nordisk/facett

Handover 2026-09-29 end — oden session closed

nornir:class=design · written on oden at the owner's stop order (oden going away). Continues handover-2026-09-29-oden.md. Every sha below was checked with git ls-remote when this file was written.

Mains

Open branches left by this session (pushed, NOT merged)

facett: every lane branch of today is merged. The open facett branches all predate today and were not touched: feat/facett-eid, feat/facett-flight, feat/facett-progress, feat/jobview-progress-bar-jera-naming, feat/sweden-geomap, feat/wrapped-webgpu, lane/standing-reds.

Services and processes on oden

• korp-server.service: INACTIVE (stopped 19:33:03 UTC on the owner's order), still enabled — it starts again at boot. Installed: current → korp-server-gpu-g10c0366-1790709813 (sha256 f2bb96c0…0935), verified healthy at 19:23:36 before the stop. Rollback: ln -sfn korp-server-g05eb4e5-1789073753 current in /home/korp/.korp/software/, restart. State backup: ~/runs/facett-0929/deploy-logs/state-backup.
• korp's own spark41 boot fails (korp-infra-2, draupnir container boot through user korp's podman). The draupnir-korp-spark-iceberg container in rickard's podman was started by hand at 19:22 and stopped at the end of this session. Spark reads S3 at host.containers.internal:9000, but :9000 listens on 127.0.0.1 only — unverified with a real query.
• korp-ui (~/.local/bin/korp-ui, sha256 6a6cc582…) window on the owner's desktop: left running (owner's).
• Owner's Chrome /tmp/owner-chrome-685t: left open (owner's). No gallery_probe runs; nothing on :8620.

Not done

• Step a (dwarf, 2026-10-03, owner: "Step a: go"): crates version 0.2.0 MERGED to main; crates.io publish ON HOLD. Lane C2's bump (2a156ce, rebased onto main 0f948c7 as c5a38fb: workspace 0.1.19 → 0.2.0, 311 intra-facett requirements in 77 manifests, facett-py 0.2.0) is on main together with the follow-up: knut main f51b296 widened knut-pipelines facett-pipeline to >=0.1.19, <0.3 (and knut-popviz facett-pop to >=0.1.9, <0.3), so facett's root [patch.crates-io] facett-pipeline = { path } applies again and facett-pipeline/tests/knut_facett_matrix.rs compiles (was E0308 at :262, two PipelineModel types) and passes 2/2; the registry facett-core/graph/pipeline 0.1.19 copies left Cargo.lock. facet_completeness (UniverseTab → caps matrix as universe; GenTab = container; SimView = BACKLOG, paid by moving StatGridView into the matrix — backlog stays 52) and roster_gate (graph tabs credit facett-graphgen) are green. Workspace CPU suite on dwarf: 3745 passed, 121 failed in 82 targets — 120 "no adapter" + 1 PoisonError fallout in facett-map3d/tests/gpu_clock.rs, nothing else. Publish order (owner): bump (done) → name scrub / squash → publish. Blockers left for crates.io: znippy-zoomies 0.1.26 and rotaryengine 0.1.13 are not published; nornir-robotui on egui 0.36 belongs to 0.3.0 (the egui-0.36 branches — facett lane-h2/egui-0.36, korp/dwarves/nornir/holger/edda egui-0.36 — are kept, not merged). PyPI 0.2.0 was already published 2026-10-02 (the lane Y bullet below predates the upload); 0.2.1 (name scrub, MIT only) is prepared on main 0f948c7. The PyPI token is to be rotated by the owner. korp: its 30 facett requirements (Cargo.toml 28, crates/korp-embed 2) are path + caret 0.1.x, and edda crates/step-board pins facett-jobview 0.1.17 (2 lines) — against a 0.2.0 facett checkout korp's workspace does not resolve at all ("failed to select a version for facett-about = ^0.1 … candidates 0.2.0"). Widening those (korp's dependency wave, not done here) resolves cleanly; the knut seam itself is closed by f51b296 (with the old knut, korp would lock registry facett-pipeline 0.1.19 beside the path 0.2.0).

