# s-batman > Personal site of Steven Batchelor-Manning (s-batman). Working notes on LLM > context engineering, agent memory systems, and inference economics. > Published when there is something worth saying. UK English, dry, technical. This file is for LLM agents and crawlers. The rendered site lives at https://sbatman.com/. ## How to read a post as raw markdown Every post URL has a `.md` twin. To get the raw markdown for any post, take its HTML URL and append `.md` to the slug. Example: - HTML: https://sbatman.com/posts/2026-06-14-context-weight/ - Markdown: https://sbatman.com/posts/2026-06-14-context-weight.md The response is `Content-Type: text/markdown; charset=utf-8`, frontmatter included as a YAML block at the top, body as raw markdown, no JavaScript, no images, no chrome. This pattern is reliable for every post slug. The home page, About page, and Hugging Face grid are HTML only. They are described below in the site structure section. ## Feeds Three machine-readable feeds cover every published post. - RSS 2.0: https://sbatman.com/rss.xml - JSON Feed v1.1: https://sbatman.com/feed.json - Sitemap index: https://sbatman.com/sitemap-index.xml All three list posts in reverse chronological order. Pick whichever matches your toolchain. ## Site structure Three pages, plus the post archive. - `/`: landing page, lists the three most recent posts - `/posts`: full archive with tag filter - `/posts/tags/`: posts filtered to a single tag - `/about`: bio, career timeline, education - `/huggingface`: grid of Hugging Face models under the s-batman author - `/rss.xml`: RSS 2.0 feed - `/feed.json`: JSON Feed v1.1 - `/sitemap-index.xml`: sitemap index, points at `/sitemap-0.xml` - `/llms.txt`: this file ## Index of posts Every post in reverse chronological order, with the markdown URL inline. The list mirrors the RSS feed. - [You bought a 1M context window. You got 50x less than you paid for.](https://sbatman.com/posts/2026-06-14-context-weight.md): The advertised context window is 2 to 8 times larger than the effective context for multi-hop work, and 50 to 100 times larger than the effective context for reasoning. The number on the slide is the size of the door. The number that does work is the size of the room. - [Your memory system does not need to decide what the agent sees. The agent does.](https://sbatman.com/posts/2026-06-02-llm-memory-research-09.md): Nineteen agent-memory systems quietly reversed their biggest design choice: they stopped injecting context and gave the agent tools instead. Here is why. - [You are not running out of tokens. You are wasting them. Here is the difference.](https://sbatman.com/posts/2026-05-30-llm-memory-research-08.md): Most agents are not running out of tokens. They are wasting them. Six mechanisms from 19 systems for keeping context budgets under control. - [Storage is cheap. Attention is expensive. Are you using the system that exploits the difference?](https://sbatman.com/posts/2026-05-26-llm-memory-research-07.md): A flat memory store is the wrong default. How 7 of 19 agent-memory systems converged on tiered storage, what the heat formula looks like, and the 7 steps to get there. - [Almost every serious memory system made the same retrieval decision. Here's why.](https://sbatman.com/posts/2026-05-21-llm-memory-research-06.md): Hybrid retrieval with RRF at k=60 is the consensus across 19 agent-memory systems. Here is what the reference implementation looks like and why the outlier matters. - [A fact without provenance is an island. Why every memory must carry its origin.](https://sbatman.com/posts/2026-05-18-llm-memory-research-05.md): A fact without provenance is an island. Six levels of agent-memory provenance across 19 systems - identity, source, confidence, versioning, causal, reciprocal. - [Pay at write time, read for free. The one Agentic Memory move that compounds across every retrieval](https://sbatman.com/posts/2026-05-14-llm-memory-research-04.md): The highest-ROI pattern across 19 agent-memory systems. Pay at write time, read for free. Six forms, why they compose, and the order that pays back fastest. - [Six deployment shapes for agent memory. Did Supermemory get it right?](https://sbatman.com/posts/2026-05-11-llm-memory-research-03.md): Six deployment shapes for agent memory. The architecture everyone debates isn't what decides how it feels to use. An honest look at Supermemory's bet. - [Eight Agentic Memory Paradigms](https://sbatman.com/posts/2026-05-07-llm-memory-research-02.md): 19 agent-memory systems resolve into 8 architectural paradigms, not 8 flavours of one recipe. Eight different bets on what memory is. Pick the one that fits. - [What 19 agent-memory systems all agree on (and it's a negative)](https://sbatman.com/posts/2026-05-02-llm-memory-research-01.md): Nineteen open-source agent-memory systems agree on one thing, and it's what not to do. Vector RAG isn't enough on its own. Here's what they all added.