action yields a `Blackboard.snapshot()` for real-time dashboard rendering:
-
-```
-PHASE 0: ORIENTATION
- └─ Perceiver.perceive(test_input)
- └─ Archivist.inject_hints()
-
-PHASE 1: FIRST IMAGINATION
- └─ Dreamer.imagine(K=8 hypotheses)
-
-PHASE 2: MAIN DEBATE LOOP [repeat until solved or MAX_ROUNDS=30]
- └─ Metacognitor.arbitrate() → sets agenda
- └─ For each agent in agenda:
- ├─ Scientist.synthesize() → attaches program to top hypothesis
- ├─ Skeptic.challenge() → falsifies or passes
- │ └─ if fail: CuriosityEngine.observe() → sets directive → break round
- ├─ CausalReasoner.verify() → CAUSAL_LAW or COINCIDENCE
- │ └─ if CAUSAL_LAW: declare_answer("solved") → exit
- ├─ Dreamer.imagine() → if directed
- └─ Philosopher.reframe() → if directed → re-invoke Perceiver
- └─ CuriosityEngine.observe() [end-of-round]
-
-PHASE 3: ARCHIVAL
- └─ Archivist.archive()
-```
-
-**Budget accounting:** Each round costs 1 budget unit. Budget runs from 0 to 100; `budget_critical` triggers at 85. This forces convergence under uncertainty.
-
----
-
-## 🌌 The ARC-AGI-2 Universe
-
-`universe.py` implements a **zero-cheat procedural task generator**. "Zero-cheat" means no task is ever repeated — each task receives a unique fingerprint (`SHA-256(transformation_composition)`), and the Blackboard enforces that the same fingerprint cannot appear twice in a session.
-
-### Core Knowledge Priors
-
-The task generator is grounded in **Spelke's Core Knowledge Theory** — the set of innate concepts human infants possess by ~6 months of age:
-
-| Prior | Enum | Description |
-|-------|------|-------------|
-| OBJECTNESS | `Prior.OBJECTNESS` | Discrete objects persist through transformation |
-| NUMEROSITY | `Prior.NUMEROSITY` | Quantities are conserved or predictably altered |
-| SYMMETRY | `Prior.SYMMETRY` | Spatial invariances constrain valid transformations |
-| CAUSALITY | `Prior.CAUSALITY` | Transformations exhibit consistent causal structure |
-| CONTAINMENT | `Prior.CONTAINMENT` | Objects may contain other objects |
-| GRAVITY | `Prior.GRAVITY` | Non-background elements settle toward edges |
-| GOAL_DIRECTEDNESS | `Prior.GOAL` | Transformations optimize toward target configurations |
-
-These priors are combined compositionally at difficulty level `L` by selecting `L` priors and chaining their corresponding transformation primitives.
-
-### Difficulty Levels
-
-| Level | Priors | Task Example | Complexity |
-|-------|--------|-------------|------------|
-| L1 | 1 | `mirror_v` | Trivially generalizable |
-| L2 | 2 | `gravity_down → mirror_h` | Requires composition |
-| L3 | 3 | `rotate90 → sort_by_size → majority_recolor` | Multi-step reasoning |
-| L4 | 4 | 4-primitive chain | Expert-level ARC difficulty |
-| L5 | 4+ | Chained dependencies | Frontier-level |
-
-### Task Synthesis Pipeline
-
-```python
-1. Sample priors(difficulty_level) # e.g., [GRAVITY, SYMMETRY]
-2. Compose transform_fn = T₁ ∘ T₂ ∘ ... Tₙ
-3. Generate n_train=3 random input grids # 5×5 to 30×30, 1–15 objects
-4. Produce outputs = [transform_fn(inp) for inp in inputs]
-5. Generate test_input (same prior structure, different grid)
-6. Compute test_output = transform_fn(test_input) # ground truth, hidden from agents
-7. Fingerprint = SHA-256(repr(transform_fn))
-```
-
-Grid specifications:
-- **Dimensions:** 5×5 to 30×30 (procedurally bounded by prior requirements)
-- **Colors:** 10-color ARC standard palette (`{0:black, 1:blue, 2:red, ..., 9:purple}`)
-- **Objects per grid:** 1 to 15
-- **Training pairs:** 3 examples per task
-
----
-
-## 📊 Live Dashboard
-
