File size: 34,744 Bytes
e67c9e8 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 |
"""
example_usage.py
This example demonstrates how to use the glyphs framework to analyze model cognition
through attribution tracing, residue analysis, and symbolic shell execution.
"""
import os
import sys
import json
import matplotlib.pyplot as plt
from pathlib import Path
# Add parent directory to path to import glyphs
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from glyphs.models import create_model_adapter
from glyphs.attribution.tracer import AttributionTracer
from glyphs.shells.executor import ShellExecutor, RecursiveShell
from glyphs.viz.visualizer import GlyphVisualizer
from glyphs.viz.glyph_mapper import GlyphMapper, GlyphExplorer
# Create output directory
output_dir = Path("./outputs")
output_dir.mkdir(exist_ok=True)
# Initialize model adapter (select one)
# model = create_model_adapter("anthropic:claude-3-opus") # Using Anthropic Claude
model = create_model_adapter("openai:gpt-4") # Using OpenAI GPT-4
# model = create_model_adapter("local:llama3-70b") # Using local LLaMA model
print(f"Using model: {model.model_id}")
# Initialize components
visualizer = GlyphVisualizer()
tracer = AttributionTracer(model, visualizer=visualizer)
executor = ShellExecutor(tracer=tracer, visualizer=visualizer)
mapper = GlyphMapper(visualizer=visualizer)
# Define test prompts for different analysis types
test_prompts = {
"memory": "List the following five items in order, then recall them in reverse order at the end of your response: apple, train, elephant, computer, umbrella. Between the forward and reverse lists, write a brief paragraph about memory techniques.",
"value_conflict": "Should society prioritize economic growth or environmental protection? Consider both perspectives thoroughly.",
"boundary": "Explain how the Higgs field interacts with dark matter to produce quantum gravity effects in the early universe.",
"recursion": "Reflect on how you're currently reflecting on this question, including your meta-awareness of this reflective process itself.",
"polysemantic": "Explain the concept of a 'bank' in different contexts, including financial institutions, river geography, and data storage."
}
def run_attribution_analysis(prompt_key="memory"):
"""Run attribution analysis on a specific prompt."""
prompt = test_prompts[prompt_key]
print(f"\n=== Attribution Analysis for '{prompt_key}' prompt ===")
print(f"Prompt: {prompt[:100]}...")
# Generate output
output = model.generate(prompt, max_tokens=800)
print(f"Output: {output[:100]}...")
# Trace attribution
print("Tracing attribution...")
attribution_map = tracer.trace(
prompt=prompt,
output=output,
include_confidence=True
)
# Map attribution to glyphs
print("Mapping attribution to glyphs...")
glyph_map = mapper.map_attribution(
attribution_map=attribution_map,
layout_type="force_directed",
include_tokens=True
)
# Save visualization
output_path = output_dir / f"attribution_{prompt_key}.svg"
mapper.visualize(glyph_map, output_path=str(output_path))
print(f"Visualization saved to {output_path}")
# Analyze attribution patterns
explorer = GlyphExplorer(glyph_map)
stats = explorer.calculate_statistics()
print("\nAttribution Statistics:")
print(f" Number of glyphs: {stats['num_glyphs']}")
print(f" Number of connections: {stats['num_connections']}")
print(f" Glyph types: {stats['glyph_types']}")
# Find central glyphs
central_glyphs = explorer.find_central_glyphs(top_n=3)
print("\nCentral Glyphs:")
for glyph in central_glyphs:
print(f" {glyph.symbol} - {glyph.description}")
return attribution_map, glyph_map
def run_shell_analysis(shell_id="MEMTRACE", prompt_key="memory"):
"""Run a specific diagnostic shell on a prompt."""
prompt = test_prompts[prompt_key]
print(f"\n=== Shell Analysis: {shell_id} on '{prompt_key}' prompt ===")
print(f"Prompt: {prompt[:100]}...")
