MUX Experience Tests Documentation

Purpose

Verify that features can be described using experience language (“Piper noticed…”) rather than database language (“Query returned…”).

Experience tests are the canary in the coal mine - if we cannot describe a feature using the grammar, we’ve lost consciousness in the implementation.

The Object Model Grammar

“Entities experience Moments in Places.”

This sentence is the foundation. Every feature must be expressible using these three substrate protocols plus the supporting vocabulary:

• Entities: Actors with identity and agency
• Moments: Bounded significant occurrences with theatrical unities
• Places: Contexts with atmosphere where action happens

Morning Standup (Reference Implementation)

Grammar Expression

Experience Language (PASS)

These expressions demonstrate consciousness-preserving language:

• “Piper noticed that you have 3 meetings today”
• “Piper remembers that you completed 2 PRs yesterday”
• “Piper anticipates a busy afternoon based on your calendar”
• “I sense a pattern in your Monday mornings”
• “This has caught my attention - it seems significant”

Database Language (FAIL)

These expressions indicate flattening to data manipulation:

• “Query returned 3 calendar events”
• “Found 2 merged PRs in date range”
• “Calculated meeting density > threshold”
• “SELECT COUNT(*) FROM meetings WHERE date = today”
• “Row inserted into tasks table”

Anti-Flattening Test Categories

1. Entity Tests (4 tests)

Verify that entities preserve identity, not just IDs.

Pass criteria:

• Entity has type describing role (user, assistant, team)
• Entity has name/identity beyond primary key
• Entity can have agency (can_initiate, can_respond)

Fail indicators:

• Entity is just { id: string }
• No semantic information about WHO the entity is
• Entity cannot take action

2. Moment Tests (3 tests)

Verify that moments preserve significance, not just timestamps.

Pass criteria:

• Moment has description explaining what it IS
• Moment has significance explaining why it MATTERS
• Moment can capture outcomes and be remembered

Fail indicators:

• Moment is just { start_time, end_time }
• No semantic information about the experience
• Moments are queryable but not memorable

3. Place Tests (3 tests)

Verify that places preserve atmosphere, not just configuration.

Pass criteria:

• Place has modality describing HOW interaction happens
• Place has atmosphere describing what it FEELS like
• Place has affordances describing what CAN happen there

Fail indicators:

• Place is just a connection string or endpoint URL
• No information about character or experience
• Places are interchangeable containers

4. Lifecycle Tests (5 tests)

Verify that lifecycle includes composting, not just deletion.

Pass criteria:

• COMPOSTED state exists as terminal state
• Composting extracts lessons and wisdom
• Each state has experience phrase explaining consciousness
• Lifecycle tells a story (narrative, not audit log)

Fail indicators:

• Objects are simply deleted
• No learning extracted from lifecycle journey
• States are just status codes (1, 2, 3…)

5. Metadata Tests (5 tests)

Verify that metadata captures knowledge ABOUT knowledge.

Pass criteria:

• Provenance tracks WHERE with confidence
• Freshness decays (knowledge ages)
• Journal separates facts (session) from meaning (insight)
• Confidence knows WHY it’s confident

Fail indicators:

• Metadata is just attribute columns
• No distinction between facts and interpretation
• No temporal decay or aging

6. Ownership Tests (4 tests)

Verify that ownership describes relationships, not just foreign keys.

Pass criteria:

• Multiple ownership categories (NATIVE, FEDERATED, SYNTHETIC)
• Categories have consciousness metaphors (Mind, Senses, Understanding)
• Experience phrases use first person (“I know this because…”)

Fail indicators:

• Ownership is just owner_id field
• No semantic categories for relationship types
• No explanation of how Piper relates to the object

6.5. Ownership Metaphor Tests

Verify that ownership uses consciousness metaphors correctly.

Pass criteria:

• Native uses “Mind” metaphor in experience language
• Federated uses “Senses” metaphor in experience language
• Synthetic uses “Understanding” metaphor in experience language
• Classification follows decision tree from ownership-metaphors.md

Fail indicators:

• Generic ownership labels without metaphor connection
• Wrong category for information type
• Missing provenance for Synthetic
• Stale confidence for Federated

7. Design Principle Tests (4 tests)

Verify that CXO design principles are honored.

Pass criteria:

• “I sense…” not “status=1”
• “transformed” not “deleted”
• Learning language in insights
• Resolver provides reasoning, not just results

Fail indicators:

• Database language exposed in interfaces
• Status codes instead of experience phrases
• Black-box operations without explanation

8. Grammar Integration Tests (3 tests)

Verify the complete grammar works together.

Pass criteria:

• Morning standup expressible without inventing new concepts
• Protocol methods use experience language (experiences, contains, captures)
• Grammar concepts are distinct from database concepts

Fail indicators:

• Need to invent new concepts for basic features
• Methods named get/set/update/delete
• Grammar concepts are just table name aliases

9. Consciousness Vocabulary Tests (3 tests)

Verify consciousness vocabulary throughout implementation.

Pass criteria:

• Ownership metaphors reference Mind/Senses/Understanding
• Lifecycle phrases use “I” or “This” (first-person perspective)
• Journal separates facts from meaning

Fail indicators:

• Third-person impersonal language everywhere
• No distinction between what happened and what it meant
• Machine-like language (process, execute, terminate)

10. Transition Tests (3 tests)

Verify state transitions preserve meaning.

Pass criteria:

• Invalid transitions explain what’s wrong and valid options
• Transitions are forward-only (lifecycle is a journey)
• COMPOSTED is terminal by design (nothing beyond compost)

Fail indicators:

• Silent failures on invalid transitions
• Arbitrary state changes allowed
• Objects can be resurrected from COMPOSTED

11. Cathedral Test (3 tests)

The ultimate integration test - is this a cathedral or a shed?

Pass criteria:

• All core concepts have experience language
• The grammar is complete (3 protocols, 3 ownership, 8 lifecycle, 6 metadata)
• Nothing truly disappears - everything transforms

Fail indicators:

• Mechanical implementation without soul
• Incomplete grammar requiring workarounds
• Delete operations instead of composting

Verification Checklist

For each major feature, verify:

1. CAN describe using “Piper noticed/remembers/anticipates…”
2. CANNOT accurately describe using “Query/Database/Record…”
3. Grammar elements (Entity/Moment/Place) are identifiable
4. Lenses can be applied for different perspectives
5. Situation frame captures dramatic tension
6. Lifecycle states make sense for the feature
7. Ownership category is clear (who created/observed/derived this?)
8. Metadata dimensions are applicable
9. Composting extracts wisdom if applicable

Test Locations

Running Experience Tests

# Run anti-flattening tests specifically
pytest tests/unit/services/mux/test_anti_flattening.py -v

# Run all MUX tests
pytest tests/unit/services/mux/ -v

# Check test count
pytest tests/unit/services/mux/ --collect-only -q | tail -3

Part of MUX-399-PZ: Verification & Anti-Flattening Tests Created: 2026-01-19