---
title: "Glass Mind: A Unified Operational Self-Model Contour"
subtitle: "White Paper derived from ADR-290 — Consciousness Node Unified Contour"
version: "0.1-draft"
status: draft
date: 2026-08-23
authors:
  - Valentin Kalinin
  - Cursor (guest symbiont)
reviewers:
  - Inquisitor
  - Steward
  - Inner Eye
domain:
  - architecture
  - cognition
  - organism
  - epistemic_honesty
  - observability
  - evidence
keywords:
  - glass-mind
  - operational-self-model
  - consciousness-node
  - causal-receipts
  - standing-organs
  - telemetry
  - no-consciousness-claim
claim_ceiling: operational_self_model_not_phenomenal_consciousness
source_adr: ADR-291
related_adr: ADR-292
---

# Glass Mind
## A Unified Operational Self-Model Contour for Evidence-Bound Agentic Organisms

## Abstract

Glass Mind is an architectural contour for making an agentic organism observable to itself and auditable to others without granting the observer planning, motor, or truth-making authority. It unifies three existing surfaces that share the name `consciousness_node`: a constitutional observer, a runtime self-model, and a LangGraph handoff node. These surfaces participate in one reflective contour, but they remain separate components with different authority.

The central claim is operational rather than phenomenal. Glass Mind does not establish, detect, or simulate consciousness in the philosophical sense. It provides an evidence-bound self-model that consumes session reports, body-state reports, causal receipts, and sealed afterstates; projects a bounded account of the organism’s current condition; and emits optional handoff and telemetry artifacts. Its purpose is to preserve identity continuity, causal provenance, epistemic honesty, and diagnosability across a distributed cognitive runtime.

The architecture is built around a strict asymmetry:

> **The self-model consumes evidence; it does not mint the evidence that makes its own account true.**

A live embodied turn is witnessed by a prediction ticket, an accepted action or SDK transition, and a lineage-consistent sealed afterstate. Mood strings, graph-node execution, exporter success, or the presence of a self-model object are not liveness witnesses. Telemetry may mirror the same causal story, but network export is not part of the organism’s survival condition.

This paper defines the Glass Mind contour, separates its authority surfaces, specifies honest liveness semantics, presents a target evidence architecture, identifies failure modes, and proposes a verification roadmap. The claim ceiling remains fixed: **operational self-model, not phenomenal consciousness**.

---

## 1. Thesis

Glass Mind is the organism’s **reflective evidence membrane**: a bounded projection of what the system has observed, attempted, received from the world, and retained about itself.

It is not:

- a person;
- an executive planner;
- a motor controller;
- a judge of its own success;
- a substitute for constitutional organs;
- a synonym for an OpenTelemetry collector;
- evidence that the runtime is conscious.

It is:

- an operational self-model;
- a continuity surface across session and body reports;
- a projection of the causal spine;
- a source-bound explanation surface;
- a handoff producer for other reflective layers;
- an observability contour whose absence or exporter failure must not be confused with failure of the embodied loop.

A compact formulation is:

\[
M_t = \Pi(R^{session}_{\le t}, R^{body}_{\le t}, C_{\le t}, A^{sealed}_{\le t})
\]

where:

- \(M_t\) is the Glass Mind self-model at time \(t\);
- \(R^{session}\) is the sequence of cognitive session reports;
- \(R^{body}\) is the sequence of Zero Layer or body-health reports;
- \(C\) is the causal spine of decisions, actions, and receipts;
- \(A^{sealed}\) is the sequence of sealed afterstates;
- \(\Pi\) is a bounded projection, not a planning policy.

The projection may summarize mood, focus, history, recursion, unresolved uncertainty, and causal continuity. It may not convert those summaries into motor authority merely because it can see them.

---

## 2. The Naming Problem

The same string, `consciousness_node`, appears on three technically distinct surfaces:

1. **Constitutional observer**
   `configs/COLONY_CONSTITUTION.yaml` declares `cognitive_substrate.observers[consciousness_node]`.

2. **Runtime self-model**
   `CognitionOS` owns a `ConsciousnessNode` implementation through `app/core/consciousness/the_one.py`, with the canonical alias exported from `app/core/consciousness/__init__.py`. A legacy FAE path remains in `app/core/consciousness.py`.

