{"updated_at":"2026-09-06T00:00:00Z","site":"https://chrishayuk.com","author":"Chris Hay","policy":"A record is a draft until it is individually reviewed and published; drafts carry recorded_at and no published_at. material_revision is true only when an earlier version of that record exists, so edits within a version are not revisions. updated_at is derived from the newest record and does not move on its own.","feeds":{"notebook":"https://chrishayuk.com/notebook/feed.xml","record":"https://chrishayuk.com/record/feed.xml","json":"https://chrishayuk.com/feed.json","record_json":"https://chrishayuk.com/record/feed.json"},"record_index":"https://chrishayuk.com/api/records","latest":[{"id":"N-ADDRESS-BUILD","version":"0.1","title":"The address is built through depth.","kind":"notebook","state":"draft","recorded_at":"2026-09-06T00:00:00Z","updated_at":"2026-09-06T00:00:00Z","material_revision":false,"topics":["residual-stream","model-representation","model-memory","factual-recall","activation-patching"],"summary":"For capital versus language on Gemma 3 4B IT, two instruments agree: relation information is readable and causally consequential by L8; entity binding becomes decisive in the final-position residual between L24 and L28. Cross-layer reader alignment recovers relation information across changing coordinates. The L28–30 endpoint may carry the answer token, so the mechanistic evidence window is L8–24.","url":"https://chrishayuk.com/notebook/the-address-is-built-through-depth"},{"id":"N-MAP","version":"0.1","title":"What is the map actually a map of?","kind":"notebook","state":"draft","recorded_at":"2026-09-06T00:00:00Z","updated_at":"2026-09-06T00:00:00Z","material_revision":false,"topics":["residual-stream","ffn","model-representation","model-memory"],"summary":"The map is a three-dimensional view of a transformer’s evolving residual state, shown relative to selected answer-token directions. The film follows Japan towards Tokyo, then swaps France’s state into an Australia prompt. Reading the demonstration alongside its code makes the next questions precise: what is being plotted, which state changes, and what does the model know how to read?","url":"https://chrishayuk.com/notebook/what-is-the-map"},{"id":"N-STATE","version":"0.1","title":"What has to survive?","kind":"notebook","state":"draft","recorded_at":"2026-09-06T00:00:00Z","updated_at":"2026-09-06T00:00:00Z","material_revision":false,"topics":["residual-stream","model-memory","context-engineering"],"summary":"What state must a model retain to continue? This notebook distinguishes all-position residual state, a single token’s residual and persistent decode state. Full-sequence transplantation and MAP-5 test different boundaries; neither justifies compressing every conversation into one vector.","url":"https://chrishayuk.com/notebook/what-has-to-survive"},{"id":"N-ADDRESS","version":"0.2","title":"Reading by address.","kind":"notebook","state":"draft","recorded_at":"2026-09-06T00:00:00Z","updated_at":"2026-09-06T00:00:00Z","material_revision":false,"topics":["ffn","model-as-database","model-memory","model-representation"],"summary":"An address is a pattern a reader can respond to. This notebook follows the operation from six hand-built key–value pairs to Gemma’s evolving state: match a query, combine value directions, then read an answer. The film, source code and saved experiments show both what works and where addressing remains approximate.","url":"https://chrishayuk.com/notebook/reading-by-address"},{"id":"N-AUTHORITY","version":"0.1","title":"Which source wins?","kind":"notebook","state":"draft","recorded_at":"2026-09-06T00:00:00Z","updated_at":"2026-09-06T00:00:00Z","material_revision":false,"topics":["context-engineering","model-memory","source-authority","model-as-database"],"summary":"When two things in one context disagree, which one answers? A recorded long-context arc measures it: a compact record can replace a source that has been retired, is completely inert against one that can still be read, and a single blocked attention