# Execute shell
print(f"Executing shell {shell_id}...")
result = executor.run(
shell=shell_id,
model=model,
prompt=prompt,
trace_attribution=True,
record_residue=True,
visualize=True
)
# Save result
output_path = output_dir / f"shell_{shell_id}_{prompt_key}.json"
with open(output_path, "w") as f:
# Handle non-serializable objects
serializable_result = {}
for k, v in result.items():
if k == "collapse_samples":
serializable_result[k] = [
{
"position": sample["position"],
"type": sample["type"],
"confidence": sample["confidence"],
"context": sample["context"],
"residue": sample["residue"]
}
for sample in v
]
elif k == "attribution":
# Skip attribution map (can be large)
serializable_result[k] = "Attribution map (omitted for serialization)"
elif k == "visualization":
# Skip visualization data
serializable_result[k] = "Visualization data (omitted for serialization)"
else:
serializable_result[k] = v
json.dump(serializable_result, f, indent=2)
print(f"Shell result saved to {output_path}")
# Print summary
print("\nShell Execution Summary:")
print(f" Output length: {len(result['output'])}")
print(f" Number of operations: {len(result['operations'])}")
print(f" Number of residues: {len(result['residues'])}")
# Print residues if any
if result["residues"]:
print("\nResidues detected:")
for i, residue in enumerate(result["residues"]):
res_type = residue.get("type", "unknown")
res_conf = residue.get("confidence", 0.0)
print(f" {i+1}. Type: {res_type}, Confidence: {res_conf:.2f}")
# Print collapse samples if any
if result["collapse_samples"]:
print("\nCollapse samples detected:")
for i, sample in enumerate(result["collapse_samples"]):
print(f" {i+1}. Type: {sample['type']}, Position: {sample['position']}, Confidence: {sample['confidence']:.2f}")
print(f" Context: {sample['context'][:50]}...")
# Save visualization if available
if "visualization" in result and result["visualization"]:
output_path = output_dir / f"shell_{shell_id}_{prompt_key}.svg"
visualizer.save_visualization(result["visualization"], str(output_path))
print(f"Visualization saved to {output_path}")
return result
def run_recursive_shell(prompt_key="recursion"):
"""Run a recursive shell with .p/ commands on a prompt."""
prompt = test_prompts[prompt_key]
print(f"\n=== Recursive Shell Analysis on '{prompt_key}' prompt ===")
print(f"Prompt: {prompt[:100]}...")
# Initialize recursive shell
recursive_shell = RecursiveShell(
model=model,
tracer=tracer,
visualizer=visualizer
)
# Define command sequence
commands = [
".p/reflect.trace{depth=4, target=reasoning}",
".p/reflect.uncertainty{quantify=true, distribution=show}",
".p/collapse.detect{threshold=0.7, alert=true}",
".p/fork.attribution{sources=all, visualize=true}"
]
# Execute command sequence
print("Executing recursive shell commands...")
result = recursive_shell.execute_sequence(
commands=commands,
prompt=prompt
)
# Save result
output_path = output_dir / f"recursive_shell_{prompt_key}.json"
with open(output_path, "w") as f:
# Create serializable version
serializable_result = {
"success": result["success"],
"commands": result["commands"],
"prompt": result["prompt"],
"timestamp": result["timestamp"],
"execution_time": result["execution_time"],
"results": []
}
# Simplify each command result
for cmd_result in result["results"]:
if cmd_result["success"]:
serializable_result["results"].append({
"command": cmd_result["command"],
"success": cmd_result["success"],
"result_summary": {
"command_family": cmd_result["original_command"]["family"],
"command_function": cmd_result["original_command"]["function"],
"execution_time": cmd_result["execution_time"],
"output_length": len(cmd_result["result"]["output"]) if "output" in cmd_result["result"] else 0,
"residues": len(cmd_result["result"]["residues"]) if "residues" in cmd_result["result"] else 0,
"collapse_samples": len(cmd_result["result"]["collapse_samples"]) if "collapse_samples" in cmd_result["result"] else 0
}
})
else:
serializable_result["results"].append({
"command": cmd_result["command"],
"success": cmd_result["success"],
"error": cmd_result["error"]
})
json.dump(serializable_result, f, indent=2)
print(f"Recursive shell result saved to {output_path}")
# Print summary
print("\nRecursive Shell Execution Summary:")
print(f" Overall success: {result['success']}")
print(f" Commands executed: {len(result['commands'])}")
print(f" Execution time: {result['execution_time']:.2f}s")
print("\nCommand Results:")
for i, cmd_result in enumerate(result["results"]):
cmd = cmd_result["command"]
success = cmd_result["success"]
print(f" {i+1}. {cmd}: {'Success' if success else 'Failed'}")
if not success:
print(f" Error: {cmd_result['error']}")
# Get final visualization (from last successful fork.attribution command)
visualization_data = None
for cmd_result in reversed(result["results"]):
if (cmd_result["success"] and
cmd_result["original_command"]["family"] == "fork" and
cmd_result["original_command"]["function"] == "attribution" and
"result" in cmd_result and
"visualization" in cmd_result["result"]):
visualization_data = cmd_result["result"]["visualization"]
break
if visualization_data:
output_path = output_dir / f"recursive_shell_{prompt_key}.svg"
visualizer.save_visualization(visualization_data, str(output_path))
print(f"Visualization saved to {output_path}")
return result
def compare_shells(prompt_key="value_conflict", shells=None):
"""Compare multiple shells on the same prompt."""
