# Reels-RSI: Program-Level Self-Improvement for Code-Rendered Explainer Videos

arXiv: https://arxiv.org/abs/2610.14495 · HTML: https://mlsanity.com/paper/2610.14495/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 23 of its 24 references that are in the map.
- First posted on arXiv 10 Oct 2026.

## Builds on (most cited first, 23 of 23)

- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Reflexion: Language Agents with Verbal Reinforcement Learning](https://mlsanity.com/paper/2303.11366/) (2023)
- [Voyager: An Open-Ended Embodied Agent with Large Language Models](https://mlsanity.com/paper/2305.16291/) (2023)
- [ExpeL: LLM Agents Are Experiential Learners](https://mlsanity.com/paper/2308.10144/) (2023)
- [GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning](https://mlsanity.com/paper/2507.19457/) (2025)
- [Automated Design of Agentic Systems](https://mlsanity.com/paper/2408.08435/) (2024)
- [Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models](https://mlsanity.com/paper/2510.04618/) (2025)
- [Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents](https://mlsanity.com/paper/2505.22954/) (2025)
- [VBench-2.0: Advancing Video Generation Benchmark Suite for Intrinsic Faithfulness](https://mlsanity.com/paper/2503.21755/) (2025)
- [Inducing Programmatic Skills for Agentic Tasks](https://mlsanity.com/paper/2504.06821/) (2025)
- [SkillRevise: Improving LLM-Authored Agent Skills via Trace-Conditioned Skill Revision](https://mlsanity.com/paper/2606.01139/) (2026)
- [Code2Video: A Code-centric Paradigm for Educational Video Generation](https://mlsanity.com/paper/2510.01174/) (2025)
- [TheoremExplainAgent: Towards Video-based Multimodal Explanations for LLM Theorem Understanding](https://mlsanity.com/paper/2502.19400/) (2025)
- [KNighter: Transforming Static Analysis with LLM-Synthesized Checkers](https://mlsanity.com/paper/2503.09002/) (2025)
- [More Convincing, Not More Correct: Self-Play Reward Hacking of Reference-Free LLM Judges](https://mlsanity.com/paper/2607.05904/) (2026)
- [VideoWeaver: Evaluating and Evolving Skills for Agentic Long Video Generation](https://mlsanity.com/paper/2606.08091/) (2026)
- [On the Fragility of Self-Improving Agents: Variance, Task Order, and Underspecification](https://mlsanity.com/paper/2608.18066/) (2026)
- [LLM-as-a-Judge Is Not an Oracle: Why Self-Improving Agents Need Deterministic Guardrails](https://mlsanity.com/paper/2609.02246/) (2026)
- [Write Your Own CodeChecker: An Automated Test-Driven Checker Development Approach with LLMs](https://mlsanity.com/paper/2411.06796/) (2024)
- [Aligning Progress and Feasibility: A Neuro-Symbolic Dual Memory Framework for Long-Horizon LLM Agents](https://mlsanity.com/paper/2604.02734/) (2026)
- [ManimAgent: Self-Evolving Multimodal Agents for Visual Education](https://mlsanity.com/paper/2606.30296/) (2026)
- [Findings of the First Teaching Monster Challenge: A Benchmark of Pedagogical Content Knowledge in AI Agents](https://mlsanity.com/paper/2608.08852/) (2026)
- [CineForge: Self-Improving Agents for Long-Horizon Video Generation](https://mlsanity.com/paper/2608.29621/) (2026)

## Abstract (arXiv)

Explainer videos -- a worked geometry problem, why the sky is blue, how a dynasty fell -- are usually one more target for pixel video generation. We argue they are a different kind of artifact: a program (script, storyboard and per-frame animation code) that a deterministic renderer turns into frames. A program can be read, diffed, tested and audited; pixels can only be watched. Reels-RSI is an open-source reel director for any agent harness; its self-improvement happens at the program level rather than in weights or prompts. Every revision pass is logged with a code snapshot; a retrospective pairs each check finding with the diff that fixed it and proposes lessons; a lesson that a static check can catch is compiled into a rule with a must-fail and a must-pass example, so the memory is unit-tested; and a rule learned today runs over everything made before it. A bounded, benchmark-gated step lets an agent rewrite the director's playbook, kept only when blind pairwise judgments on held-out topics prefer it; nothing enters memory without a human decision. Across 111 reels -- a field study, two playbook evolutions, a 2x2 memory ablation, a weaker director, a 48-reel learning curve and a judge-reliability study -- we find that the rule learned in the field caught four latent defects in frames authored before it existed and none after; that rules are an audit for a strong director (eight unguarded reels, no violation) and a guard for a weaker one (thirty violations unguarded versus five); that lessons shorten revision loops but do not compound or move the judge; and that the first rewrite's 11--1 verdict was a position bias in our vote design (judges pick the second-shown reel 62--81% of the time), masking scripts that traded thirteen facts for craft. We document four failure modes of the preference signal, which check caught each, and the fixes. Code, rules, lessons and data are public.