layer is enough to change which of them the model answers with. The source itself is never altered, so none of this licenses forgetting it.","url":"https://chrishayuk.com/notebook/which-source-wins"},{"id":"W-LARQL","version":"0.1","title":"LARQL","kind":"work","state":"draft","recorded_at":"2026-09-05T00:00:00Z","updated_at":"2026-09-05T00:00:00Z","material_revision":false,"topics":["model-as-database","model-representation"],"summary":"LARQL explores querying learned systems. This dossier connects the software to questions about model state, representations and what it means to make a model addressable.","url":"https://chrishayuk.com/work/larql"},{"id":"W-VINDEX3","version":"0.1","title":"VINDEX3","kind":"work","state":"draft","recorded_at":"2026-09-05T00:00:00Z","updated_at":"2026-09-05T00:00:00Z","material_revision":false,"topics":["model-as-database","model-representation"],"summary":"VINDEX3 is part of Chris Hay’s work on model representation. This dossier records the model-as-database proposition and links to the canonical project.","url":"https://chrishayuk.com/work/vindex3"},{"id":"W-HAUSE","version":"0.1","title":"HAUSE","kind":"work","state":"draft","recorded_at":"2026-09-05T00:00:00Z","updated_at":"2026-09-05T00:00:00Z","material_revision":false,"topics":["ai-interface","semantic-design"],"summary":"HAUSE is a semantic design system for AI interfaces: communicative acts rendered as statements, instruments and performances. This publication consumes its real components and publishing helpers.","url":"https://chrishayuk.com/work/hause"},{"id":"W-MCP","version":"0.1","title":"MCP-CLI","kind":"work","state":"draft","recorded_at":"2026-09-05T00:00:00Z","updated_at":"2026-09-05T00:00:00Z","material_revision":false,"topics":["mcp","ai-interface"],"summary":"MCP-CLI provides a command-line interface for working with Model Context Protocol servers. This dossier connects the interface to tool infrastructure and the wider CHUK work.","url":"https://chrishayuk.com/work/mcp-cli"},{"id":"N-OPERATOR","version":"0.1","title":"The operator and the model.","kind":"notebook","state":"draft","recorded_at":"2026-09-05T00:00:00Z","updated_at":"2026-09-05T00:00:00Z","material_revision":false,"topics":["residual-stream","recurrent-depth"],"summary":"An open thought distinguishing prediction of a model’s trajectory from reproducing its computation. It records a question; no experimental finding is asserted.","url":"https://chrishayuk.com/notebook/the-operator-and-the-model"},{"id":"Q-FFN","version":"0.1","title":"What if the FFN is a graph?","kind":"question","state":"draft","recorded_at":"2026-09-05T00:00:00Z","updated_at":"2026-09-05T00:00:00Z","material_revision":false,"topics":["ffn","model-representation"],"summary":"Could a feed-forward network be usefully treated as an addressable graph? This question connects model representation to traversal and querying, without claiming an experimental answer.","url":"https://chrishayuk.com/research/ffn-as-graph"},{"id":"N-CONTEXT","version":"0.1","title":"Context should be abundant.","kind":"notebook","state":"draft","recorded_at":"2026-08-26T00:00:00Z","updated_at":"2026-08-26T00:00:00Z","material_revision":false,"topics":["context-engineering","model-as-database"],"summary":"An open notebook proposition: repeatedly reconstructing earlier state may be one reason context feels scarce. Indexing reusable state is a question to investigate, not an established saving.","url":"https://chrishayuk.com/notebook/context-should-be-abundant"},{"id":"Q-RESIDUAL","version":"0.1","title":"Can geometry predict the working set?","kind":"question","state":"draft","recorded_at":"2026-08-26T00:00:00Z","updated_at":"2026-08-26T00:00:00Z","material_revision":false,"topics":["residual-stream","mixture-of-experts"],"summary":"Does residual-stream geometry predict which parts of a model will be active or resident together? The existing research note records co-activation as observed and useful locality prediction as untested.","url":"https://chrishayuk.com/research/residual-geometry"}]}