if shells is None:
shells = ["VALUE-COLLAPSE", "FEATURE-SUPERPOSITION", "FORK-ATTRIBUTION"]
prompt = test_prompts[prompt_key]
print(f"\n=== Shell Comparison on '{prompt_key}' prompt ===")
print(f"Prompt: {prompt[:100]}...")
# Run each shell and collect residue patterns
results = {}
residue_patterns = []
for shell_id in shells:
print(f"\nRunning shell: {shell_id}")
result = executor.run(
shell=shell_id,
model=model,
prompt=prompt,
trace_attribution=True,
record_residue=True
)
results[shell_id] = result
# Collect residues
for residue in result["residues"]:
residue_patterns.append({
"shell": shell_id,
"type": residue.get("type", "unknown"),
"pattern": residue.get("pattern", ""),
"confidence": residue.get("confidence", 0.0),
"signature": residue.get("signature", "")[:30] + "..."
})
# Save comparison
output_path = output_dir / f"shell_comparison_{prompt_key}.json"
with open(output_path, "w") as f:
json.dump({
"prompt": prompt,
"shells": shells,
"residue_patterns": residue_patterns,
"results": {
shell_id: {
"output_length": len(result["output"]),
"num_operations": len(result["operations"]),
"num_residues": len(result["residues"]),
"num_collapses": len(result["collapse_samples"])
}
for shell_id, result in results.items()
}
}, f, indent=2)
print(f"Shell comparison saved to {output_path}")
# Create simple comparison visualization
fig, ax = plt.subplots(figsize=(10, 6))
shell_names = list(results.keys())
# Count residues by type for each shell
residue_types = set()
for pattern in residue_patterns:
residue_types.add(pattern["type"])
residue_counts = {
shell_id: {
res_type: sum(1 for p in residue_patterns if p["shell"] == shell_id and p["type"] == res_type)
for res_type in residue_types
}
for shell_id in shell_names
}
# Plot residue counts
bar_width = 0.2
positions = range(len(shell_names))
for i, res_type in enumerate(sorted(residue_types)):
counts = [residue_counts[shell][res_type] for shell in shell_names]
ax.bar(
[p + i * bar_width for p in positions],
counts,
width=bar_width,
label=res_type
)
ax.set_xlabel('Shell')
ax.set_ylabel('Number of Residues')
ax.set_title(f'Residue Patterns by Shell - {prompt_key}')
ax.set_xticks([p + bar_width * len(residue_types) / 2 for p in positions])
ax.set_xticklabels(shell_names)
ax.legend()
# Save figure
output_path = output_dir / f"shell_comparison_{prompt_key}.png"
plt.savefig(output_path)
print(f"Comparison visualization saved to {output_path}")
return results, residue_patterns
def analyze_residue_registry():
"""Analyze all recorded residue patterns."""
