LLMography in the Age of Loop and Graph Engineering
A working paper extending LLMography from conversation traces to loop- and graph-structured Human–AI decision systems.
Mohammed Bousmah · 2026-08-01 · Working paper
Human–AI decision provenance
LLMography
The science of auditable Human–AI intelligence.
Reconstruct the decision trajectories formed between humans, LLMs, agents, tools, and autonomous loops — so intelligence can be observed, attributed, and audited.
Founded by Dr. Mohammed Bousmah, PhD
Decision trajectory
01 · Seed trace
A decision begins as a short Human → LLM → Decision path.
Every interaction leaves a trace.
The problem
AI decisions are becoming harder to reconstruct.
Modern systems moved from prompts to chains, agents, loops, and multi-agent graphs. Logs show infrastructure. They do not explain how a Human–AI decision emerged.
01
Prompts became chains.
02
Chains became agents.
03
Agents became loops.
04
Loops became decision systems.
Traditional telemetry
Human–AI decision provenance
Request logs, latency, tokens, status codes.
Who asked, what was delegated, what was verified, who decided.
Infrastructure events in time.
Decision events in relation.
What the system did.
How a Human–AI decision emerged.
What is LLMography?
A new analytical layer for Human–AI systems.
LLMography reconstructs Human–AI decision trajectories so they can be observed, attributed, measured, reproduced, explained, and audited.
Layer 1
Interaction provenance
Prompts, replies, tool calls, human interventions, and verification events as they occurred.
Layer 2
Loop provenance
How observation, revision, feedback, and stopping conditions evolve inside a decision loop.
Layer 3
System provenance
How humans, agents, models, tools, memory, policy, and orchestration relate across the system.
Signature object
From interaction logs to decision trajectories.
Hover a node to inspect the metadata LLMography recovers: actor, action, time, and reason.
A human frames a decision, an LLM proposes a path, an agent delegates a tool call, verification rejects an unsupported claim, the human revises, and a decision is issued.. Human — Goal framing. Request issued (Need a defensible recommendation). Prompt — Task specification. Instruction recorded. LLM — Draft reasoning. Reply generated. Agent — Orchestration. Delegation. Tool — Evidence lookup. Executed. Observation — Returned evidence. Observed. Verification — Claim check. Rejected (Unsupported conclusion). Memory — Prior context. Recalled. Policy — Oversight rule. Constraint. Human — Intervention. Revision requested (Unsupported conclusion). Decision — Issued outcome. Approved
Research
Built from research, not marketing.
LLMography begins as a scientific framework. The platform is a research preview derived from that work.
Concepts
The language of auditable Human–AI systems.
Concept
LLMography
The science of reconstructing, measuring, and auditing Human–AI decision trajectories.
Concept
Human–AI decision trajectory
The reconstructed path through which a Human–AI system arrives at a decision.
Concept
Loop engineering
The design of iterative observe–act–verify–revise cycles in agentic systems.
Concept
Graph engineering
The design of stateful, branching execution graphs for agentic work.
Concept
Interaction provenance
The record of prompts, replies, tool calls, interventions, and verification events.
Concept
System provenance
How humans, agents, models, tools, memory, and policy relate across a system.
Evolution
From prompting to auditable decision systems.
01
Prompting
A single human instruction produces a model output.
02
Chains
Multiple sequential LLM operations are composed into a path.
03
Agents
Models select tools and actions rather than only generating text.
04
Agentic AI
Systems pursue goals across steps with partial autonomy.
05
Loop engineering
Systems iterate, observe, verify, and self-correct.
06
Graph engineering
Execution becomes stateful, branching, and structured.
07
LLMography
The resulting Human–AI decision trajectory becomes observable and auditable.
Research preview
Trace. Reconstruct. Measure. Audit.
The LLMography Platform is a research preview for reconstructing Human–AI decision trajectories from traces.
Research preview
Coming soon
Decision trajectory
The reconstructed path from goal to issued decision.
Provenance graph
Actors, tools, memory, and verification as a structured graph.
Human interventions
Where a person asked, revised, approved, or stopped the loop.
Verification events
Checks that accepted, rejected, or returned work.
Agent activity
Delegations, tool calls, and orchestration steps.
Loop metrics
Proposed analytical indicators — not certified scores.
Timeline
Ordered events with residual gaps made visible.
Audit summary
What can be reconstructed, and what cannot.
Research network
Follow the research
Receive major LLMography research updates, papers, and platform announcements.
Research network
Share your details and we will follow up when there is meaningful research or platform news.
Intelligence is becoming collaborative.Its history should not disappear.
LLMography is building the scientific and technical foundations for reconstructing how humans and AI systems reach decisions together.