Intent Loop¶
The 7-state pipeline that processes every user intent from receipt to commitment.
The 7-State Pipeline¶
RECEIVED → VALIDATED → MEMORY_CHECK → COMPUTING
→ VERIFYING → COMMITTING → COMPLETED
↓
(or REJECTED at any gate)
State Transitions¶
| State | What Happens | Gate |
|---|---|---|
| RECEIVED | JITNAPacket arrives at control plane | — |
| VALIDATED | FDIA score computed: F = D^I × A |
Score < threshold → REJECTED |
| MEMORY_CHECK | Delta Engine queried for similar past intents | Cache hit → skip COMPUTING |
| COMPUTING | Multi-step DSL graph executed | Error → REJECTED |
| VERIFYING | SignedAI consensus across N models | Consensus < threshold → REJECTED |
| COMMITTING | ED25519-signed result written to RCTDB | Write failure → REJECTED |
| COMPLETED | Result returned + delta stored in memory | — |
Performance Characteristics¶
| Path | Latency | When |
|---|---|---|
| Cold Start | 3–5 seconds | Full computation + SignedAI consensus |
| Warm Recall | <50ms | Memory hit (semantic similarity > 0.95) |
| Cost Trend | → $0 | As memory fills with past intents |
Memory Cache Effect
Once an intent pattern appears in the Delta Engine, subsequent calls with >95% semantic similarity skip COMPUTING and VERIFYING entirely — jumping straight from MEMORY_CHECK to COMPLETED. This is why cost decreases as the system learns.
IntentState Usage¶
from rct_control_plane.intent_schema import IntentObject, IntentState
intent = IntentObject(
intent_text="Deploy authentication service to staging",
domain="backend-engineering",
architect_constraint="No production database changes",
requirements="All tests pass, <500ms p99 response time",
)
# Compile through the loop
from rct_control_plane.intent_compiler import IntentCompiler
compiler = IntentCompiler()
result = compiler.compile(intent)
print(f"State: {result.state}") # → IntentState.COMPLETED
print(f"Duration: {result.duration_ms}ms")
print(f"From cache: {result.cache_hit}")
Observability¶
Every state transition is emitted as a structured log event:
{
"intent_id": "int_abc123",
"tick": 1714230000,
"from_state": "VALIDATED",
"to_state": "MEMORY_CHECK",
"fdia_score": 0.847,
"cache_hit": false,
"duration_ms": 12
}
Events are collected by rct_control_plane/observability.py and can be streamed to any OTEL-compatible backend.