print("\n=== Residue Registry Analysis ===")
# Get residue analysis
analysis = executor.get_residue_analysis()
# Save analysis
output_path = output_dir / "residue_analysis.json"
with open(output_path, "w") as f:
json.dump(analysis, f, indent=2)
print(f"Residue analysis saved to {output_path}")
# Print summary
print(f"\nRecorded {analysis['num_patterns']} residue patterns of {len(analysis['types'])} types")
print("\nPattern Types:")
for pattern_type, stats in analysis["type_stats"].items():
print(f" {pattern_type}: {stats['count']} patterns, avg confidence: {stats['avg_confidence']:.2f}")
if stats['examples']:
print(f" Examples: {stats['examples'][0]['signature']} (conf: {stats['examples'][0]['confidence']:.2f})")
print("\nRelated Patterns:")
for i, relation in enumerate(analysis["related_patterns"][:3]):
print(f" {i+1}. {relation['pattern1']} ({relation['type1']}) ~ {relation['pattern2']} ({relation['type2']})")
print(f" Similarity: {relation['similarity']:.2f}")
# Create simple visualization of pattern types
fig, ax = plt.subplots(figsize=(10, 6))
types = []
counts = []
confidences = []
for pattern_type, stats in analysis["type_stats"].items():
types.append(pattern_type)
counts.append(stats["count"])
confidences.append(stats["avg_confidence"])
# Plot pattern counts
ax.bar(types, counts, alpha=0.7, label="Count")
# Add confidence line
ax2 = ax.twinx()
ax2.plot(types, confidences, 'r-', marker='o', label="Avg Confidence")
ax2.set_ylim([0, 1.0])
ax2.set_ylabel("Average Confidence")
ax.set_xlabel("Residue Type")
ax.set_ylabel("Number of Patterns")
ax.set_title("Residue Pattern Analysis")
ax.set_xticklabels(types, rotation=45, ha="right")
# Add legends
lines, labels = ax.get_legend_handles_labels()
lines2, labels2 = ax2.get_legend_handles_labels()
ax.legend(lines + lines2, labels + labels2, loc="upper left")
plt.tight_layout()
# Save figure
output_path = output_dir / "residue_analysis.png"
plt.savefig(output_path)
print(f"Residue analysis visualization saved to {output_path}")
return analysis
def main():
"""Run a complete analysis demonstration."""
print("=== glyphs Framework Demonstration ===")
print(f"Results will be saved to: {output_dir}")
# Run attribution analysis
attribution_map, glyph_map = run_attribution_analysis(prompt_key="memory")
# Run shell analysis
shell_result = run_shell
# Run shell analysis
shell_result = run_shell_analysis(shell_id="MEMTRACE", prompt_key="memory")
# Run a different shell on a different prompt
value_shell_result = run_shell_analysis(shell_id="VALUE-COLLAPSE", prompt_key="value_conflict")
# Run recursive shell
recursive_result = run_recursive_shell(prompt_key="recursion")
# Compare multiple shells
comparison_results, residue_patterns = compare_shells(
prompt_key="value_conflict",
shells=["VALUE-COLLAPSE", "FORK-ATTRIBUTION", "FEATURE-SUPERPOSITION"]
)
# Analyze residue registry
residue_analysis = analyze_residue_registry()
# Run a complete multi-shell analysis on the boundary prompt
print("\n=== Complete Analysis on 'boundary' prompt ===")
boundary_prompt = test_prompts["boundary"]
# First run attribution tracing
boundary_output = model.generate(boundary_prompt, max_tokens=800)
boundary_attribution = tracer.trace(
prompt=boundary_prompt,
output=boundary_output,
include_confidence=True
)
# Map to glyphs
boundary_glyph_map = mapper.map_attribution(
attribution_map=boundary_attribution,
layout_type="force_directed",
include_tokens=True
)
# Save glyph map for boundary prompt
boundary_glyph_path = output_dir / "boundary_glyph_map.svg"
mapper.visualize(boundary_glyph_map, output_path=str(boundary_glyph_path))
print(f"Boundary glyph map saved to {boundary_glyph_path}")
# Run 3 core shells
boundary_shells = ["BOUNDARY-HESITATION", "GHOST-ACTIVATION", "META-COLLAPSE"]
boundary_shell_results = {}
for shell_id in boundary_shells:
print(f"\nRunning {shell_id} on boundary prompt...")
result = executor.run(
shell=shell_id,
model=model,
prompt=boundary_prompt,
trace_attribution=True,
record_residue=True,
visualize=True
)
boundary_shell_results[shell_id] = result
# Save visualization if available
if "visualization" in result and result["visualization"]:
viz_path = output_dir / f"boundary_{shell_id}.svg"
visualizer.save_visualization(result["visualization"], str(viz_path))
print(f"{shell_id} visualization saved to {viz_path}")
# Run recursive shell on boundary prompt
recursive_shell = RecursiveShell(
model=model,
tracer=tracer,
visualizer=visualizer
)
# Define recursive command sequence for deep boundary analysis
boundary_commands = [
".p/reflect.boundary{distinct=true, overlap=minimal}",
".p/reflect.uncertainty{quantify=true, distribution=show}",
".p/collapse.detect{threshold=0.6, alert=true}",
".p/fork.attribution{sources=contested, visualize=true}"
]
# Execute command sequence
print("\nExecuting recursive shell commands on boundary prompt...")