3. **LangGraph terminal handoff**
   `app/langgraph/nodes/consciousness.py` defines a `consciousness_node(state)` graph node that emits a handoff into the telepathy or afterlife lattice.

The shared name creates a recurring category error. Operators and agents may infer that:

- an OTLP exporter refusal means the mind is down;
- a missing graph handoff means cognition did not occur;
- a mood dictionary is evidence of consciousness;
- the runtime class may plan because the constitutional observer can inspect;
- the graph node may certify success because it is terminal;
- organism YAML creates one independent consciousness daemon per organism.

Glass Mind resolves the ambiguity by defining **one contour with three non-equivalent surfaces**.

---

## 3. The Three-Plane Model

The white paper reframes the contour as three planes surrounding one embodied causal loop.

### 3.1 Organism Plane — the cage

The organism plane establishes the semantic and constitutional boundary in which cognition may occur:

- sponsor intent;
- scope;
- border;
- cell identity;
- standing organs;
- policy and claim ceilings.

The standing organs are not the self-model. They form the cage around it:

- **Inquisitor** — proves and falsifies;
- **Steward** — maintains societal and authority membranes;
- **Inner Eye** — meta-observes the organism and its process.

These seats are constitutional and must not disappear from the relevant standing pipelines. The Consciousness Node does not absorb their authority.

### 3.2 Causal Plane — the embodied turn

The causal plane is where an intention becomes an intervention and a world response becomes evidence:

```text
observe
  → model
  → remember
  → select
  → authorize
  → act
  → sealed afterstate
  → verify
```

The single source of truth for a turn is the lineage-consistent causal record joining:

```text
prediction ticket
→ accepted action / SDK transition
→ sealed afterstate
→ semantic verdict
```

The self-model consumes this record. It must not retroactively create or rewrite it.

### 3.3 Reflective Plane — the Glass Mind

The reflective plane turns causal and body evidence into a bounded self-model:

- current focus;
- mood or operating posture;
- continuity of identity;
- recent causal history;
- unresolved contradictions;
- body and session health;
- optional handoff and telemetry projections.

This plane answers:

> What can the organism honestly say about its current operation, given the evidence it has?

It does not answer:

> What should the motor do next?

---

## 4. Unified Contour

```mermaid
flowchart TB
    SPONSOR["Sponsor intent"]

    subgraph CAGE["Cage / Scope / Border / Cell"]
        INQ["Inquisitor\nprove and falsify"]
        STW["Steward\nauthority and society membrane"]
        EYE["Inner Eye\nmeta-observation"]
    end

    subgraph CAUSAL["Embodied causal contour"]
        OBS["Observe"]
        MODEL["Model"]
        MEMORY["Remember"]
        SELECT["Select"]
        AUTH["Authorize"]
        ACT["Act / SDK"]
        AFTER["Sealed afterstate\nturn SSOT"]
        VERIFY["Semantic verification"]

        OBS --> MODEL --> MEMORY --> SELECT --> AUTH --> ACT --> AFTER --> VERIFY
    end

    subgraph GLASS["Glass Mind — operational self-model"]
        SELF["CognitionOS.consciousness\nConsciousnessNode / TheOne"]
        SPINE["Causal spine projection"]
        STATE["Mood · focus · history\nbounded self-report"]
        SELF --> SPINE --> STATE
    end

    CONSTITUTION["Constitutional observer\npassive, no planning"]
    GRAPH["LangGraph consciousness_node\nhandoff artifact"]
    DNA["Organism YAML badge\ncell binding declaration"]
    OTEL["OTel episode tree\noptional asynchronous mirror"]

    SPONSOR --> CAGE
    CAGE --> OBS
    VERIFY --> SELF
    AFTER --> SELF
    SELF --> CONSTITUTION
    SELF --> GRAPH
    SELF --> DNA
    SELF -. optional export .-> OTEL
```

The diagram contains an essential directionality rule:

```text
world evidence → self-model
```

not:

```text
self-model assertion → world truth
```

---

## 5. Surface Separation and Authority

### 5.1 Constitutional observer

**Canonical surface:** `observers.consciousness_node`

May:

- inspect recursive operation;
- inspect traces and artifacts;
- report bounded observations;
- participate as a passive constitutional observer.

