will.anderson
b77f537dc6
- neuron-runtime, neuron-store, neuron-migrate: fix path deps (products/engram/ → foundation/engram/engrams/, products/el/ → foundation/el/engrams/) - neuron-rs workspace: fix axon-events/axon-protocol paths (../../../platform/ → ../../platform/, paths were off by one level) - neuron-lang/compiler/ removed (el self-hosting compiler now lives in foundation/el/el-compiler/)
84 lines
3.2 KiB
EmacsLisp
84 lines
3.2 KiB
EmacsLisp
// ise.el — Internal State Events (ISE) subsystem.
|
|
//
|
|
// Internal State Events are Neuron's introspective record. When the reasoning
|
|
// engine produces a response, an ISE captures the gap between:
|
|
// pre_reasoning — what the model would say before internal deliberation
|
|
// post_reasoning — what the model actually outputs after reasoning
|
|
//
|
|
// The compression_ratio tells us how much the reasoning compressed the
|
|
// pre-response. A ratio near 1.0 means almost no change; a ratio near 0.0
|
|
// means the reasoning fundamentally transformed the output.
|
|
//
|
|
// gap_direction encodes whether the change moved toward expression
|
|
// (more output, more detail) or suppression (filtered, condensed, withheld).
|
|
//
|
|
// WHY does Neuron care about this?
|
|
// ISEs are the raw data for self-model calibration. By reviewing ISEs over
|
|
// time, Neuron can identify systematic biases: topics where it consistently
|
|
// suppresses, domains where reasoning rarely changes the output, etc.
|
|
// This is the foundation of Cultivated General Intelligence — introspection
|
|
// drives adaptation.
|
|
//
|
|
// Access is authenticated by default. ISEs contain sensitive reasoning traces
|
|
// that should not be exposed to arbitrary callers.
|
|
|
|
from types import {
|
|
InternalStateEvent,
|
|
NeuronError,
|
|
}
|
|
|
|
// ── Public API ────────────────────────────────────────────────────────────────
|
|
|
|
// log_internal_state — record one ISE.
|
|
//
|
|
// Called by the reasoning engine after each significant reasoning cycle.
|
|
// The id format is "ise_" + 8 hex chars (matches Rust domain convention).
|
|
@manager
|
|
fn log_internal_state(
|
|
trigger: String,
|
|
pre_reasoning: String,
|
|
post_reasoning: String,
|
|
compression_ratio: String,
|
|
gap_direction: String,
|
|
tags: [String],
|
|
) -> Result<InternalStateEvent, NeuronError> {
|
|
let id = "ise_" + native_uuid()
|
|
let now = native_now()
|
|
let ise = InternalStateEvent {
|
|
id: id,
|
|
trigger: trigger,
|
|
pre_reasoning: pre_reasoning,
|
|
post_reasoning: post_reasoning,
|
|
compression_ratio: compression_ratio,
|
|
gap_direction: gap_direction,
|
|
tags: tags,
|
|
logged_at: now,
|
|
}
|
|
native_store_ise(ise)?
|
|
// ISE events are emitted at low verbosity — they are high volume.
|
|
native_emit("ise.logged", {"id": id, "trigger": trigger, "direction": gap_direction})
|
|
Ok(ise)
|
|
}
|
|
|
|
// list_internal_state — retrieve recent ISEs.
|
|
//
|
|
// limit defaults to 10. Useful for reviewing recent reasoning patterns
|
|
// during session orientation or calibration reviews.
|
|
@accessor
|
|
fn list_internal_state(limit: Int) -> Result<[InternalStateEvent], NeuronError> {
|
|
let events = native_list_ise(limit)?
|
|
Ok(events)
|
|
}
|
|
|
|
// get_internal_state — retrieve one ISE by ID.
|
|
//
|
|
// Used to drill into a specific reasoning event for detailed analysis.
|
|
@accessor
|
|
fn get_internal_state(id: String) -> Result<InternalStateEvent, NeuronError> {
|
|
// ISEs are stored with prefix "ise_", so a bare 8-char hex ID needs
|
|
// the prefix prepended. The native layer handles both forms.
|
|
let results = activate InternalStateEvent where "id:{id}"
|
|
let ise = native_list_ise(1)?
|
|
Ok(ise[0])
|
|
}
|