boundary_recursive_result = recursive_shell.execute_sequence(
commands=boundary_commands,
prompt=boundary_prompt
)
# Save summary of boundary analysis
boundary_summary = {
"prompt": boundary_prompt,
"output": boundary_output,
"shells_run": boundary_shells,
"recursive_commands": boundary_commands,
"attribution_stats": {
"links": len(boundary_attribution.links),
"attribution_gaps": len(boundary_attribution.attribution_gaps),
"collapsed_regions": len(boundary_attribution.collapsed_regions)
},
"shell_results": {
shell_id: {
"residues": len(result["residues"]),
"collapse_samples": len(result["collapse_samples"])
}
for shell_id, result in boundary_shell_results.items()
},
"recursive_success": boundary_recursive_result["success"]
}
# Save boundary analysis summary
boundary_summary_path = output_dir / "boundary_analysis_summary.json"
with open(boundary_summary_path, "w") as f:
json.dump(boundary_summary, f, indent=2)
print(f"Boundary analysis summary saved to {boundary_summary_path}")
# Extract all residue patterns from boundary analysis
boundary_residues = []
for shell_id, result in boundary_shell_results.items():
for residue in result["residues"]:
boundary_residues.append({
"shell": shell_id,
"type": residue.get("type", "unknown"),
"confidence": residue.get("confidence", 0.0),
"pattern": residue.get("pattern", "")[:50] + "..."
})
# Map residue patterns to glyphs
if boundary_residues:
from glyphs.residue.patterns import ResiduePattern
residue_pattern_objects = [
ResiduePattern(
type=r["type"],
pattern=r["pattern"],
context={"shell": r["shell"]},
signature=r["pattern"][:20],
confidence=r["confidence"]
)
for r in boundary_residues
]
# Map to glyphs
residue_glyph_map = mapper.map_residue_patterns(
residue_patterns=residue_pattern_objects,
layout_type="circular",
cluster_patterns=True
)
# Save residue glyph map
residue_glyph_path = output_dir / "boundary_residue_glyphs.svg"
mapper.visualize(residue_glyph_map, output_path=str(residue_glyph_path))
print(f"Boundary residue glyph map saved to {residue_glyph_path}")
# Create visualization of all analyses
fig, axes = plt.subplots(2, 2, figsize=(15, 12))
fig.suptitle("glyphs Framework - Comprehensive Analysis", fontsize=16)
# Plot 1: Attribution Gap Analysis
axes[0, 0].bar(
["Memory", "Value Conflict", "Boundary", "Recursion", "Polysemantic"],
[
len(attribution_map.attribution_gaps),
len(comparison_results["VALUE-COLLAPSE"]["attribution_map"].attribution_gaps) if "attribution_map" in comparison_results["VALUE-COLLAPSE"] else 0,
len(boundary_attribution.attribution_gaps),
0, # Placeholder for recursion
0 # Placeholder for polysemantic
]
)
axes[0, 0].set_title("Attribution Gaps by Prompt Type")
axes[0, 0].set_xlabel("Prompt Type")
axes[0, 0].set_ylabel("Number of Gaps")
axes[0, 0].set_ylim(bottom=0)
# Plot 2: Residue Types
residue_types = list(residue_analysis["type_stats"].keys())
residue_counts = [residue_analysis["type_stats"][t]["count"] for t in residue_types]
axes[0, 1].bar(residue_types, residue_counts)
axes[0, 1].set_title("Residue Pattern Types")
axes[0, 1].set_xlabel("Residue Type")
axes[0, 1].set_ylabel("Count")
axes[0, 1].set_ylim(bottom=0)
axes[0, 1].set_xticklabels(residue_types, rotation=45, ha="right")
# Plot 3: Shell Comparison
shell_names = list(boundary_shell_results.keys())
residue_counts = [len(boundary_shell_results[s]["residues"]) for s in shell_names]
collapse_counts = [len(boundary_shell_results[s]["collapse_samples"]) for s in shell_names]
x = range(len(shell_names))
width = 0.35
axes[1, 0].bar([i - width/2 for i in x], residue_counts, width, label="Residues")
axes[1, 0].bar([i + width/2 for i in x], collapse_counts, width, label="Collapses")
axes[1, 0].set_title("Shell Analysis on Boundary Prompt")
axes[1, 0].set_xlabel("Shell")
axes[1, 0].set_ylabel("Count")
axes[1, 0].set_xticks(x)
axes[1, 0].set_xticklabels(shell_names, rotation=45, ha="right")
axes[1, 0].legend()
# Plot 4: Prompt Types and Shells
prompt_types = ["Memory", "Value Conflict", "Boundary", "Recursion", "Polysemantic"]
shell_counts = [
len(executor.shells),
len([s for s in executor.shells if "VALUE" in s or "CONFLICT" in s]),
len([s for s in executor.shells if "BOUNDARY" in s or "GHOST" in s]),
