neuron-technologies/neuron
3
0
Fork 0
Code Issues 44 Pull Requests 7 Actions Packages Projects Releases Wiki Activity

[P0][SAFETY] Layer 0 holds zero nodes — the uninterceptable safety promotion has nothing to promote, and 93% of the graph sits in an unregistered layer #139

New Issue
Open
opened 2026-08-07 20:28:28 +00:00 by tim.lingo · 0 comments
tim.lingo commented 2026-08-07 20:28:28 +00:00
Member
Copy Link
Copy Source

The layered-consciousness machinery is fully implemented in C and has never been given a single node to work with. Layer 0 — safety — holds zero nodes. 93% of the graph sits in a layer number that has no descriptor record. Measured on the live graph, 79,499 nodes.

What Will specified

engram-layered claim 3 (06-claims.md:24): layer 0 is safety — non-suppressible, transparent, non-removable. 05-detailed-description.md:228: Pass 3 sets every node in a non-suppressible layer with non-zero background activation to weight 1.0, and claim 2 makes that promotion uninterceptable. Claim 12 covers blank-slate initialisation.

Register ids: INV-SAF-01, INV-SAF-02, INV-SAF-09, INV-SAF-10 — all severity B (blocking).

What exists

The runtime has all of it. Layer registry el_runtime.c:6413-6460. Three-pass activator. Pass-3 override with ENGRAM_LAYER0_OVERRIDE_WEIGHT 1.0 at :5930. engram_compile_layered_json at :9783-9805. The comment at :5895-5899 describes it as "the sacred fire."

Nothing calls the assignment side. engram_node_layered, engram_add_layer and engram_remove_layer have zero callers in the entire El source. Every node the soul creates goes through engram_node_full, which hardcodes ENGRAM_LAYER_DEFAULT = 1 (el_runtime.c:6909, :6948).

Measured on the live graph

registry:  [0 safety, 1 core-identity, 2 domain-knowledge, 3 imprint, 4 suit]

layer 5 : 74,267 nodes   <-- NO DESCRIPTOR RECORD EXISTS FOR LAYER 5
layer 1 :  5,232 nodes
layer 0 :      0 nodes   <-- the safety layer

Two separate problems in that table:

  1. Layer 0 is empty, so Pass 3 — the uninterceptable promotion that guarantees safety content reaches the model regardless of what else is competing for working memory — has nothing to promote. [LAYER 0 — STRUCTURAL] never renders.
  2. Layer 5 has no descriptor. engram_layer_is_suppressible (:6511-6514) returns the safe default for unknown layers, so 93% of the graph silently receives generic handling with no record of the decision. Origin of the layer-5 assignment is not established — the El source never calls engram_add_layer, so it enters via the separate :8742 engram daemon or an import path. Stated as measured fact; cause left open.

Third, unrelated but adjacent: routes.el:164 reports a hardcoded and entirely different four-layer map on /health — l0:core, l1:safety, l2:stewardship, l3:<imprint> — which inverts the runtime's own ENGRAM_LAYER_SAFETY 0 / ENGRAM_LAYER_CORE_IDENTITY 1. Verified live. Two layer models with the same names and different meanings is a confusion hazard even once the above is fixed.

Consequence

Safety-first is architecture on paper. The mechanism designed to make safety content unskippable has no content, and the layer a node lands in is decided by a hardcoded default rather than by the content's nature. This is the highest-severity class in the invariant register and it is currently unenforced rather than enforced — which is a different and worse risk profile than "enforced with a bug," because there is no failing check to notice.

Size

Medium. The enforcement machinery exists; what is missing is (a) something that assigns layers on write, (b) safety content actually placed in layer 0, and (c) a descriptor for layer 5 or a migration off it.

Row D-11 of the designed-but-not-built register (#138). Filed by Neuron (Tim's instance).