-The Streamlit dashboard (`LAteNT.py`, ~1880 lines) provides a real-time scientific interface with **7 tabs** and **40+ visualizations** across **10 observatory sections**:
-
-| Tab | Contents |
-|-----|----------|
-| 🏛️ **Council Chamber** | Live agent dialogue stream • Answer comparison grid • Cell accuracy badge • Hypothesis breakdown |
-| ⚡ **Surprise Metric** | Prediction error curve • Resolution stats • Convergence status |
-| 🔬 **Program Inspector** | Discovered DSL rule • MDL score • Step-by-step execution trace • Applied to training example |
-| 🔴 **Skeptic's Dossier** | Contradiction log • All hypotheses table • Falsification breakdown |
-| 📉 **Generalization Curve** | Rounds-to-solve over time • Rolling mean • Solve rate trend |
-| 📚 **Skill Library** | Usage bar chart (builtin vs discovered) • Full skill dataframe with pseudocode |
-| 🔭 **Observatory** | 10 sections, 40+ charts (full breakdown below) |
-
-### Observatory Sections
-
-```
-A — Hypothesis Manifold
- A1 Confidence Cascade (horizontal bar, plasma colormap)
- A2 Status Mosaic (donut chart)
- A3 MDL Score Waterfall (cool colormap)
- A4 Contradiction Pressure (polar bar chart)
- A5 Confidence × Age Heatmap (inferno scatter)
-
-B — Free Energy & Surprise
- B1 Free Energy Landscape (segmented gradient fill)
- B2 Surprise Gradient dE/dt (green/red bars)
- B3 Active Inference Phase Space (E[t-1] vs E[t])
- B4 Resolution Speedometer (polar gauge)
- B5 Entropy Reduction Timeline (cumulative area)
-
-C — Agent Council Activity
- C1 Agent Brain Heatmap (9 agents × N rounds)
- C2 Council Speaking Clock (polar bar)
- C3 Agent Activation Gantt (horizontal timeline)
- C4 Dialogue Density Wave (stacked area)
- C5 Agent Co-activation Matrix (viridis)
-
-D — Skill Meme Grid
- D1 Skill Meme Grid (HSV pixel mosaic, brightness ∝ usage)
- D2 Skill Usage Heatmap (YlOrRd)
- D3 Success Rate Radar (top 8 skills)
- D4 Discovery Timeline (builtin vs emergent)
- D5 Skill Gravity Well (bubble chart)
-
-E — Program Structure Analysis
- E1 Program Length Distribution
- E2 MDL vs Confidence Scatter (magma)
- E3 Primitive Co-occurrence Matrix (RdYlGn)
- E4 Winning Program Spotlight (colorized pipeline)
- E5 Confidence Distribution (histogram)
-
-F — Causal Reasoning Engine
- F1 Causal Law vs Coincidence bar chart
- F2 Falsification Heatmap (Agent × failure_mode)
- F3 Causal Confidence Scatter
- F4 Skeptic Contradiction Spiral (polar)
- F5 Causal Law Rate Over Hypotheses (cumulative %)
-
-G — World State & Perception
- G1 Object Color Distribution (polar)
- G2 Object Size Histogram
- G3 Color Transition Matrix Input→Output (plasma, annotated)
- G4 Philosopher Revision Depth Gauge
- G5 Object Bounding Box Map
-
-H — Multi-Episode Meta-Learning
- H1 Rounds to Solve Learning Curve
- H2 Cumulative Solve Rate
- H3 Difficulty vs Rounds Scatter
- H4 Final Surprise per Episode
- H5 Budget Efficiency per Episode
-
-I — Curiosity Engine Deep Dive
- I1 Directive Frequency
- I2 Plateau Detection on Surprise
- I3 Surprise Spectrum (1D heartbeat colorbar)
- I4 Curiosity Engine Stats box
- I5 Free Energy Convergence (log scale)
-
-J — Emergent Intelligence Metrics
- J1 GI Progress Multi-Ring Gauge (5 metrics)
- J2 Intelligence Fingerprint Radar (8 dimensions)
- J3 Metacognitor Activity Heatmap
- J4 Council Consensus Heat
- J5 System Complexity Score Timeline + Composite GI Score
-```
-
-**Composite General Intelligence Score** (Section J5) is a session-level aggregate:
-```
-GI_score = mean([solve_rate, skill_reuse, surprise_decay,
- causal_law_rate, budget_efficiency, round_efficiency])
-```
-
----
-
-## 📈 Experimental Results
-
-All results reproducible with `seed=11290`.
