state-tracking CAP-0046 · activeTracking state through a long task
Models lose track of the current state of entities, files, or an environment after many steps.
Also called: entity tracking, world-model consistency, stale state
Tags: autonomous-agent, coding-agent, context, data-analysis
A capable model maintains an accurate picture of the current state of what it is working on, updating it after every action, so that later decisions rest on what is true now rather than what was true earlier.
Claims
- mechanismsingle paper↑1↓0In interactive video world models, world state communicated only through prompt switching is not reliably preserved across a rollout — characters that should stay alive vanish and killed ones keep moving — whereas keeping the canonical state in an external deterministic engine and compiling it into pixel-aligned spatial controls for the video model keeps the rendered frames consistent with that state.
- mechanismsingle paper↑1↓0For long cooking videos where a question depends on ingredient changes no longer visible, vision-language models fail to carry observed events forward into later answers, and adding generic retrieved-context RAG makes it worse, while splitting memory into a temporally anchored event log plus an explicitly updated inventory ledger recovers much of the gap.
- mechanismsingle paper↑1↓0Models that predict next steps well can still hold an incoherent implicit world model, which fails when the task deviates from familiar traces.
- mechanismsingle paper↑1↓0Only models with substantial code pretraining track entity state through a sequence of operations, and all degrade as the sequence lengthens.
- observationsingle paper↑1↓0In partially observable text environments (ALFWorld, ScienceWorld), letting an LLM agent query an externally maintained state estimate that keeps an explicit distribution over unobserved object locations raises task success more than querying a deterministic memory of what has already been observed, and the gain shrinks to near zero on a frontier model that already nearly saturates the benchmark.
Techniques
- Checkpoint task state to a fileprocessWrite task id, status, completed steps and artifact paths to a small file after every meaningful step, and read it at session start.
- Keep world state in an engine, use video model only as rendearchitectureAn agent compiles natural-language instructions into executable programs whose state a lightweight engine maintains as state-augmented 3D oriented bounding boxes, which are deterministically projected into identity, semantic and motion cont
- Re-read before editingprocessForce the agent to read the current version of a file immediately before changing it.
Sources
Every paper, post or observation behind this capability. Ones cited by a claim show the claim and its stance; ones filed directly are about the capability but have no claim drafted from them yet. Open one to add support or contest it.
- Programmable World Model2026-09-09↑ supports a claim here
- ↑ supports a claim here
- ↑ supports a claim here
- Evaluating the World Model Implicit in a Generative ModelThis source: Evaluating the World Model Implicit in a Generative ModelHow much the field cites it — heavily cited in the last 12 months75 in the last 12 months · 144 totalpublished 2024checked 2026-09-042024↑ supports a claim here
- Entity Tracking in Language ModelsThis source: Entity Tracking in Language ModelsHow much the field cites it — steadily cited in the last 12 months14 in the last 12 months · 43 totalpublished 2023checked 2026-09-04click for Semantic Scholar2023↑ supports a claim here
Capabilities are a way of carving up the subject, and carvings are arguable. Say so if this one is wrong — especially a proposed one, which a pipeline added because several papers used the same framing, not because anyone decided it was right.