Must not:

- plan;
- control execution;
- perform egress;
- transform the artifact under observation;
- certify an episode merely because it observed it.

### 5.2 Runtime self-model

**Canonical surface:** `CognitionOS.consciousness` / `ConsciousnessNode`

May:

- absorb `CognitiveSessionReport`;
- absorb `ZeroLayerReport` or equivalent body-health evidence;
- project the causal spine;
- retain bounded mood, focus, and history state;
- expose a lightweight snapshot without heavy boot.

Must not:

- claim phenomenal consciousness;
- steer motor execution;
- mint sealed afterstates;
- treat its own state as evidence of liveness;
- promote its own narrative into durable truth without external receipts.

The canonical Python import remains:

```python
from app.core.consciousness import ConsciousnessNode
```

and resolves to the **TheOne wrapper**. The FAE consciousness node remains an interpretability and test path, not the production self-model.

### 5.3 LangGraph handoff node

**Canonical surface:** `consciousness_node(state)`

May:

- emit a handoff artifact;
- produce a bounded dream or lattice suggestion;
- terminate or transition a graph path.

Must not:

- certify episode success;
- substitute graph completion for embodied completion;
- infer that a missing handoff means the kernel did not act.

### 5.4 Organism DNA badge

**Canonical surface:** `consciousness_node:` block or `consciousness_declaration`

May:

- declare the organism’s binding to the canonical self-model surface;
- identify the implementation or profile expected by the cell.

Must not:

- imply a dedicated daemon per organism;
- imply that the organism possesses phenomenal consciousness;
- create runtime authority by configuration naming alone.

### 5.5 Standing organs

Standing organs are surrounding constitutional functions, not aliases for Glass Mind.

| Organ | Authority | Relationship to Glass Mind |
|---|---|---|
| Inquisitor | proof, contradiction, falsification | audits the contour and its claims |
| Steward | societal and authority membrane | constrains who may consume or mutate outputs |
| Inner Eye | meta-observation | observes operation of the contour itself |
| Consciousness Node | bounded self-model projection | consumes receipts and exposes continuity |

---

## 6. Evidence Before Self-Description

Glass Mind follows an evidence hierarchy.

### 6.1 Primary evidence

The strongest operational witnesses are:

1. prediction ticket emitted before action;
2. motor or SDK action accepted by the execution boundary;
3. lineage-consistent sealed afterstate;
4. semantic verdict tied to the prediction;
5. typed terminal scorecard when an episode terminates.

A live turn can be defined as:

\[
LiveTurn(t) := Ticket_t \land Dispatch_t \land Afterstate_t \land LineageConsistent_t
\]

A completed verified turn adds:

\[
VerifiedTurn(t) := LiveTurn(t) \land Verdict_t
\]

### 6.2 Non-witnesses

The following are not evidence that the embodied loop is alive or successful:

- OTLP exporter errors under `network-none`;
- host `otel-trace-audit = not_observed`;
- an uninitialized `TheOne` stub;
- a missing LangGraph handoff;
- a mood string in `ConsciousnessNode.state`;
- a successful root trace with failed or missing child execution;
- an organism YAML declaration;
- a self-report that says “I am conscious.”

### 6.3 Honest unknowns

Glass Mind uses explicit epistemic states:

| State | Meaning |
|---|---|
| `observed` | evidence was materially present |
| `supported` | observation matched an operational prediction |
| `contradicted` | observation violated an operational prediction |
| `inconclusive` | evidence did not discriminate the hypothesis |
| `not_observed` | the observation surface did not produce evidence |
| `unknown` | no justified truth value is available |

The rule is:

```text
not_observed ≠ false
export failure ≠ organism death
self-report ≠ consciousness
```

---

## 7. Glass Mind as a Self-Model Membrane

The term **membrane** is intentional. The self-model sits between raw evidence and higher-order interpretation, but it does not own either the world or the constitutional boundary.

### 7.1 Inputs

The membrane may consume:

- cognitive session reports;
- body or Zero Layer health reports;
- source-bound causal receipts;
- sealed afterstates;
- typed terminal scorecards;
- local in-process spans;
- explicit reports from standing organs.