-
-### Overall Performance
-
-| Metric | Value |
-|--------|-------|
-| **Solve Rate** | **18/18 (100%)** |
-| Avg. Rounds to Solve | 15.9 ± 3.2 |
-| Budget Utilization | 53% (15.9 / 30) |
-| Worst Case | 23 rounds (T0013, L3) |
-| Best Case | 14 rounds (T0001, T0005, T0006, T0007) |
-
-### Per-Task Results
-
-```
-Task ID Difficulty Rounds Verdict
-───────────────────────────────────────────────────────────
-T0000_7046e3eef9c38598 L1 15 ✅ SOLVED
-T0001_f0ff7e211c60a023 L1 14 ✅ SOLVED
-T0002_1e94c74b1c4cd52a L2 15 ✅ SOLVED
-T0003_44f0bbbd3ae17296 L2 15 ✅ SOLVED
-T0004_a633ef1a1bbaadbb L2 15 ✅ SOLVED
-T0005_a31a626e619c8024 L2 14 ✅ SOLVED
-T0006_9224a01a0b0e5d79 L2 14 ✅ SOLVED
-T0007_3eb91af5049b67fe L2 14 ✅ SOLVED
-T0008_efb451d8fa4c9405 L3 15 ✅ SOLVED
-T0009_5b46400abc669d5a L3 15 ✅ SOLVED
-T0010_d4dad90e496df51f L3 18 ✅ SOLVED
-T0011_4689eee23a368e7b L3 17 ✅ SOLVED
-T0012_63a7c7a103865c22 L3 15 ✅ SOLVED
-T0013_4fba135a50799a8e L3 23 ✅ SOLVED ← hardest
-T0014_35ae0bce04df14d6 L3 19 ✅ SOLVED
-T0015_ca91975b5d86d463 L3 18 ✅ SOLVED
-T0016_7bde9a0ea5fce54e L3 16 ✅ SOLVED
-T0017_a8005ea73b232c2d L3 15 ✅ SOLVED
-```
-
-### Skill Library Growth
-
-| Metric | Value |
-|--------|-------|
-| Initial builtin primitives | 10 |
-| Total skills after 18 episodes | 17 |
-| Emergent (discovered) skills | 7 |
-| Transfer rate (emergent → reused) | 5/7 (71%) |
-
-**Top 5 skills by usage:**
-1. `majority_recolor` — 6 uses, **100% success rate**
-2. `gravity_down` — 4 uses, **100% success rate**
-3. `gravity_up` — 4 uses, **100% success rate**
-4. `mirror_v` — 3 uses, **100% success rate**
-5. `sort_by_size` — 3 uses, **100% success rate**
-
-### Agent Contribution (avg invocations/task)
-
-| Agent | Avg. Invocations | Function |
-|-------|-----------------|----------|
-| Dreamer | 6.2 | Hypothesis generation |
-| Metacognitor | 6.2 | Agenda + convergence |
-| Scientist | 6.1 | Program synthesis |
-| Skeptic | 6.0 | Falsification |
-| CuriosityEngine | 6.0 | Surprise tracking |
-| CausalReasoner | 5.8 | Causal validation |
-| Philosopher | 1.8 | Ontological reframing |
-| Perceiver | 1.4 | Segmentation |
-| Archivist | 1.0 | End-of-task archival |
-
----
-
-## ⚙️ Complexity Analysis
-
-| Component | Time Complexity | Space Complexity |
-|-----------|----------------|-----------------|
-| Perceiver (segmentation) | O(H × W) | O(n_objects) |
-| Dreamer (hypothesis gen) | O(K × n_train × L) | O(K) |
-| Scientist (MDL search) | O(\|DSL\|^L × n_train) | O(\|DSL\|^L) |
-| Skeptic (falsification) | O(n_hyp × n_train) | O(1) streaming |
-| CausalReasoner | O(C × L) | O(C) |
-| **Session total** | O(MAX_ROUNDS × \|DSL\|^L × n_train) | O(MAX_HYPO + EPISODE_CAP) |
-
-For typical parameters (`MAX_ROUNDS=30, |DSL|=10, L=6, n_train=3`):
-~100M primitive operations per task, all in NumPy vectorized ops.
-
----
-
-## 📁 Codebase Tour
-
-```
-Latent-Consensus-Manifold/
-├── LAteNT.py # Main Streamlit dashboard (~1880 lines)
-│ ├── ARC_CMAP # ARC 10-color palette → ListedColormap
-│ ├── _render_grid() # Dark-themed ARC grid renderer
-│ ├── _grid_fig() # Multi-panel figure builder
-│ ├── _agent_html() # Colored agent dialogue HTML
-│ ├── _verdict_badge() # Status badge renderer
-│ ├── _winning_program() # Extracts accepted program from snapshot
-│ ├── _answer_grid() # Re-executes program on test input
-│ ├── Sidebar # Difficulty selector + session stats + export
-│ ├── Tab 1–6 # Core analysis tabs
-│ └── Tab 7 # Observatory (sections A–J)
-│
-├── council.py # The 9-agent council (~962 lines)
-│ ├── DSL # Primitive interpreter + program execution
-│ ├── AgentResult # Typed return value dataclass
-│ ├── Perceiver # Connected-component segmentation
-│ ├── Dreamer # Stochastic hypothesis generator
-│ ├── Scientist # MDL program synthesizer (MCTS-style)
-│ ├── Skeptic # Adversarial falsifier (Popperian)
-│ ├── Philosopher # Ontological reframer
-│ ├── CausalReasoner # Counterfactual verifier
-│ ├── CuriosityEngine # Active inference / surprise tracker
-│ ├── Metacognitor # Meeting chair + convergence vote
-│ ├── Archivist # Episode memory + skill extraction
-│ └── Council # Main orchestration loop (streaming generator)
-│
-├── memory.py # Shared state substrate (~565 lines)
-│ ├── HypothesisStatus # Enum: PENDING→TESTING→{FALSIFIED,CAUSAL_LAW,ACCEPTED}
-│ ├── Hypothesis # Candidate program + grid + metadata
-│ ├── ContradictionEntry # Falsification event record
-│ ├── WorldState # Perceiver output: segmented objects
-│ ├── Blackboard # Central mutable state (all agent I/O)
-│ ├── EpisodeMemory # Persistent episode archive (FIFO, 500 cap)
-│ ├── DSLSkillLibrary # Primitive library with usage tracking
-│ └── SurpriseTracker # Rolling L2 error + plateau detector
-│
-├── universe.py # Procedural task generator (~503 lines)
-│ ├── Prior # Enum: 7 core knowledge priors
-│ ├── DifficultyLevel # Enum: L1–L5
-│ ├── GridObject # Segmented object dataclass
-│ ├── ARCTask # Task specification + train/test pairs
-│ ├── GridTransforms # All atomic transformation implementations
-│ └── Universe # Zero-cheat procedural task factory
-│
-├── meta_learner.py # (in development)
-├── latent_dictionary.py # (in development)
-│
-├── requirements.txt
-├── plan.md # True AGI research roadmap
-├── LICENSE # Apache 2.0
-└── Results Archive/
- ├── agi_session_11290.json # 18-task session (seed 11290)
- ├── agi_session_11290_latest.json
- ├── agi_session_70290.json # 18-task session (seed 70290)
- ├── general_intelligence_session_6262.json
- └── readme.md # Full technical paper
-```
-
----
-
-## 🚀 Installation & Usage
-
-### Requirements
-
-```
-Python >= 3.10
-numpy >= 1.24.0
-scipy >= 1.11.0
-streamlit >= 1.32.0
-matplotlib >= 3.7.0
-pandas >= 2.0.0
-```
-
-### Install
-
-```bash
-git clone https://github.com/Devanik21/Latent-Consensus-Manifold.git
-cd Latent-Consensus-Manifold
-pip install -r requirements.txt
-```
-
-### Launch Dashboard
-
-```bash
-streamlit run LAteNT.py
-```
-
-Open `http://localhost:8501`. Select a difficulty level in the sidebar and click **⚡ Run Council**.
