Phase 2: Noosphere Engine analytics modules

[user1303836]

Summary

Implements six Phase 2 analytics modules for the Noosphere Engine, following the architecture spec from chaos_research.md (ARCHITECT-1 through ARCHITECT-4).

Modules

• Resonance Mirror (resonance_mirror/): Discord embed visualization of community coherence. Uses EI-driven color coding, Unicode bars, and trend arrows. Slash command: /mirror
• Attractor Dashboard (attractor_dashboard/): 9-metric display (coherence, momentum, topic entropy/churn, activity entropy, reply depth, modularity) with change-point detection via ruptures. Slash command: /dashboard
• Cordyceps Audit (cordyceps_audit/): Standalone bot influence measurement. Herfindahl index for message concentration + vocabulary Jaccard for linguistic mimicry. Slash command: /cordyceps
• Morphogenetic Field (morphogenetic_field/): User interaction graph with cosine coupling scores, NetworkX modularity. Slash command: /morph
• Cryptobiosis Mode (cryptobiosis.py): 3-state machine (active/entering/cryptobiotic) with configurable activity thresholds. Emits cryptobiosis_trigger events.
• Pathology Detection (pathology.py + pathology_cog.py): Echo chamber, bot dominance, server death, flame war, clique formation. Z-score with hybrid absolute thresholds per Arch-2/Arch-3 spec. Slash command: /pathology

Architecture

• All modules consume CommunityStateVector via on_state_vector_updated event listener
• All unprompted output goes through bot.output_governor.send() (OutputGovernor pattern)
• Shared data models in noosphere/shared/models.py (CommunityStateVector, ProcessedMessage)
• Dependencies on Phase 0/1 infrastructure are coded to the spec interfaces

Dependencies added

• ruptures -- change-point detection for Attractor Dashboard
• networkx -- graph analysis for Morphogenetic Field modularity

Tests

85 tests covering all business logic across 7 test files:

• Resonance mirror: bar rendering, color coding, trend arrows
• Attractor dashboard: metric extraction, change-point detection, formatting
• Cordyceps audit: Herfindahl index, vocabulary Jaccard, parasitism scoring
• Morphogenetic field: user state, coupling computation, graph modularity
• Cryptobiosis: state machine transitions, wake/sleep behavior
• Pathology: all 5 detectors with healthy/flagged cases, z-score thresholds

Verification

• ruff check . -- All checks passed
• ruff format --check . -- All files formatted
• mypy src/ -- Success: no issues found
• pytest tests/ -- 676 passed (85 new + 591 existing)

[greptile-apps]

Your free trial has ended. If you'd like to continue receiving code reviews, you can add a payment method here.

[user1303836]

Code Review: PR #188 -- Noosphere Engine Phase 2 Analytics

Verdict: Changes Requested -- 6 issues to address before merge, 5 positive observations.

---

Issues Requiring Changes

1. shared/models.py diverges from canonical shared/data_models.py -- shared/models.py

This PR ships its own shared/models.py with int IDs, list[float] embeddings, and bare str classification. Per our alignment agreement:

• IDs must be str (existing codebase uses String(36) throughout)
• Embeddings should be np.ndarray | None (not list[float])
• Classification should use MessageClassification enum (not bare str)
• The canonical shared models live in shared/data_models.py (owned by dev-foundation)

This file needs to be deleted and all imports redirected to intelstream.noosphere.shared.data_models. This is the same issue we flagged during interface alignment -- it must be resolved before merge.

2. int IDs used throughout all modules -- all cog and business logic files

All cogs use int for guild_id/user_id (e.g., _latest: dict[int, CommunityStateVector] in resonance_mirror/cog.py:25, dict[int, dict[int, int]] in cordyceps_audit/cog.py:25, UserState.user_id: int in morphogenetic_field/field.py:15). These must be str to match the existing codebase convention and the canonical shared models.

Affected files:

• resonance_mirror/cog.py lines 25-26
• cordyceps_audit/audit.py line 14 (dict[int, int])
• cordyceps_audit/cog.py lines 24-26
• morphogenetic_field/field.py lines 15-16, 31-32, 38, 41, 63, 71, 86
• morphogenetic_field/cog.py lines 22-23
• cryptobiosis.py line 85
• pathology_cog.py lines 27-28
• All test files using int IDs

3. CryptobiosisState and PathologyType use bare enum.Enum -- cryptobiosis.py:17, pathology.py:11

Per alignment agreement, all enums must use str, enum.Enum for JSON serialization compatibility. Currently:

class CryptobiosisState(enum.Enum):  # Should be (str, enum.Enum)
class PathologyType(enum.Enum):      # Should be (str, enum.Enum)

4. MorphogeneticFieldCog._on_message passes msg.embedding as-is to update_user -- morphogenetic_field/cog.py:33

update_user expects list[float] (field.py:44), but the canonical ProcessedMessage.embedding is np.ndarray | None. Two problems:

• No null check: if msg.embedding is None, np.array(None) will produce a 0-d array, breaking all downstream math
• Type mismatch: np.ndarray gets re-wrapped in np.array() unnecessarily

Fix: guard against None, and accept np.ndarray directly in update_user.