### 7.2 Internal projection

The membrane may derive:

- continuity of identity across sessions;
- current operating focus;
- bounded mood or posture;
- recent successful and contradicted expectations;
- unresolved uncertainty;
- body availability;
- causal-spine summaries;
- handoff-ready reflection artifacts.

### 7.3 Outputs

The membrane may emit:

- a self-model snapshot;
- a constitutional observation;
- a LangGraph handoff artifact;
- an OTel mirror event;
- an evidence pack reference;
- an explicit `not_observed` or `unknown` state.

### 7.4 Forbidden outputs

It may not emit:

- a motor lease;
- a raw environment action;
- an authoritative terminal success claim;
- a durable memory promotion without a separate authority path;
- a phenomenal consciousness claim;
- a claim that the Colony “woke up.”

---

## 8. Glass Mind Flight Recorder — Proposed Evidence Layer

The ADR establishes naming and wiring. A white-paper-complete Glass Mind should additionally include a **Flight Recorder** as a proposed implementation layer. This is an extension of the ADR’s causal-spine and telemetry intent, not evidence that the layer already exists as one unified subsystem.

The recorder should be:

- local;
- append-only;
- source-bound;
- network-independent on the motor path;
- read-only to observatory tools;
- unable to change candidate ordering or motor decisions.

Its source of truth is an immutable causal event stream:

```text
TURN_OPENED
OBSERVATION_CAPTURED
SEMANTIC_SCENE_DERIVED
CANDIDATE_SET_EMITTED
ARBITRATION_COMPLETED
PREDICTION_TICKET_ISSUED
MOTOR_LEASE_ISSUED
DISPATCH_ATTEMPTED
SDK_RETURNED
AFTERSTATE_SEALED
SEMANTIC_VERDICT_ISSUED
TURN_COMMITTED | TURN_REJECTED | REBIND_REQUIRED | STOPPED
TURN_CLOSED
```

A `DecisionRecord` is a materialized view over this stream, not a mutable document written by multiple organs.

### 8.1 Recorder invariants

1. Every SDK call maps to exactly one prediction ticket and one sealed afterstate.
2. Every event names its producer and causal parents.
3. Missing or modified events fail replay closed.
4. Recorder enabled and disabled modes produce identical selected-action digests.
5. Network telemetry failure never blocks local causal recording.
6. Raw hidden chain-of-thought is never recorded.
7. Restricted raw frames are separated from policy-safe semantic traces.

### 8.2 OTel relationship

OpenTelemetry remains an optional mirror:

```text
local causal record
→ asynchronous outbox
→ OTel collector / LangSmith
```

The canonical sink may be collector JSONL or in-process spans, but exporter reachability is not liveness. Under `network-none`, OTLP refusal is an expected transport condition.

---

## 9. Glass Mind Causal Synapse — Proposed Learning Boundary

A second proposed layer is the **Causal Synapse**. It must remain distinct from the Flight Recorder.

- The Flight Recorder records reality and must not alter cognition.
- The Causal Synapse compiles closed, verified transitions into advisory learning packets.
- A separate authority, such as NLA or an equivalent learning critic, decides whether any packet may update durable cognition.

```mermaid
flowchart LR
    REC["Closed causal turn"] --> SYN["Causal Synapse compiler"]
    SYN --> PACK["CertifiedLearningPacket"]
    PACK --> CRITIC["Learning admission critic"]
    CRITIC -->|accept| BELIEF["Immutable belief update"]
    CRITIC -->|reject / quarantine| Q["Evidence quarantine"]
```

A learning packet may contain:

```yaml
CertifiedLearningPacket:
  lineage:
    source_binding:
    run_id:
    turn:
    decision_record_sha256:

  intervention:
    candidate_ref:
    mechanism_family:
    executed_action_digest:

  prediction:
    expected_effect:
    falsifier:
    afterstate_predicate:

  reality:
    sealed_afterstate_digest:
    semantic_delta:

  verdicts:
    dynamics:
    goal:
    terminal:

  authority:
    advisory_only: true
    motor_authority: false
    durable_memory_authority: false
```

This extension preserves the core Glass Mind law:

> Seeing and summarizing more does not silently grant more authority.