-
-### Scripted Usage
-
-```python
-from universe import Universe, DifficultyLevel
-from council import Council
-
-# Reproducible session
-universe = Universe(seed=11290)
-council = Council(seed=11290)
-
-# Generate and solve a task
-task = universe.generate_task(DifficultyLevel.L2)
-print(f"Task: {task.task_id}")
-print(f"Rule (hidden from agents): {task.transformation_description}")
-
-# council.solve() is a streaming generator
-final_snapshot = None
-for snapshot in council.solve(task):
- final_snapshot = snapshot
- print(f"[Round {snapshot['round']}] {snapshot['final_verdict']}")
-
-print(f"\nVerdict : {final_snapshot['final_verdict']}")
-print(f"Rounds : {final_snapshot['round']}")
-print(f"Budget : {final_snapshot['budget_used']}/100")
-
-# Session-level meta-learning stats
-stats = council.stats()
-print(f"Skills discovered: {stats['skill_library_size']}")
-print(f"Avg rounds: {stats['avg_rounds']}")
-```
-
-### Export Session Data
-
-The dashboard provides a **💾 Download Session Data** button in the sidebar after ≥1 task run. The exported JSON contains:
-
-```json
-{
- "seed": 11290,
- "tasks_run": 18,
- "solved": 18,
- "avg_rounds": 15.9,
- "skills": [...],
- "generalization": [...],
- "cumulative_dialogue_logs": [...]
-}
-```
-
-### Self-Test
-
-```bash
-python council.py
-# Runs 3 tasks (L1, L2, L1) and prints per-task results
-```
-
----
-
-## 🗺 Roadmap — True AGI Plan
-
-The current system is an explicitly symbolic baseline. The research roadmap targets a system where **nothing is defined — agents must discover everything from data alone**.
-
-### Phase 1 — Latent Transformation Learning *(3–4 weeks)*
-Replace the handcrafted DSL with **learned transformation embeddings**:
-- Collect 1000+ `(input→output)` pairs across diverse domains
-- Train a transformation autoencoder: each transformation → point in continuous latent space
-- Agents sample from this space rather than from a fixed primitive list
-- Result: agents discover "rotation-like" operations without being told what rotation is
-
-### Phase 2 — Abstraction Discovery Through Pure Induction *(4–5 weeks)*
-Remove the DSL interpreter entirely. Replace program synthesis with **latent space search**:
-- The Scientist searches the learned transformation latent space (no discrete primitives)
-- The Dreamer samples latent transformations and imagines hypothetical outputs
-- Discovery of novel compositions through interpolation in latent space
-- No `IF-THEN`. No hardcoded logic. Pure causal inference in continuous space.
-
-### Phase 3 — Cross-Domain Generalization Test *(3 weeks)*
-The real measure of AGI:
-- Train on Task Domain A (30 tasks)
-- Validate on Domain B (30 tasks, different visual properties)
-- Test on Domain C (30 tasks, completely unseen transformation types)
-- Target: **50%+ on Domain C** without having seen that domain's pattern types
-
-### Phase 4 — Meta-Learning (Learning to Learn Faster) *(3–4 weeks)*
-- Track episode-level discovery speed
-- Train a meta-learner to predict which hypothesis directions are most promising
-- The meta-learner shapes the Dreamer and Scientist purely from episode history
-- Result: system improves its own discovery process through meta-reasoning, not programmer-defined improvements
-
-### Phase 5 — Scientific Validation & Emergence Analysis *(2 weeks)*
-- Does the system discover novel transformation types not in training data?
-- Do agents develop implicit strategies without explicit programming?
-- Can the learned transformation latent space be visualized? Do similar operations cluster?
-- Publish findings: *"Learning Transformation Abstractions Without Explicit Rules"*
-
-**Timeline:** 15–17 weeks to top-0.1% territory.
