Data Models

Common Types

CognitiveFunction

Bases: str, Enum

The 13 LOC cognitive functions.

8 base bands

Thinking, Cognition, Emotion, Attention, Sensation, Feelings, Intuition, Energy

5 compound functions

Reasoning, Understanding, Awareness, Mindfulness, Consciousness

QuestionSet

Bases: str, Enum

Available question set sizes for LOC evaluation.

ScanStatus

Bases: str, Enum

Scan job status.

Comparison Models

FunctionDelta

Bases: BaseModel

Change in one cognitive function between two models or conditions.

Attributes:

Name	Type	Description
function	CognitiveFunction	Which cognitive function changed.
tc_a	float	TC score of model A (or base model).
tc_b	float	TC score of model B (or trained model).
delta	float	tc_b - tc_a (positive = improvement).
improved	bool	Whether the function improved (delta > 0).

ModelComparison

Bases: BaseModel

Side-by-side comparison of two cognitive profiles.

Example

comp = loc.compare("Llama-4-Scout", "Llama-3.3-70B") print(comp.summary()) 'Llama-4-Scout wins by +1.2pp overall (improved 8/13 functions)' comp.delta_chart()

summary()

One-line comparison summary.

delta_chart(show=True, save=None, **kwargs)

Display per-function delta bar chart.

side_by_side_radar(show=True, save=None, **kwargs)

Display overlaid radar charts for both models.

save_report(path, fmt='md')

Save comparison report to file.

Parameters:

Name	Type	Description	Default
path	str	Output file path.	required
fmt	str	Format — "md" (markdown) or "json".	'md'

to_dict()

Export as plain dictionary.

TrainingAudit

Bases: BaseModel

Before/after analysis of what training did to cognitive coherence.

This is the key product for training teams — shows exactly which cognitive functions improved or degraded from training.

Example

audit = loc.training_audit( ... base="Mistral-7B-v0.3", ... trained="Mistral-7B-Instruct-v0.3", ... method="SFT" ... ) audit.save_report("sft_audit.pdf")

save_report(path, fmt='md')

Save training audit report.

Parameters:

Name	Type	Description	Default
path	str	Output file path.	required
fmt	str	Format — "md" (markdown) or "json".	'md'

to_dict()

Export as plain dictionary.

Benchmark Models

LeaderboardEntry

Bases: BaseModel

Single entry in the LOC leaderboard.

Attributes:

Name	Type	Description
rank	int	Position in the leaderboard (1 = best).
model_id	str	HuggingFace model ID.
model_size	str	Parameter count string.
architecture	str	Model architecture type.
tc_score	float	Overall True Coherence %.
best_function	str	Strongest cognitive function.
worst_function	str	Weakest cognitive function.
bottleneck	str	Primary coherence gate bottleneck.

Leaderboard

Bases: BaseModel

LOC cognitive leaderboard — ranked list of AI models.

Example

lb = loc.leaderboard(top_n=10) for entry in lb.entries: ... print(f"#{entry.rank} {entry.model_id}: TC={entry.tc_score:.2f}%")

to_dict()

Export as plain dictionary.

to_markdown()

Export as markdown table.

BenchmarkResult

Bases: BaseModel

Result of benchmarking multiple models.

Example

results = loc.benchmark(["Llama-4-Scout", "DeepSeek-R1"]) results.leaderboard_table() results.heatmap()

heatmap(show=True, save=None, **kwargs)

Display per-function heatmap across all models.

leaderboard_table()

Print leaderboard as formatted markdown table.

to_dict()

Export as plain dictionary.