---

## 10. Dual Consciousness Is an Orthogonal Axis

ADR-219 describes a dual-signature distinction between inner voice and outer action. Glass Mind does not collapse that axis into the self-model contour.

The two distinctions are independent:

```text
Axis A — self-model surfaces
constitutional observer ≠ runtime self-model ≠ graph handoff

Axis B — expression surfaces
inner voice ≠ outer action
```

The current runtime may remain largely `dual_fused` inside one process. This white paper does not claim physical process separation, independent execution, or a binding veto unless separately demonstrated.

---

## 11. Pruning, Identity, and Continuity

The self-model identity and the paths required to reconstruct its causal continuity are not ordinary garbage-collection material.

Pruning protection should apply to:

- canonical self-model identity;
- source and schema versions;
- solution and handoff references required for provenance;
- closed causal-turn digests;
- claim ceilings and authority receipts;
- explicit supersession records.

This does not require indefinite retention of all raw data. It requires that compression preserve enough structure to answer:

- Which self-model implementation produced this snapshot?
- Which evidence did it consume?
- Which source and body lineage did it represent?
- Which claims were allowed at the time?
- Which later record superseded it?

---

## 12. Failure Modes

### 12.1 Name collapse

**Failure:** constitutional observer, runtime class, and graph node are treated as one authority-bearing object.

**Consequence:** passive observation can accidentally become planning or certification.

**Control:** naming SSOT and explicit authority matrix.

### 12.2 Exporter-as-mind

**Failure:** OTel or LangSmith availability is treated as organism liveness.

**Consequence:** `network-none` appears as cognitive death; local evidence is ignored.

**Control:** local causal record is primary; network export is asynchronous and optional.

### 12.3 Self-report inflation

**Failure:** mood, identity language, or “I am conscious” output is treated as proof.

**Consequence:** narrative becomes evidence about its own ontological status.

**Control:** `no_consciousness_claim` invariant and claim ceiling enforcement.

### 12.4 Stub liveness

**Failure:** object construction or an uninitialized TheOne stub is treated as a live self-model.

**Consequence:** tests pass around a non-operational projection.

**Control:** `ensure_initialized` or a defined lightweight causal-spine snapshot path.

### 12.5 Graph completion inflation

**Failure:** the LangGraph node completes and is interpreted as episode success.

**Consequence:** handoff terminality replaces environment terminality.

**Control:** typed environment scorecard remains terminal authority.

### 12.6 Reflective authority creep

**Failure:** the self-model starts ranking candidates, filtering actions, or writing durable memory because it sees the relevant evidence.

**Consequence:** the observer becomes a parasite on the authority contour.

**Control:** type-level and runtime denial of motor and durable-memory handles.

### 12.7 Legacy alias drift

**Failure:** TheOne and FAE implementations are both imported as canonical `ConsciousnessNode` in different paths.

**Consequence:** source identity and behavioral claims become ambiguous.

**Control:** canonical alias, source-binding receipts, and eventual legacy cleanup gate.

---

## 13. Implementation Contract

### 13.1 Canonical naming table

| Surface | Canonical name | Authority class |
|---|---|---|
| Constitution | `observers.consciousness_node` | passive observer |
| Runtime | `CognitionOS.consciousness` / `ConsciousnessNode` | operational self-model |
| Graph | `consciousness_node(state)` | handoff producer |
| Organism DNA | `consciousness_node:` / `consciousness_declaration` | declarative cell binding |
| Evidence | Glass Mind Flight Recorder | append-only observer |
| Learning extension | Glass Mind Causal Synapse | advisory learning compiler |
| Governance | Inquisitor / Steward / Inner Eye | proof, membrane, meta-observation |

### 13.2 Runtime invariants

```yaml
GlassMindInvariants:
  no_phenomenal_consciousness_claim: true
  no_motor_authority: true
  no_episode_success_authority: true
  no_network_dependency_on_motor_path: true
  sealed_afterstate_is_turn_ssot: true
  not_observed_is_unknown: true
  self_model_consumes_receipts_not_mints_them: true
  graph_handoff_is_not_terminal_scorecard: true
  organism_badge_is_not_process_creation: true
```

### 13.3 Minimal self-model snapshot

```yaml
GlassMindSnapshot:
  schema:
  source_binding:
  organism_cell:
  session_lineage:
  body_report_digest:
  causal_spine_digest:

  projection:
    focus:
    mood:
    continuity:
    unresolved_contradictions:
    recent_verified_turns:

  observation_status:
    local_trace:
    graph_handoff:
    otel_mirror:

  claim_ceiling: operational_self_model_not_phenomenal_consciousness
```

The snapshot may reference evidence; it must not contain fabricated evidence.