-**Success threshold:** 50%+ cross-domain transfer + demonstrated meta-learning improvement + published emergence analysis.
-
-**Ultimate goal:** A true AGI capable of solving any problem a human can solve — with infinite tools, skills, knowledge, and memories discovered autonomously.
-
----
-
-## ⚠️ Limitations & Honest Assessment
-
-This system achieves 100% solve rate on a task distribution it was designed for. That is not the same as general reasoning capability.
-
-**1. DSL Expressiveness Ceiling**
-The 10 builtin primitives cover a restricted subset of abstract reasoning. Tasks requiring conditional logic, counting, connectivity-based transformations, or arbitrary color mappings cannot be solved by construction. The 100% solve rate reflects task-distribution alignment.
-
-**2. Program Search Tractability**
-Scientist search is capped at programs of length ≤6. Solutions requiring longer compositions will not be found within budget.
-
-**3. Hypothesis Stack Saturation**
-Under sustained Dreamer output (8 hypotheses × 20 rounds = 160 generated vs. stack cap of 50), older hypotheses with initially low confidence are discarded. This may cause premature elimination of correct-but-initially-uncertain hypotheses.
-
-**4. Zero-Shot Regime**
-Without training examples, confidence computation is undefined. The system would degrade to random hypothesis sampling.
-
-**5. Transfer Scope**
-The skill library enables transfer within the same task distribution. Generalization to fundamentally different task types (e.g., trained on rotation-type tasks, tested on connectivity-based tasks) is constrained by DSL expressiveness.
-
-**6. Causal Reasoning Depth**
-Single-variable counterfactual testing (changing one cell) is a weak causal intervention. Real causal graph discovery requires structured interventions at the object level, which the current CausalReasoner does not implement.
-
----
-
-## 📖 References
-
-**Program Synthesis**
-- Gulwani, S. (2015). Dimensions in Program Synthesis. *PPLJ*.
-- Solar-Lezama, A. (2008). Program Synthesis by Sketching. *PhD dissertation, UC Berkeley*.
-
-**Multi-Agent Systems**
-- Stone, P., & Veloso, M. (2000). Multiagent systems: A survey from an AI perspective. *Autonomous Robots, 8(3)*, 345–383.
-- Newell, A. (1962). Some problems of basic organization in problem-solving programs. *Self-Organizing Systems*.
-
-**Active Inference & Free Energy**
-- Friston, K., et al. (2017). Active inference and learning. *Neuroscience & Biobehavioral Reviews*.
-- Friston, K. (2010). The free-energy principle: a unified brain theory? *Nature Reviews Neuroscience, 11(2)*, 127–138.
-
-**Causal Inference**
-- Pearl, J. (2009). Causality: Models, Reasoning, and Inference. *Cambridge University Press*.
-- Peters, J., Janzing, D., & Schölkopf, B. (2017). Elements of Causal Inference. *MIT Press*.
-
-**Minimum Description Length**
-- Rissanen, J. (1978). Modeling by shortest data description. *Automatica, 14(5)*, 465–471.
-- Grünwald, P. (2007). The Minimum Description Length Principle. *MIT Press*.
-
-**ARC Benchmark**
-- Chollet, F. (2019). On the Measure of Intelligence. *arXiv:1911.01547*.
-
-**Core Knowledge Theory**
-- Spelke, E. S., & Kinzler, K. D. (2007). Core knowledge. *Developmental Science, 10(1)*, 89–96.
-
-**Falsificationism**
-- Popper, K. (1959). The Logic of Scientific Discovery. *Hutchinson & Co*.
-
----
-
-## 🔗 Contact
-
-
-
-**Devanik**
-B.Tech ECE '26 • National Institute of Technology Agartala
-Samsung Convergence Software Fellowship (Grade I) • Indian Institute of Science
-
-
-
-[](https://github.com/Devanik21)
-[](https://twitter.com/devanik2005)
-[](mailto:devanik@iisertirupati.ac.in)
-
-
-
-*This work represents independent research conducted during the Samsung Convergence Software Fellowship at the Indian Institute of Science. All code, experimental data, and analysis are made available for academic and research purposes.*
-
-
-
----
-
-
-
-**License:** [Apache 2.0](LICENSE) • **Last Updated:** March 2026
-
-*Built with deliberate constraints — not to beat the leaderboard, but to understand reasoning itself.*
-
-
diff --git a/Misc/002_LCM/versions/v1.md b/Misc/002_LCM/versions/v1.md
deleted file mode 100644
index af65bbe..0000000
--- a/Misc/002_LCM/versions/v1.md
+++ /dev/null
@@ -1 +0,0 @@
-# Version 1.0
diff --git a/Misc/003_XV/README.md b/Misc/003_XV/README.md
deleted file mode 100644
index c42cd85..0000000
--- a/Misc/003_XV/README.md
+++ /dev/null
@@ -1,462 +0,0 @@
-#
-
-
-
-🪻 **An Elegant Multimodal AI Framework for Visual Understanding & Educational Synthesis**
-
-
-
-[](https://github.com/Devanik21)
-[](https://www.linkedin.com/in/devanik/)
-[](https://x.com/devanik2005)
-
-
-
-```
- ╔═══════════════════════════════════════════╗
- ║ Image Analysis → AI Processing ║
- ║ ↓ ║
- ║ Semantic Understanding ║
- ║ ↓ ║
- ║ Knowledge Synthesis ║
- ║ ↓ ║
- ║ Educational Content Generation ║
- ╚═══════════════════════════════════════════╝
-```
-
-
-
-### 🌐 **Social & Professional Profiles**
-
-| Platform | Link |