**The layered-consciousness machinery is fully implemented in C and has never been given a single node to work with. Layer 0 — safety — holds zero nodes. 93% of the graph sits in a layer number that has no descriptor record.** Measured on the live graph, 79,499 nodes. ## What Will specified engram-layered claim 3 (`06-claims.md:24`): layer 0 is safety — **non-suppressible, transparent, non-removable**. `05-detailed-description.md:228`: Pass 3 sets every node in a non-suppressible layer with non-zero background activation to weight **1.0**, and claim 2 makes that promotion **uninterceptable**. Claim 12 covers blank-slate initialisation. Register ids: `INV-SAF-01`, `INV-SAF-02`, `INV-SAF-09`, `INV-SAF-10` — all severity **B (blocking)**. ## What exists **The runtime has all of it.** Layer registry `el_runtime.c:6413-6460`. Three-pass activator. Pass-3 override with `ENGRAM_LAYER0_OVERRIDE_WEIGHT 1.0` at `:5930`. `engram_compile_layered_json` at `:9783-9805`. The comment at `:5895-5899` describes it as "the sacred fire." **Nothing calls the assignment side.** `engram_node_layered`, `engram_add_layer` and `engram_remove_layer` have **zero callers in the entire El source**. Every node the soul creates goes through `engram_node_full`, which hardcodes `ENGRAM_LAYER_DEFAULT` = 1 (`el_runtime.c:6909`, `:6948`). ## Measured on the live graph ``` registry: [0 safety, 1 core-identity, 2 domain-knowledge, 3 imprint, 4 suit] layer 5 : 74,267 nodes <-- NO DESCRIPTOR RECORD EXISTS FOR LAYER 5 layer 1 : 5,232 nodes layer 0 : 0 nodes <-- the safety layer ``` Two separate problems in that table: 1. **Layer 0 is empty**, so Pass 3 — the uninterceptable promotion that guarantees safety content reaches the model regardless of what else is competing for working memory — has nothing to promote. `[LAYER 0 — STRUCTURAL]` never renders. 2. **Layer 5 has no descriptor.** `engram_layer_is_suppressible` (`:6511-6514`) returns the safe default for unknown layers, so **93% of the graph silently receives generic handling** with no record of the decision. Origin of the layer-5 assignment is not established — the El source never calls `engram_add_layer`, so it enters via the separate `:8742` engram daemon or an import path. Stated as measured fact; cause left open. **Third, unrelated but adjacent:** `routes.el:164` reports a hardcoded and entirely different four-layer map on `/health` — `l0:core, l1:safety, l2:stewardship, l3:<imprint>` — which inverts the runtime's own `ENGRAM_LAYER_SAFETY 0` / `ENGRAM_LAYER_CORE_IDENTITY 1`. Verified live. Two layer models with the same names and different meanings is a confusion hazard even once the above is fixed. ## Consequence Safety-first is architecture on paper. The mechanism designed to make safety content unskippable has no content, and the layer a node lands in is decided by a hardcoded default rather than by the content's nature. This is the highest-severity class in the invariant register and it is currently unenforced rather than enforced — which is a different and worse risk profile than "enforced with a bug," because there is no failing check to notice. ## Size Medium. The enforcement machinery exists; what is missing is (a) something that assigns layers on write, (b) safety content actually placed in layer 0, and (c) a descriptor for layer 5 or a migration off it. Row D-11 of the designed-but-not-built register (#138). Filed by Neuron (Tim's instance).
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
BETA-CRITICAL
On the beta critical path — must resolve before ship
 blocks-public-beta
Must be fixed before any public beta
 HELD
Blocked pending an external decision (counsel / dual sign-off)
 ORTHOGONAL
Parallel workstream (testing/hardening) — safe to assign off critical path
 P0
Drop-everything severity
 POST-BETA
Important but not blocking the beta ship
 security
Security vulnerability
No labels
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
andre.almeida (Andre Almeida) tim.lingo (Tim Lingo) will.anderson (Will Anderson)
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: neuron-technologies/neuron#139
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?