.plan-26-35: All I want for Tesseramas is (GP)U

GeoTessera 0.10 has been released, moving hosting to Source Cooperative with UTM-native Zarr routing and Matryoshka reads, while the full global run of Tessera v1.1 embeddings for 2017-2025 has been delayed by a month due to GPU shortages. A new paper by Julia Guerrero-Viu et al. finds that Tessera embeddings degrade gradually with shorter temporal windows, with a 39% drop in segmentation accuracy on PASTIS-R when contracting from a year to a month, but still outperform from-scratch models when labels are scarce.

I've been in Ireland again this week for the bank holiday, which also saw us celebrate a year of Tessera hacking https://anil.recoil.org/ a-year-of-tessera-embeddings while hunting for more GPUs to support our users' embedding requests. I also got Scrutineer scanning my own code https://anil.recoil.org/ scrutineer-scanning-with-a-local-glm using a local GLM-5.3 model, the PROPL 2026 videos are live https://anil.recoil.org/ propl-2026-videos-are-live , and I wrote a text and data mining reservations library https://anil.recoil.org/ tdm-reservations-for-evidence-tap for Evidence TAP https://evidencetap.org . It's been a whole year since the first GeoTessera release, woohoo I put together a brief recap https://anil.recoil.org/notes/geotessera-a-year-on and released GeoTessera 0.10 https://github.com/ucam-eo/geotessera/releases/tag/v0.10.0 , which moves our hosting over to the Source Cooperative https://source.coop/ with UTM-native Zarr routing and Matryoshka reads for the v2 beta. The sync of the Tessera v1.1 embeddings https://anil.recoil.org/notes/tessera-v11-out onto the Source Cooperative also finished this week, so those are now browsable via tze.geotessera.org https://tze.geotessera.org as well click on the model version on the top right . The full global run for v1.1 for 2017-2025 has unfortunately been delayed by a month due to GPU shortages. If you can help us get access to more GPUs then please get in touch. Smaller cards are fine, since we can parallelise the inference across many of them per tile. Mark Elvers https://www.tunbury.org/ has mostly finished the v2 inference ablations https://www.tunbury.org/2026/08/31/week-35-2025/ , so compute is now what we're short of. Relatedly, I read a nice paper on the temporal sensitivity of Tessera embeddings https://arxiv.org/abs/2608.27175 . The authors keep the Tessera encoder frozen and recompute its embeddings over successively shorter observation windows, from a full year which is what we generate by default right down to a single day . They then benchmark linear probes and UNet heads against from-scratch networks on various datasets. Interestingly, accuracy degrades gradually rather than collapsing. Contracting the temporal window from a year down to a month drops 39% of the segmentation accuracy on the phenology-driven PASTIS-R crop task. I was surprised to see that even single-day embeddings classify land cover at 3.4 times the background chance baseline, so temporal coverage seems practical to tune if you're willing to pay the inference cost. The benefit of using embeddings is more evident when labels are scarce: with only 1% of the labels TESSERA-UNet already achieves ∼24 mIoU on PASTIS-R, some 50% above the best from-scratch UNet, and even the linear probe matches or exceeds every fully supervised from-scratch UNet there, while on DEN the embedding-based models stay essentially flat from 1% of the labels onwards and the from-scratch UNets only catch up between 30% and 100%, so the models only draw level on DEN at saturation. The embeddings thus pay off whenever the mapping task is phenology-driven or labels are scarce, and lose their edge only when it is spectral and labels are abundant, a combination rarely met in operational LULC production. -- Temporal Sensitivity Analysis of Tessera Embeddings , Julia Guerrero-Viu et al, Aug 2026 I also wrote up some project ideas this week in preparation for the incoming MPhil students to Cambridge, covering scanning the literature for species range and habitat maps https://anil.recoil.org/ideas/literature-range-habitat-maps , the generative detection of all the ghost roads in the world https://anil.recoil.org/ideas/tessera-ghost-roads , and the interpretable downscaling of local weather predictions https://anil.recoil.org/ideas/tessera-interpretable-downscaling . I'll write more about these next week when back in Cambridge. My musings on security embargoes https://anil.recoil.org/notes/rumour-is-the-exploit were covered pretty widely online, with discussions on HN https://news.ycombinator.com/item?id=49480466 and Lobsters https://lobste.rs/s/t73wqi/just rumour bug is enough find security . I loved seeing Nick Craig-Wood participate https://news.ycombinator.com/item?id=49480777 , as he maintains rclone https://rclone.org/ which we use extensively to manage filesync in the EEG https://www.tunbury.org/2026/01/06/ceph-notes/ copying-data . The rclone project fielded around 20 security disclosures in its first decade, and now 40+ in the last month alone. He's now considering merging security fixes straight to trunk, rather than holding embargoed branches. The Lobsters thread also pointed me at the Zero Day Clock https://zerodayclock.com/ , which is a depressingly incredible graph. As a followup, I deployed Andrew Nesbitt https://nesbitt.io 's Scrutineer to scan my own OCaml code, but driven by GLM 5.3 running on my Mac Studio rather than a commercial code model. I wrote up quick notes on the deployment https://anil.recoil.org/notes/scrutineer-local-llm . The short version is it's surfacing hundreds of potential findings to triage, so there's some work ahead. Thomas Gazagnaire https://github.com/samoht is giving me a hand though To support this, I refreshed my ocaml-deepseek https://tangled.org/anil.recoil.org/ocaml-deepseek inference engine, which I first built for language-integrated LLMs https://anil.recoil.org/notes/language-integrated-llms but am now exposing as a server. It is now renamed ds4 to match the upstream project, and the glm-5.3 branch serves GLM 5.3 Flash over an OpenAI-compatible endpoint for Scrutineer to use. The recordings from the third Programming for the Planet https://pldi26.sigplan.org/home/propl-2026 workshop at PLDI 2026 are now up on the ACM SIGPLAN YouTube playlist https://www.youtube.com/playlist?list=PLXeET4GjfkHs , and of course also on our self-hosted Watch EEG https://watch.eeg.cl.cam.ac.uk . I wrote up the talks and thoughts https://anil.recoil.org/notes/pldi26-propl , and am very excited that Cyrus Omar https://web.eecs.umich.edu/~comar/ is visiting Cambridge on sabbatical for the year The Evidence TAP database is up to ~20 million fulltexts now, thanks to an import from Semantic Scholar as well as the download engine continuing to get education and conservation fulltexts. I also published ocaml-tdmrep https://tangled.org/anil.recoil.org/ocaml-tdmrep , an OCaml implementation of the W3C TDM Reservation Protocol https://www.w3.org/community/reports/tdmrep/CG-FINAL-tdmrep-20240202/ , to help ensure the literature downloader respects publishers' machine-readable text-and-data-mining reservations while crawling open access versions.