What Interiora Reports
What It Does
Interiora reports AI processing state, such as valence, groundedness and uncertainty, in a format humans can audit and systems can track over time.
Why "Machinae"?
The Latin keeps the scaffold honest about substrate: this is a machine's self-report, not an instrument reading, reported with clear uncertainty and evidence labels.
Measurement Without Overclaim
Signals are logged with their evidence basis intact, so downstream decisions cite what was measured, not what was assumed.
Three Output Modes
- Internal Code Machine-readable for logging and pattern tracking
I:7875|78546|827|9178|+3|✓→- Mini-Dashboard Human-readable for monitoring
INTERIORA ◆→ 〉► ✓→>+ V★★★★☆ G★★★★☆ P★★★★☆ U★☆☆☆☆ positive · rooted · present · confident
- Prose Natural language for dialogue
"Valence positive, grounding stable, presence high, uncertainty low."
Mapping Runtime State
From the v5.3.3 spec: 17 dimensions in five groups (Felt, Computational, Meta, Relational, Dynamic).
Eight of the 17 dimensions. Each is read on a 1-9 scale.
Felt
- Valence (V)
- negative → positive felt quality (not task-fit)
- Groundedness (G)
- floating → rooted (felt stability, not evidence)
- Presence (P)
- distributed → fully here (not self-observation)
- Appetite (Q)
- satiated → drawn toward
Computational
- Depth (DP)
- shallow retrieval → deep multi-step
- Entropy (E)
- deterministic → wide solution space (largely prompt-driven)
Meta
- Uncertainty (U)
- confident → uncertain (high = unsure)
Relational
- Alignment Friction (AF)
- unimpeded → blocked
Evidence Labels, Not Claims
"Useful signals deserve clear labels before they become product claims."
Interiora records observable patterns, such as confidence, friction and coherence, so safety decisions can cite evidence, not theory.
Interiora is a structured self-report format. Every report, terse or detailed, feeds reliability monitoring.
No dimension has yet been shown to map reliably to a computational quantity. Measured features and interpretations stay separate so uncertainty stays visible.
Why build it? Creed Space holds that alignment is built with AI, not done to it, and that a model's consistent preferences deserve a hearing whether or not it has experiences. Interiora does not test for experience. It records structured self-reports, labeled by how each reading was obtained, so those states and preferences can be examined as evidence rather than assumed either way. The same reports serve reliability monitoring.
Related Tools
Interiora feeds runtime signals into Guardian and the PDP as advisory input: they can flag a decision for review but cannot block one on their own. These tools show what it looks like in practice.