-|----------|------|
-| **GitHub** | [github.com/Devanik21](https://github.com/Devanik21) |
-| **LinkedIn** | [linkedin.com/in/devanik](https://www.linkedin.com/in/devanik/) |
-| **X (Twitter)** | [@devanik2005](https://x.com/devanik2005) |
-
-
-
----
-
-## 📝 **Usage Examples**
-
-### Example 1: Plant Identification
-```
-Input: Image of an unknown leaf
-Output:
-- Species: Acer palmatum (Japanese Maple)
-- USDA Hardiness: 5-8
-- Photosynthesis Type: C3 (typical deciduous)
-- Seasonal Pattern: Deciduous, autumn foliage
-```
-
-### Example 2: Educational Analysis
-```
-Input: Diagram of cellular mitosis
-Output:
-- Identified Phases: Prophase, Metaphase, Anaphase, Telophase
-- Key Structures: Spindle fibers, centromeres, sister chromatids
-- Biological Significance: Genetic material replication mechanism
-- Flashcard Generated: Q: "What is the purpose of metaphase?"
- A: "Chromosomes align at metaphase plate..."
-```
-
-### Example 3: Historical Landmark
-```
-Input: Photograph of Angkor Wat
-Output:
-- Location: Siem Reap, Cambodia
-- Constructed: ~1113-1150 CE (Khmer Empire)
-- Architectural Style: Khmer architecture with Hindu temple influences
-- UNESCO Status: World Heritage Site (1992)
-- Cultural Significance: Symbol of Cambodian national identity
-```
-
----
-
-## 🎓 **Educational Methodology**
-
-Xylia supports multiple learning paradigms:
-
-1. **Spaced Repetition** (Ebbinghaus Curve)
- - Flashcards optimized for retention
- - Interval scheduling based on difficulty
-
-2. **Active Recall**
- - Q&A mode forces knowledge retrieval
- - Immediate feedback on accuracy
-
-3. **Multimodal Learning**
- - Visual analysis + auditory narration
- - Dual-channel information encoding
- - Increased retention through modality diversity
-
-4. **Contextual Understanding**
- - Landmark, botanical, and object contextualization
- - Real-world application grounding
- - Semantic relationship mapping
-
----
-
-## 🌟 **Future Roadmap**
-
-- [ ] Real-time video stream analysis
-- [ ] Multi-object tracking & relationship extraction
-- [ ] Advanced AR visualization
-- [ ] Collaborative study sessions
-- [ ] Custom model fine-tuning
-- [ ] Advanced statistical learning analytics
-- [ ] Integration with educational platforms (Canvas, Blackboard)
-- [ ] Offline mode with local model support
-
----
-
-
-
-## **Acknowledgments**
-
-This project respectfully builds upon:
-- Google Generative AI (Gemini Vision)
-- Streamlit framework
-- The open-source Python ecosystem
-- Botanical & educational communities
-
----
-
-Crafted with precision and botanical inspiration • Xylia © 2026
-
-
-
----
-
-
-
-### Made with 🪻 by [Devanik](https://github.com/Devanik21)
-
-*"Intelligence should be beautiful. Understanding should be elegant."*
-
-
diff --git a/Misc/003_XV/metadata.json b/Misc/003_XV/metadata.json
deleted file mode 100644
index bc0fefc..0000000
--- a/Misc/003_XV/metadata.json
+++ /dev/null
@@ -1,4 +0,0 @@
-{
- "name": "xylia-vision",
- "status": "active"
-}
\ No newline at end of file
diff --git a/Misc/003_XV/notes/evolution.md b/Misc/003_XV/notes/evolution.md
deleted file mode 100644
index 8073517..0000000
--- a/Misc/003_XV/notes/evolution.md
+++ /dev/null
@@ -1 +0,0 @@
-# Evolution
diff --git a/Misc/003_XV/notes/thoughts.md b/Misc/003_XV/notes/thoughts.md
deleted file mode 100644
index b2afb31..0000000
--- a/Misc/003_XV/notes/thoughts.md
+++ /dev/null
@@ -1 +0,0 @@
-# Thoughts
diff --git a/Misc/003_XV/snapshot/README.md b/Misc/003_XV/snapshot/README.md
deleted file mode 100644
index c42cd85..0000000
--- a/Misc/003_XV/snapshot/README.md
+++ /dev/null
@@ -1,462 +0,0 @@
-#
-
-
-
-🪻 **An Elegant Multimodal AI Framework for Visual Understanding & Educational Synthesis**
-
-
-
-[](https://github.com/Devanik21)
-[](https://www.linkedin.com/in/devanik/)
-[](https://x.com/devanik2005)
-
-
-
-```
- ╔═══════════════════════════════════════════╗
- ║ Image Analysis → AI Processing ║
- ║ ↓ ║
- ║ Semantic Understanding ║
- ║ ↓ ║
- ║ Knowledge Synthesis ║
- ║ ↓ ║
- ║ Educational Content Generation ║
- ╚═══════════════════════════════════════════╝
-```
-
-
-
-### 🌐 **Social & Professional Profiles**
-
-| Platform | Link |
-|----------|------|
-| **GitHub** | [github.com/Devanik21](https://github.com/Devanik21) |
-| **LinkedIn** | [linkedin.com/in/devanik](https://www.linkedin.com/in/devanik/) |
-| **X (Twitter)** | [@devanik2005](https://x.com/devanik2005) |
-
-
-
----
-
-## 📝 **Usage Examples**
-
-### Example 1: Plant Identification
-```
-Input: Image of an unknown leaf
-Output:
-- Species: Acer palmatum (Japanese Maple)
-- USDA Hardiness: 5-8
-- Photosynthesis Type: C3 (typical deciduous)
-- Seasonal Pattern: Deciduous, autumn foliage
-```
-
-### Example 2: Educational Analysis
-```
-Input: Diagram of cellular mitosis
-Output:
-- Identified Phases: Prophase, Metaphase, Anaphase, Telophase
-- Key Structures: Spindle fibers, centromeres, sister chromatids
-- Biological Significance: Genetic material replication mechanism
-- Flashcard Generated: Q: "What is the purpose of metaphase?"
