INFERENCE DIAGRAM
/ MAIN 5-1-5
UPPER • 1
Source
prompt, user, tool
UPPER • 2
Why
intent, goal
UPPER • 3
Context
RAG, memory
UPPER • 4
Constraint
policy, budget
UPPER • 5
Human
oversight
CORE
INFERENCE ENGINE
Attention Heads
GENERATING
LOWER • 1
Transform
parse, tools
LOWER • 2
Verify
ground, safety
LOWER • 3
Log
10 artifacts
LOWER • 4
Release
output, stream
LOWER • 5
Feedback
RLHF, reward
INPUT LAYER
OUTPUT LAYER
DATA FLOW:
1. Prompt → Source
›
2. Why/Context/Constraint/Human → Core
›
3. Tokens
›
4. Transform → Verify → Log (10) → Release
›
5. Feedback loop
LOG • 10 ARTIFACTS (per inference)
01 prompt
02 why-vector
03 context-hash
04 constraints
05 human-pref
06 attention-map
07 token-probs
08 transform-ops
09 verify-score
10 release-manifest