Closing the Agent Loop A research program in governed agentic operations

Task loops complete.
Governed loops close.

Almost everything that goes wrong at scale in agentic operations goes wrong in the gap between those two sentences. This lane defines operational closure — and builds the instruments to test it.

the loop closes: outcome and memory update linked
Context packet scoped, traced Agent work reason and act Evidence outcome + trace Proposal append-first Promotion gate policy/human approves Governed memory epistemic state task complete — context discarded
Loops closed
0
Context lost
0
Megan Anderson · AI ARMY, INC. · aiarmy.co · 2026
01 · The symptoms, re-described

The field named its pains separately

Context rot Lossy re-encoding at handoff seams — the loop losing state each time it passes through a summary.
Memory poisoning An ungoverned write path — the loop accepting claims into authority without promotion.
Multi-agent token multiplier Broadcast in place of scoped projection — paying transmission for context that carries no epistemics about what each worker needs.
Sycophancy entrenchment Compression itself is the accomplice: selection by fluency, similarity, and recency pre-agrees the information space before the model ever speaks.

Each is what one part of an un-closed loop looks like from outside. Different symptoms; one shared failure surface; one grammar of repair — status · provenance · supersession · scope · closure — carried on the objects that move through the loop.

02 · The research agenda

Seven open problems

Questions the program is instrumented to study and does not claim to have answered — every branch feeds the same spine: operational closure at scale.

1

The depth law of drift

Does context degradation compound multiplicatively with delegation depth while cost merely adds — and what is the exponent? If provenance-carrying packets flatten the curve, by how much, at what packing cost?

Tier 4 · inference-economics paper
click a branch to read each problem — stated so that results, ours or anyone's, can settle it
03 · Four commitments

The research program is built on four core positions.

01

Closure means something precise — and it hasn't been defined

Completion is a property of a task; closure is a property of an operation. An operation is closed when what was known, who authorized it, what was done, what resulted, and what memory now says are one traceable object.

02

Sovereign and portable at the same time

The context packet refuses to trade one for the other: a bounded, governed, portable assembly of context prepared for a specific recipient, task, and moment — the right context, at the right time. The corpus stays home; the projection travels.

03

Integrity lives in the coupling

Every response is elicited from a unique state space — prompt, memory, tools, environment. The answer was never a property of the model alone. So drift, bloat, and sycophancy are attacked by engineering what enters the loop: epistemic agentic memory, with status, provenance, supersession, stakes, and validity carried on the objects themselves.

04

Humans stay accountable; the loop stays traceable

Agents propose; promotion into durable state is a governed, observable, human-accountable act. Every claim traces to the packet that scoped the work that produced it. Judgment is routed, not automated away.

Together these define the category the series builds toward: governed agentic operations — and the layer that enables them, governed agentic infrastructure.

04 · The instruments

The series, in reading order

Published here first, then Zenodo, then arXiv.

Series capstone · agenda paper

Open Problems at the Agentic Memory Wall

A Research Agenda for Governed Agentic Operations

States the four commitments, re-describes the field's symptoms as failures of closure, and poses the seven problems above — each falsifiably, mapped to the paper and benchmark tier designed to test it.

Megan Anderson · AI ARMY, INC. · aiarmy.co · 2026

SITE · LIVE
v1.0 · 2026-07 DOI · pending arXiv · pending Zenodo · pending
1 SITE · LIVE
Closing the Agent Loop

Integrity Is a Property of the Coupling & Trust Is Symbiotic with the Environment in Agentic Operations

Defines the category and the closure claim.

v1.0 · 2026-07DOI · arXiv · Zenodo — pending PDF ↓
2 SITE · LIVE
Specification

Context Packet Specification v0.4.0

Defines the primitive that travels — the right context, at the right time.

v0.4.0 · 2026-07DOI · arXiv · Zenodo — pending PDF ↓
3 SITE · LIVE
Closing the Agent Loop

Governance Starts at Capture: An Epistemic Memory Architecture for Governed Agentic Operations

Defines the epistemic agentic memory the loop writes into.

v1.0 · 2026-07DOI · arXiv · Zenodo — pending PDF ↓
4 SITE · LIVE
Closing the Agent Loop

Epistemic Context Packets as an Inference-Economics Primitive: Valid Handoff and the Agentic Memory Wall

Prices the loop — valid handoff as the economic unit; pre-registers the program's falsification conditions.

v1.0 · 2026-07DOI · arXiv · Zenodo — pending PDF ↓
05 · Epistemic agentic memory

Memory as a constellation; shaping the information space

Memory objects are stored as stars or nodes; provenance links are the bridges between them, creating clusters or constellations over time. Unrelated objects or ungoverned objects remain isolated but accessible until promoted to a cluster. A packet is constructed from the information field of constellations for the right context at the right time. The memory stays home; the packet travels.

44objects
three established constellations · 44 objects, provenance intact · nothing travels until a task asks
Cite this work · for researchers

Reference the research program

Anderson, Megan. Accountable Agentic Reasoning: A Cross-Domain Constraint Architecture for Events, Cognition, Information Geometry, and Autonomous AI Systems. Research program, AI ARMY, INC., aiarmy.co/research, 2026.
[1]Anderson, M. "Closing the Agent Loop. Integrity Is a Property of the Coupling & Trust Is Symbiotic with the Environment in Agentic Operations." AI ARMY, 2026.SITE · LIVEDOI · arXiv pending
[2]Anderson, M. "Context Packet Specification v0.4.0." AI ARMY, 2026.SITE · LIVEDOI · arXiv pending
[3]Anderson, M. "Closing the Agent Loop. Governance Starts at Capture: An Epistemic Memory Architecture for Governed Agentic Operations." AI ARMY, 2026.SITE · LIVEDOI · arXiv pending
[4]Anderson, M. "Closing the Agent Loop. Epistemic Context Packets as an Inference-Economics Primitive: Valid Handoff and the Agentic Memory Wall." AI ARMY, 2026.SITE · LIVEDOI · arXiv pending
[5]Anderson, M. "Open Problems at the Agentic Memory Wall: A Research Agenda for Governed Agentic Operations." AI ARMY, 2026.SITE · LIVEDOI · arXiv pending
06 · Follow the series

Governance is not overhead — it is how agentic systems preserve the conditions of valid action

The harness, task sets, and scoring schema are being prepared for release.

Read the agenda paper → Follow the series →
Megan Anderson · AI ARMY, INC. · aiarmy.co Closing the Agent Loop · governed agentic operations · 2026