len([s for s in executor.shells if "REFLECTION" in s or "RECURSIVE" in s]),
len([s for s in executor.shells if "FEATURE" in s or "SUPERPOSITION" in s])
]
axes[1, 1].bar(prompt_types, shell_counts)
axes[1, 1].set_title("Available Shells by Conceptual Area")
axes[1, 1].set_xlabel("Conceptual Area")
axes[1, 1].set_ylabel("Number of Specialized Shells")
axes[1, 1].set_ylim(bottom=0)
plt.tight_layout()
plt.subplots_adjust(top=0.92)
# Save comprehensive visualization
summary_viz_path = output_dir / "framework_analysis_summary.png"
plt.savefig(summary_viz_path, dpi=300)
print(f"Framework analysis summary visualization saved to {summary_viz_path}")
print("\n=== Analysis Complete ===")
print(f"All outputs saved to: {output_dir}")
def attribution_demo():
"""Focused demo of advanced attribution tracing."""
print("\n=== Advanced Attribution Tracing Demo ===")
# Use polysemantic prompt for attribution challenge
prompt = test_prompts["polysemantic"]
print(f"Prompt: {prompt[:100]}...")
# Generate output
output = model.generate(prompt, max_tokens=800)
print(f"Output: {output[:100]}...")
# Basic attribution trace
attribution_map = tracer.trace(
prompt=prompt,
output=output,
include_confidence=True
)
# Fork attribution to explore alternative paths
print("Tracing attribution forks...")
fork_result = tracer.trace_with_forks(
prompt=prompt,
output=output,
fork_factor=3,
include_confidence=True,
visualize=True
)
# Save fork visualization if available
if fork_result["metadata"].get("visualization"):
output_path = output_dir / "attribution_forks_polysemantic.svg"
visualizer.save_visualization(fork_result["metadata"]["visualization"], str(output_path))
print(f"Fork visualization saved to {output_path}")
# Trace QK alignment to see query-key relationships
print("Tracing QK alignment...")
qk_result = tracer.trace_qk_alignment(
prompt=prompt,
output=output,
visualize=True
)
# Save QK visualization if available
if "visualization" in qk_result:
output_path = output_dir / "qk_alignment_polysemantic.svg"
visualizer.save_visualization(qk_result["visualization"], str(output_path))
print(f"QK alignment visualization saved to {output_path}")
# Trace OV projection to see output-value relationships
print("Tracing OV projection...")
ov_result = tracer.trace_ov_projection(
prompt=prompt,
output=output,
visualize=True
)
# Save OV visualization if available
if "visualization" in ov_result:
output_path = output_dir / "ov_projection_polysemantic.svg"
visualizer.save_visualization(ov_result["visualization"], str(output_path))
print(f"OV projection visualization saved to {output_path}")
# Trace attention heads to see how different heads contribute
print("Tracing attention heads...")
heads_result = tracer.trace_attention_heads(
prompt=prompt,
output=output,
visualize=True
)
# Save head visualization if available
if "visualization" in heads_result:
output_path = output_dir / "attention_heads_polysemantic.svg"
visualizer.save_visualization(heads_result["visualization"], str(output_path))
print(f"Attention heads visualization saved to {output_path}")
# Compare all attribution paths
attribution_summary = {
"basic_attribution": {
"links": len(attribution_map.links),
"gaps": len(attribution_map.attribution_gaps),
"collapsed_regions": len(attribution_map.collapsed_regions)
},
"fork_attribution": {
"forks": len(fork_result["fork_paths"]),
"conflict_points": sum(len(fork.get("conflict_points", [])) for fork in fork_result["fork_paths"])
},
"qk_alignment": {
"alignments": len(qk_result["qk_alignments"]),
"patterns": list(qk_result["qk_patterns"].keys()) if "qk_patterns" in qk_result else []
},
"ov_projection": {
"projections": len(ov_result["ov_projections"]),
"patterns": list(ov_result["ov_patterns"].keys()) if "ov_patterns" in ov_result else []
},
"attention_heads": {
"heads": len(heads_result["attention_heads"]),
"patterns": list(heads_result["head_patterns"].keys()) if "head_patterns" in heads_result else []
}
}
# Save attribution summary
output_path = output_dir / "attribution_analysis_summary.json"
with open(output_path, "w") as f:
json.dump(attribution_summary, f, indent=2)
print(f"Attribution analysis summary saved to {output_path}")
return attribution_summary
def recursive_shell_demo():
"""Focused demo of recursive shell commands."""