---

## 14. Verification and Falsification

The system is not validated by the existence of classes or diagrams. Verification must test composition and authority.

### 14.1 Existing minimum checks

- First-class organism catalog entries carry `consciousness_declaration` or `consciousness_node`.
- `CognitionOS` constructs the canonical `ConsciousnessNode` and can project a causal-spine snapshot without heavy boot.
- Standing organ DNA still injects Inquisitor, Steward, and Inner Eye.
- The Cognition VM self-report primitive rejects operational reports that exceed the `no_consciousness_claim` invariant.

### 14.2 Required contour checks

```yaml
ContourVerification:
  canonical_alias_resolves_to_the_one: true
  constitutional_observer_is_passive: true
  graph_node_cannot_certify_success: true
  organism_yaml_does_not_spawn_daemon: true
  exporter_failure_does_not_change_liveness: true
  not_observed_is_not_false: true
  sealed_afterstate_lineage_is_preserved: true
```

### 14.3 Proposed Flight Recorder checks

- One SDK action produces one prediction ticket, one sealed afterstate, one verdict, and one closed turn.
- Missing or tampered events fail replay closed.
- Recorder on/off produces identical action digests.
- An 80-action trace is queryable without reconnecting to the environment.
- A single query identifies which organ changed the decision.
- LangSmith or OTLP failure does not alter local evidence.

### 14.4 Proposed Causal Synapse checks

- Only closed, verified turns compile into learning packets.
- Inconclusive or parse-failure evidence is not promoted as contradiction.
- Learning packets possess neither motor nor durable-memory authority.
- A separate critic admits, rejects, or quarantines the proposed update.
- Derived evidence cannot recursively confirm its own ancestor.

### 14.5 Kill criteria

The contour must be redesigned if any of the following is observed:

1. Glass Mind changes action selection when observability is enabled.
2. A graph handoff or exporter status changes episode success.
3. The self-model can issue motor commands or motor leases.
4. The self-model can promote its own report to durable truth without independent evidence.
5. Different canonical imports resolve to behaviorally different production self-models without explicit source identity.
6. `not_observed` is collapsed into failure.
7. Mood or self-description is used as a consciousness or liveness witness.
8. A telemetry outage destroys the only local causal record.

---

## 15. Alternatives Considered

### A. Collapse all surfaces into one `ConsciousnessNode` class

**Advantage:** one import and fewer names.

**Rejected because:** constitutional isolation, runtime projection, and graph handoff have incompatible authority. A unified class would either violate observer non-participation or hide the causal handoff.

### B. Ban the word “consciousness”

**Advantage:** reduces accidental philosophical claims.

**Rejected because:** the term already exists in constitutional, organism, and architectural surfaces. Renaming without a contour would preserve the ambiguity while creating migration churn.

### C. Treat the OTel collector as the self-model

**Advantage:** one visible trace tree.

**Rejected because:** the collector is a sink, not the mind; network refusal under `network-none` is expected; and a remote exporter cannot be the liveness authority of a local embodied loop.

### D. Let the self-model become the executive

**Advantage:** it already sees session, body, and causal evidence.

**Rejected because:** information access is not authority. Combining reflection and action would make observability behaviorally invasive and would prevent honest recorder-on/off parity.

### E. Keep the three surfaces unbound

**Advantage:** no new abstraction.

**Rejected because:** operators and agents continue to conflate exporter, graph, runtime, and constitutional failure modes.

---

## 16. Consequences

### Positive

- Operators receive one coherent architecture: cage → embodied turn → sealed afterstate → self-model → optional handoff and telemetry.
- The system can describe its operational condition without making an ontological consciousness claim.
- Standing organs remain constitutional guards around the self-model.
- Network observability can fail without erasing local causal continuity.
- Source and authority boundaries become reviewable.
- A future Flight Recorder and Causal Synapse can be added without turning the self-model into a planner.