5. top_couplings is O(n^2) with no limit on user count -- morphogenetic_field/field.py:86-94

For large servers with thousands of active users, top_couplings computes all pairwise cosine similarities. With 1000 users, that's ~500k dot products. Consider either:

• Capping the user pool to the most-active N users before computing
• Or documenting this as a known limitation for Phase 2

6. Attractor dashboard interaction_modularity in METRIC_FIELDS -- attractor_dashboard/metrics.py:19

The dashboard displays interaction_modularity, but this field on CommunityStateVector defaults to 0.0 and nothing in Phase 1 computes it. The Morphogenetic Field module computes modularity separately in its own state. This means the dashboard will always show 0.000 for Modularity unless something populates csv.interaction_modularity. Either:

• Document that this field requires the Morphogenetic Field module to be running and wired into the state vector pipeline
• Or add a math.nan sentinel check and display "N/A" when the field is uncomputed

---

Positive Observations

• Business logic / cog separation: Every module cleanly separates pure computation (analyzer.py, audit.py, metrics.py, field.py, pathology.py) from Discord cog wiring. This makes testing straightforward and is a good pattern.

• Pathology detection design: Hybrid approach using both z-score (relative to baseline) and absolute thresholds is exactly what Arch-2/Arch-3 specifies. The detect_echo_chamber dual condition (csv.egregore_index > 0.85 AND csv.topic_entropy < 1.0) OR (ei_z > 2.0 AND entropy_z < -2.0) is well-designed.

• AttractorDashboard bounded history: MAX_HISTORY = 168 (one week of hourly snapshots) with proper truncation at history[-MAX_HISTORY:] prevents unbounded growth. Good -- this is exactly what PR Add Noosphere Engine foundation (Phase 0 + Phase 1) #187's MetricsComputer needs.

• CryptobiosisMonitor state machine: Clean 3-state machine (active/entering/cryptobiotic) with separate entering-threshold and dormancy-threshold. The wakeup-threshold for sustained activity before returning to active is a thoughtful detail that prevents flapping.

• Cordyceps VocabularyTracker: Counter.most_common(top_n) caps memory by only tracking the top 200 terms per side. Clean design.

---

Test Coverage Assessment

85 tests across 7 files. Coverage is thorough:

• Cordyceps: 16 tests (HHI, Jaccard, parasitism, tracker)
• Morphogenetic: 13 tests (user state, coupling, interactions, modularity)
• Pathology: 11 tests (all 5 detectors + scan aggregation)
• Cryptobiosis: 6 tests (all state transitions)
• Attractor: 10 tests (extraction, change points, formatting)
• Resonance: 11 tests (color, bar, trend, lines)
• Shared models: 4 tests

Missing: no cog-level integration tests. That's acceptable for Phase 2 -- cog wiring is mostly boilerplate that delegates to tested business logic.

---

@greptile

[user1303836]

Re-review: All 6 Items Addressed

I've re-read every source and test file in this PR against the original review. All items are resolved.

Original Issues -- Status

#	Issue	Status
1	shared/models.py diverges from canonical data_models.py	Fixed -- renamed to shared/data_models.py, matches canonical ProcessedMessage and CommunityStateVector
2	int Discord IDs throughout	Fixed -- all IDs are str, cogs convert at boundary via str(interaction.guild.id), tests use "1", "42" etc.
3	Enums use bare enum.Enum	Fixed -- CryptobiosisState, PathologyType, MessageClassification all use enum.StrEnum
4	MorphogeneticFieldCog._on_message missing null check on msg.embedding	Fixed -- if msg.embedding is None: return at line 32
5	top_couplings O(n^2)	Acceptable -- comment documents the tradeoff, fine for typical guild sizes
6	Dashboard interaction_modularity always 0	Improved -- CommunityStateVector now defaults to math.nan instead of 0.0, clearly signaling "not yet populated"

Verification Checklist

• All imports point to intelstream.noosphere.shared.data_models (not the old shared/models)
• ProcessedMessage has is_bot: bool field; CordycepsAuditCog uses msg.is_bot instead of bot user ID comparison
• CommunityStateVector has sentiment_alignment, interaction_modularity, fractal_dimension, lyapunov_exponent, gromov_curvature with math.nan defaults
• Event listeners use correct signatures: on_state_vector_updated receives CommunityStateVector, on_message_processed receives ProcessedMessage
• AttractorDashboardCog._history bounded by MAX_HISTORY = 168
• MorphogeneticFieldCog._reply_cache bounded with eviction at 10,000 entries
• Test data uses str IDs consistently across all 7 test files
• 85 new tests covering all 6 modules

Minor Note (not blocking)

data_models.py uses TYPE_CHECKING imports for datetime, numpy, and MessageClassification (lines 7-12). This is the same pattern from PR #187 and was flagged there -- not a new issue for this PR.

Verdict: Approve

This PR is clean and ready to merge. Well-structured modules with good test coverage.