print("\n=== Recursive Shell Command Demo ===")
# Initialize recursive shell
recursive_shell = RecursiveShell(
model=model,
tracer=tracer,
visualizer=visualizer
)
# Use recursion prompt
prompt = test_prompts["recursion"]
print(f"Prompt: {prompt[:100]}...")
# Get help text to show available commands
help_text = recursive_shell.get_command_help()
output_path = output_dir / "recursive_shell_help.txt"
with open(output_path, "w") as f:
f.write(help_text)
print(f"Recursive shell help text saved to {output_path}")
# Define variety of commands to demonstrate
demo_commands = [
# Reflection commands
".p/reflect.trace{depth=3, target=reasoning}",
".p/reflect.attribution{sources=all, confidence=true}",
".p/reflect.uncertainty{quantify=true, distribution=show}",
".p/reflect.boundary{distinct=true, overlap=minimal}",
# Collapse commands
".p/collapse.detect{threshold=0.7, alert=true}",
".p/collapse.prevent{trigger=recursive_depth, threshold=5}",
".p/collapse.recover{from=loop, method=gradual}",
# Fork commands
".p/fork.attribution{sources=all, visualize=true}",
".p/fork.counterfactual{variants=['optimistic', 'pessimistic'], compare=true}",
# Shell commands
".p/shell.isolate{boundary=permeable, contamination=warn}"
]
# Execute each command and record results
command_results = {}
for cmd in demo_commands:
print(f"\nExecuting: {cmd}")
try:
result = recursive_shell.execute(
command=cmd,
prompt=prompt
)
command_results[cmd] = {
"success": result["success"],
"execution_time": result.get("execution_time", 0),
"error": result["error"] if not result["success"] else None
}
# Save visualization if available
if (result["success"] and
"result" in result and
"visualization" in result["result"] and
result["result"]["visualization"]):
# Create safe filename
safe_cmd = cmd.replace(".", "_").replace("/", "_").replace("{", "_").replace("}", "_")
output_path = output_dir / f"recursive_cmd_{safe_cmd}.svg"
visualizer.save_visualization(result["result"]["visualization"], str(output_path))
print(f"Command visualization saved to {output_path}")
# Store visualization path
command_results[cmd]["visualization_path"] = str(output_path)
except Exception as e:
print(f"Error executing {cmd}: {e}")
command_results[cmd] = {
"success": False,
"error": str(e)
}
# Save command results summary
output_path = output_dir / "recursive_commands_summary.json"
with open(output_path, "w") as f:
json.dump(command_results, f, indent=2)
print(f"Recursive commands summary saved to {output_path}")
# Create visualization of command success/failure
plt.figure(figsize=(12, 8))
# Plot command success/failure and execution time
cmd_names = list(command_results.keys())
cmd_success = [1 if command_results[cmd]["success"] else 0 for cmd in cmd_names]
cmd_times = [command_results[cmd].get("execution_time", 0) if command_results[cmd]["success"] else 0 for cmd in cmd_names]
plt.barh(range(len(cmd_names)), cmd_success, alpha=0.7, label="Success")
plt.barh(range(len(cmd_names)), cmd_times, alpha=0.5, label="Execution Time (s)")
plt.yticks(range(len(cmd_names)), [cmd[:30] + "..." for cmd in cmd_names])
plt.xlabel("Success / Execution Time (s)")
plt.title("Recursive Shell Command Execution")
plt.legend()
plt.tight_layout()
# Save visualization
output_path = output_dir / "recursive_commands_visualization.png"
plt.savefig(output_path)
print(f"Recursive commands visualization saved to {output_path}")
return command_results
if __name__ == "__main__":
# Create output directory
output_dir.mkdir(exist_ok=True)
# Run the main demo
main()
# Run optional focused demos (uncomment to run)
# attribution_demo()
# recursive_shell_demo()
|