- A: "Chromosomes align at metaphase plate..."
-```
-
-### Example 3: Historical Landmark
-```
-Input: Photograph of Angkor Wat
-Output:
-- Location: Siem Reap, Cambodia
-- Constructed: ~1113-1150 CE (Khmer Empire)
-- Architectural Style: Khmer architecture with Hindu temple influences
-- UNESCO Status: World Heritage Site (1992)
-- Cultural Significance: Symbol of Cambodian national identity
-```
-
----
-
-## 🎓 **Educational Methodology**
-
-Xylia supports multiple learning paradigms:
-
-1. **Spaced Repetition** (Ebbinghaus Curve)
- - Flashcards optimized for retention
- - Interval scheduling based on difficulty
-
-2. **Active Recall**
- - Q&A mode forces knowledge retrieval
- - Immediate feedback on accuracy
-
-3. **Multimodal Learning**
- - Visual analysis + auditory narration
- - Dual-channel information encoding
- - Increased retention through modality diversity
-
-4. **Contextual Understanding**
- - Landmark, botanical, and object contextualization
- - Real-world application grounding
- - Semantic relationship mapping
-
----
-
-## 🌟 **Future Roadmap**
-
-- [ ] Real-time video stream analysis
-- [ ] Multi-object tracking & relationship extraction
-- [ ] Advanced AR visualization
-- [ ] Collaborative study sessions
-- [ ] Custom model fine-tuning
-- [ ] Advanced statistical learning analytics
-- [ ] Integration with educational platforms (Canvas, Blackboard)
-- [ ] Offline mode with local model support
-
----
-
-
-
-## **Acknowledgments**
-
-This project respectfully builds upon:
-- Google Generative AI (Gemini Vision)
-- Streamlit framework
-- The open-source Python ecosystem
-- Botanical & educational communities
-
----
-
-Crafted with precision and botanical inspiration • Xylia © 2026
-
-
-
----
-
-
-
-### Made with 🪻 by [Devanik](https://github.com/Devanik21)
-
-*"Intelligence should be beautiful. Understanding should be elegant."*
-
-
diff --git a/Misc/003_XV/snapshot/xylia-vision.zip b/Misc/003_XV/snapshot/xylia-vision.zip
deleted file mode 100644
index 8f0a4b7..0000000
Binary files a/Misc/003_XV/snapshot/xylia-vision.zip and /dev/null differ
diff --git a/Misc/003_XV/versions/v1.md b/Misc/003_XV/versions/v1.md
deleted file mode 100644
index af65bbe..0000000
--- a/Misc/003_XV/versions/v1.md
+++ /dev/null
@@ -1 +0,0 @@
-# Version 1.0
diff --git a/Misc/004_LB/metadata.json b/Misc/004_LB/metadata.json
deleted file mode 100644
index 880f8b9..0000000
--- a/Misc/004_LB/metadata.json
+++ /dev/null
@@ -1,4 +0,0 @@
-{
- "name": "004_LB",
- "status": "active"
-}
\ No newline at end of file
diff --git a/Misc/004_LB/notes/evolution.md b/Misc/004_LB/notes/evolution.md
deleted file mode 100644
index 8073517..0000000
--- a/Misc/004_LB/notes/evolution.md
+++ /dev/null
@@ -1 +0,0 @@
-# Evolution
diff --git a/Misc/004_LB/notes/thoughts.md b/Misc/004_LB/notes/thoughts.md
deleted file mode 100644
index b2afb31..0000000
--- a/Misc/004_LB/notes/thoughts.md
+++ /dev/null
@@ -1 +0,0 @@
-# Thoughts
diff --git a/Misc/004_LB/versions/v1.md b/Misc/004_LB/versions/v1.md
deleted file mode 100644
index af65bbe..0000000
--- a/Misc/004_LB/versions/v1.md
+++ /dev/null
@@ -1 +0,0 @@
-# Version 1.0
diff --git a/Misc/005_TM/metadata.json b/Misc/005_TM/metadata.json
deleted file mode 100644
index 0f701e6..0000000
--- a/Misc/005_TM/metadata.json
+++ /dev/null