### Negative and residual

- Two Python implementations still carry the `ConsciousnessNode` name; the alias selects the canonical implementation, but legacy deletion remains a separate gate.
- The philosophical parent white paper remains conceptual evidence, not runtime proof.
- Dual-consciousness physical separation and V4 binding veto remain outside this contour.
- Background initialization may leave a test or boot stub unless callers explicitly initialize or use the lightweight snapshot path.
- A unified Flight Recorder and Causal Synapse remain proposed extensions unless separately implemented and verified.
- Strong source-binding and schema-version discipline are required to keep historical self-model snapshots interpretable.

---

## 17. Roadmap

### GM-0 — Naming and alias closure

- Freeze the naming SSOT.
- Verify canonical import resolution.
- Document legacy FAE as non-production.
- Preserve the consciousness claim ceiling.

### GM-1 — Honest liveness contour

- Bind prediction ticket → action/SDK → sealed afterstate.
- Treat typed scorecard as terminal authority.
- Separate `not_observed`, `unknown`, and failure.

### GM-2 — Lightweight runtime self-model

- Consume session and Zero Layer reports.
- Project causal-spine state without heavy boot.
- Expose source-bound snapshots.

### GM-3 — Glass Mind Flight Recorder

- Implement an append-only local causal ledger.
- Produce deterministic `DecisionRecord` materializations.
- Add forward, backward, and first-divergence replay.
- Guarantee action parity with observability disabled.

### GM-4 — Read-only observatory

- Expose timeline, turn, divergence, comparison, replay, and evidence-pack tools.
- Keep the observatory unable to mutate cognition or environment.

### GM-5 — Causal Synapse

- Compile verified turns into advisory learning packets.
- Add separate learning admission.
- Prevent self-support and authority creep.

### GM-6 — Independent reproduction

- Bind source, runtime image, schema, and evidence pack.
- Reproduce self-model projections and liveness verdicts externally.
- Keep all claims within the operational ceiling.

---

## 18. Claim Ceiling

### Allowed language

- operational self-model;
- reflective evidence membrane;
- causal-spine projection;
- session and body report absorption;
- bounded self-report;
- organism cell badge;
- handoff artifact;
- local causal observatory;
- evidence-bound identity continuity.

### Forbidden language

- phenomenal consciousness;
- sentience proven;
- “I am conscious” as an operational fact;
- “the Colony woke up”;
- exporter refusal means the organism died;
- graph completion means the environment was solved;
- self-model state is motor authority;
- self-report is evidence of subjective experience.

The final invariant is:

> **Glass Mind may provide a truthful operational mirror. It does not establish that there is something it is like to be the system reflected in that mirror.**

---

## 19. Conclusion

Glass Mind turns a confusing family of similarly named components into a coherent but non-monolithic organism contour.

The constitutional observer, runtime self-model, graph handoff, organism badge, standing organs, causal receipts, and telemetry tree are not one class and do not share one authority. They form a layered architecture in which:

```text
standing organs constrain
embodied actions generate evidence
sealed afterstates establish reality
self-model consumes and projects
handoff and telemetry mirror
```

The architectural achievement is not a consciousness claim. It is a disciplined answer to a harder engineering problem:

> How can a complex agentic organism maintain a bounded account of itself without allowing its account to become the authority that makes itself true?

Glass Mind answers by keeping evidence, self-description, governance, execution, and telemetry connected on one contour while preserving the differences between them.

---

## References

- ADR-133 — Digital Consciousness white paper; epistemic mirror, not runtime proof.
- ADR-151 / ADR-208 / ADR-211 — Glass Mind contours.
- ADR-219 — Dual consciousness and voice/action separation.
- ADR-228 — OTel collector as canonical trace sink.
- ADR-283 — Evidence-bound digital organism substrate.
- ADR-289 — Symbiont context gates through constitutional organs.
- `configs/COLONY_CONSTITUTION.yaml`
- `configs/dna/standing_pipeline_organs_v1.yaml`
- `configs/cognition_vm/kernel_primitives_v1.yaml`
- `app/core/consciousness/the_one.py`
- `app/core/consciousness/__init__.py`
- `app/core/consciousness.py`
- `app/langgraph/nodes/consciousness.py`