The Foundations of Computer Science notes are now done as a Forest with live OCaml toplevels, and Part II projects sorted.

Development tools report what type an expression has, but not why it has that type. This paper develops a theory of type slicing: a programmer selects a term, queries any part of its type information, and receives a program slice that is sufficient to reproduce the queried type. We formulate type slicing for bidirectional type systems, where synthesis slices explain the type a term synthesises and analysis slices explain the type expected by its surrounding context. The theory applies to any bidirectional system equipped with precision orders on types and terms satisfying a downwards static graduality property. We develop the metatheory over a core calculus with holes, products, sums, and explicit polymorphism, based on the Hazelnut and marked lambda calculi. We prove that every query has a minimal slice and that refining a query monotonically shrinks its minimal slices. We then show how to calculate these slices both exactly and approximately. Finally, integrating type slicing with error marking theory extends these results to arbitrary ill-typed programs, so a single mechanism explains both types and type errors in complete, incomplete, and erroneous code. The metatheory is mechanised in Agda, and a linear-time approximation of type slicing is implemented for the Hazel programming environment.

Serving LIFE v1.01 as a streamable Zarr v3 store, and three new clients in JavaScript, Python and R.

R access library for Tessera. Contribute to ucam-eo/geotessera-r development by creating an account on GitHub.

Reflections on the end of my sabbatical, with a reading list on societal change, Tessera's embeddings finish converting, Scrutineer continues to find security fixes galore, good news from Scotland.

A refreshed set of Part II and MPhil projects, split between environmental science problems that need computer science, and more conventional systems and programming languages topics.

Stdlib-only OCaml URL helpers for the free, no-key Real Dental Costs open data API: U.S. dental procedure cash prices by state and what Medicaid actually paid dentists per CDT code (HHS/CMS T-MSIS ...

Deploying Alpha-Omega's Scrutineer on a VM, driving its scans with a local GLM 5.3 on my Mac Studio, and thinking through practical workflows

Videos from the third Programming for the Planet (PROPL) workshop at PLDI 2026, and Cyrus Omar is visiting Cambridge on sabbatical for a year to drive followups!

Adds an OCaml profile so opam/dune projects get a working toolchain, dependency install, and reproducer path. Prompted by https://anil.recoil.org/notes/rumour-is-the-exploit which name-checks scrut...

TESSERA 1.0 is now fully available as Zarr with global RGB previews, a weather downscaling preprint, and Evidence TAP gets a public website with progress on the downloader and parser.

Porting the FoCS lecture notes to Forester, including transclusions and stable section URLs, and a live OCaml toplevel compiled into the browser.

New paper showing that a frozen TESSERA embedding improves probabilistic weather downscaling, improving local temperature and wind predictions at never-observed sites.

GETMobile micro-NPU article and package calculus published, TESSERA embeddings conversion using Fargate Spot, and starting to shift to Forester for teaching.

How the OxCaml overlay's guard packages keep incompatible releases out, and how to contribute to it with your own packages.

An OpenUK openness interview, prototyping the Evidence TAP site with exe.dev, Windows eio support finds a user in Forester, and a new grant for optimising ranger patrols.

Back from the Arctic into a heatwave, hacking on Eio for the TESSERA sync engine, the Conservation Evidence team demoing at Parliament, and TESSERA on stage at the RAISE Summit in Paris.

The Rewilding the Web workshop (Edinburgh, May 2026), recast as a wild early-web desktop. Drag the windows. Read the talks. Feed the counters.

OCaml Code-generation Toolbox

Spoke at CHIA's annual conference on AI for a changing world, as well as the first Cloud-Native Geospatial Forum outside the US, and started moving TESSERA's embeddings onto Source Cooperative.

The FreeBSD CI workers get slower over time. Is this a build-up on ZFS snapshots or something else? The FreeBSD `jails` setup is very effective, and a freshly built FreeBSD CI worker runs very quickly. Over time, though, when there are thousands of ZFS snapshots it starts to crawl. I’ve seen this on macOS too. Setting a high `--obuilder-prune-threshold` on FreeBSD mitigates the problem it’s not the silver bullet I had hoped for. Today when I was updating the machine to OCaml 5.5.0, base images, I noticed piles of `git-daemon` processes. I’d seen this before and had dismissed it, but today was the day to diagnose the where these came from. root@rosemary:~ # ps -ax | grep git 2261 - IsJ 0:00.00 /usr/local/libexec/git-core/git-daemon --base-path=. ... --port=9419 ... 2342 - IsJ 0:00.00 /usr/local/libexec/git-core/git-daemon --base-path=. ... --port=9419 ... ... 33 of these, the oldest six weeks old ... The `J` in the state column indicates that each one of these is running inside a jail. Cross-referencing with `jls` showed 33 of these daemons spread across 33 leftover obuilder jails, each holding a devfs mount and a ZFS snapshot open. obuilder runs each build step in a jail, created roughly like: jail -c name=... command=/usr/bin/su -l opam -c '<the build>' A non-persistent jail is automatically removed once its last process exits. That is usually fine: the build command finishes, the jail exits, and FreeBSD unmounts the devfs along with it. Very tidy. When a build step fails, the jail isn’t so tidy, and the failure path in obuilder tidies up the leftover devfs mount. However, a problem occurs when a build step leaves something running. An opam package’s test suite spins up a throwaway git server with `git daemon --detach` to test clone and push. The `--detach` flag makes it double-fork into the background, and because a vnet jail has no `init` of its own, the orphaned daemon reparents to host PID 1. Every leaked daemon confirmed it: root@rosemary:~ # ps -axww -o pid,ppid,jid,command | grep git-daemon 2261 1 44440 ...git-daemon... --port=9419... `PPID 1`. The build command exits, but this straggler is still alive inside the jail, so the jail never auto-removes. It sits there forever, holding its mount and snapshot. To make it worse, while the daemon is alive, it holds the in-jail filesystems busy, so obuilder’s tidy-up `umount` calls fail with `EBUSY`. The devfs mount also remains. My fix is to use the `jail -r` to remove jail, which sends `SIGKILL` to every process in the prison, which is what the failure path did before. Now this runs on success as well, which will typically be a harmless attempt to remove an already closed jail, but it will also tidy away anything lingering. My PR is ocurrent/obuilder#210. I can trace back which packages install the opam package `conf-git-daemon`, and found `dune`, `git-kv`, and `git-net`. Of these, `git-kv` is the only one which starts the process with `--port-9419`. Interestingly, the code does tidy up with `Fun.protect ~finally:(fun () -> kill_git pid) (...)`; however, the code never triggers because `git-daemon`’s readiness is polled with `lsof`, which isn’t installed on the worker and silently fails, leaving exactly one `git-daemon` running. I’ve opened Issue#15 on git-kv.

Sir David Attenborough drops by for the Cambridge Conservation Initiative's tenth birthday in the DAB, and I spend the afternoon demoing TESSERA and the Dash of Life to a campus packed with visitors.

Notes from Andrew Balmford's Royal Society Environment Medal lecture, on why shifting diets and cutting food waste are necessary but not sufficient, and why sustainable high-yield farming tied to land sparing is the key to slowing the extinction crisis.