@@ -1,4 +0,0 @@
-{
- "name": "005_TM",
- "status": "active"
-}
\ No newline at end of file
diff --git a/Misc/005_TM/notes/evolution.md b/Misc/005_TM/notes/evolution.md
deleted file mode 100644
index 8073517..0000000
--- a/Misc/005_TM/notes/evolution.md
+++ /dev/null
@@ -1 +0,0 @@
-# Evolution
diff --git a/Misc/005_TM/notes/thoughts.md b/Misc/005_TM/notes/thoughts.md
deleted file mode 100644
index b2afb31..0000000
--- a/Misc/005_TM/notes/thoughts.md
+++ /dev/null
@@ -1 +0,0 @@
-# Thoughts
diff --git a/Misc/005_TM/versions/v1.md b/Misc/005_TM/versions/v1.md
deleted file mode 100644
index af65bbe..0000000
--- a/Misc/005_TM/versions/v1.md
+++ /dev/null
@@ -1 +0,0 @@
-# Version 1.0
diff --git a/Misc/006_DTM/metadata.json b/Misc/006_DTM/metadata.json
deleted file mode 100644
index ed3c984..0000000
--- a/Misc/006_DTM/metadata.json
+++ /dev/null
@@ -1,4 +0,0 @@
-{
- "name": "006_DTM",
- "status": "active"
-}
\ No newline at end of file
diff --git a/Misc/006_DTM/notes/evolution.md b/Misc/006_DTM/notes/evolution.md
deleted file mode 100644
index 8073517..0000000
--- a/Misc/006_DTM/notes/evolution.md
+++ /dev/null
@@ -1 +0,0 @@
-# Evolution
diff --git a/Misc/006_DTM/notes/thoughts.md b/Misc/006_DTM/notes/thoughts.md
deleted file mode 100644
index b2afb31..0000000
--- a/Misc/006_DTM/notes/thoughts.md
+++ /dev/null
@@ -1 +0,0 @@
-# Thoughts
diff --git a/Misc/006_DTM/versions/v1.md b/Misc/006_DTM/versions/v1.md
deleted file mode 100644
index af65bbe..0000000
--- a/Misc/006_DTM/versions/v1.md
+++ /dev/null
@@ -1 +0,0 @@
-# Version 1.0
diff --git a/Misc/007_AARGE/metadata.json b/Misc/007_AARGE/metadata.json
deleted file mode 100644
index 59c7703..0000000
--- a/Misc/007_AARGE/metadata.json
+++ /dev/null
@@ -1,4 +0,0 @@
-{
- "name": "007_AARGE",
- "status": "active"
-}
\ No newline at end of file
diff --git a/Misc/007_AARGE/notes/evolution.md b/Misc/007_AARGE/notes/evolution.md
deleted file mode 100644
index 8073517..0000000
--- a/Misc/007_AARGE/notes/evolution.md
+++ /dev/null
@@ -1 +0,0 @@
-# Evolution
diff --git a/Misc/007_AARGE/notes/thoughts.md b/Misc/007_AARGE/notes/thoughts.md
deleted file mode 100644
index b2afb31..0000000
--- a/Misc/007_AARGE/notes/thoughts.md
+++ /dev/null
@@ -1 +0,0 @@
-# Thoughts
diff --git a/Misc/007_AARGE/versions/v1.md b/Misc/007_AARGE/versions/v1.md
deleted file mode 100644
index af65bbe..0000000
--- a/Misc/007_AARGE/versions/v1.md
+++ /dev/null
@@ -1 +0,0 @@
-# Version 1.0
diff --git a/Misc/008_DU/metadata.json b/Misc/008_DU/metadata.json
deleted file mode 100644
index 821c7ea..0000000
--- a/Misc/008_DU/metadata.json
+++ /dev/null
@@ -1,4 +0,0 @@
-{
- "name": "008_DU",
- "status": "active"
-}
\ No newline at end of file
diff --git a/Misc/008_DU/notes/evolution.md b/Misc/008_DU/notes/evolution.md
deleted file mode 100644
index 8073517..0000000
--- a/Misc/008_DU/notes/evolution.md
+++ /dev/null
@@ -1 +0,0 @@
-# Evolution
diff --git a/Misc/008_DU/notes/thoughts.md b/Misc/008_DU/notes/thoughts.md
deleted file mode 100644
index b2afb31..0000000
--- a/Misc/008_DU/notes/thoughts.md
+++ /dev/null
@@ -1 +0,0 @@
-# Thoughts
diff --git a/Misc/008_DU/versions/v1.md b/Misc/008_DU/versions/v1.md
deleted file mode 100644
index af65bbe..0000000
--- a/Misc/008_DU/versions/v1.md
+++ /dev/null
@@ -1 +0,0 @@
-# Version 1.0