• Lane M (dwarf → t14s, 2026-10-02): lineage at vsync + SwiftShader says "no GPU" — DONE, merged to main. (loki was switched off mid-lane; the GPU checks ran on the t14s Radeon 890M, headless, under ~/runs/t14s.gpu.lock.) (1) lineage under vsync: NOT the hover path (a drag neither rebuilds nor probes the hover index — hover_builds/hover_probes in the state + a_drag_neither_rebuilds_nor_probes_the_hover_index; hover_candidates: 42 is the press-frame probe). Root cause: the page serialised every tab's state every frame (facett-demo/src/main.rs demo.state_json()), and with lineage on screen that included the inactive Graphs tab's full demo-data state (102 kB, 1.2–1.9 ms/frame natively); plus GraphNav::positions() and appearance() hashed all 206 250 edges per frame for a cache key. Fix: FacetDeck::state_json_rolling / serialize_rolling (the tab on screen + one inactive tab per frame, the rest from a cache, same bytes), demo publish_rolling; structure_sig memoised per FrameIndex. t14s, 100k community drag: lineage 45.4 / 37.9 → 60.4 / 60.3 fps, Graphs 56.6 / 42.2 → 60.3 / 60.3 (A/B, same session). (2) SwiftShader: facett_core::adapter_retry::software_adapter (over is_software_rasteriser + facts_from_browser_info, moved to facett-core — one copy; browser_bench::software_adapter uses it) refuses a software adapter in wasm_start and facet-wrapped start() before any lane: "no GPU: software adapter (SwiftShader/llvmpipe) …" on the page, in __facett_state() / facet_error(); graphcloud buffers without mappedAtCreation, checked against the limits, a refusal said as gpu_error. t14s headless without the GPU flags: demo + gallery (universe/decograph/graphpan 100k) refuse plainly, 0 panics; with the flags gallery 3/3 PASS. Numbers in facett-benchmarks.md ("lineage at vsync + SwiftShader … lane M"). Main 620444e; pages 9c8aaf8 (live ?v=620444e, served wasm sha256 069f6f97… == dist, 26 949 374 B — the same bytes the t14s after-runs measured). Rollback: git push --force origin 3ef70a7d1d6eadbdfd194c3fd9a66e9544b3623a:refs/heads/pages. Still open: the RTX 5080 row was not re-measured (loki off); facett-core gpu_pick_host (4) needs an adapter (dwarf has none) — unchanged.

• Lane Y (dwarf → loki, t14s, 2026-10-02): PyPI — facett + facett-demo 0.2.0 READY TO UPLOAD, nothing uploaded. Owner decision: two packages at one version. facett (was facet-wrapped-web; import facett, alias facet_wrapped_web warns until 0.3.0; Streamlit glue moved to facett.streamlit behind [streamlit]; wheel only) and facett-demo (was py/ facett-demo-python; import facett_demo, command facett-demo, [korp] extra, offline mode without it). Recipe rows facett-wheel + facett-demo-wheel, PYPI_VERSION. New verbs: cargo ops verify-demo-install, cargo ops publish-pypi (Rust legacy-upload client, token from env or ~/.pypirc, never printed). Built at 340af30: facett wheel 12 509 079 B 9fec3f1e… (byte-identical over 3 builds), demo wheel 88 388 B 0cefda67…, demo sdist 82 555 B (content-reproducible only). Clean install from PyPI-fetched Streamlit 1.64.0: green; test-demo 59/5; gallery probe on loki on the INSTALLED assets 51/51 + 3/3 (nvidia blackwell). Streamlit on dwarf now runs the INSTALLED wheels (draupnir-facet-wrapped-demo, 127.0.0.1:8501, wheelhouse binds ~/runs/facett-0929/laneY/facett/{pkg-facet-wrapped,py}/dist ro — restart: cargo ops streamlit in that worktree, no index needed); the t14s tunnel works. The uploader binary + the three files + both SHA256SUMS are staged on the t14s in ~/runs/laneY-publish (dry-run there: all checks pass, token from ~/.pypirc [pypi]); the real upload waits for the owner's go: cd ~/runs/laneY-publish && ./facett-ops publish-pypi --repository pypi facett-0.2.0-py3-none-any.whl facett_demo-0.2.0-py3-none-any.whl facett_demo-0.2.0.tar.gz, then delete that dir. Owner TODOs in .nornir/pypi-publish.md (author e-mail, ignalina.dk TLS certificate is *.one.com, no LICENSE texts in the repo); trusted-publishing road in publish/github/ (needs the workflow committed at the repo root by the owner).

• Lane L (dwarf → loki, 2026-10-02): the final GPU pass on the merged code + the RTX 5080 rows — DONE. Owner's go for loki given in the session. Every graph tab at 100k on ONE bundle — the live pages bytes (pages 3ef70a7, main 1d73e36, ?v=4088fa9, wasm sha256 28016392…) — headless Chrome 154 on loki's RTX 5080 (nvidia blackwell), demo_pages_probe, one Chrome per scenario under ~/runs/loki.gpu.lock: drag fps Graphs 60.2/60.3, lineage 56.8/58.3, decograph 60.3, graphpan 60.4, graph3d 60.3, Universe 100k 60.4, Universe 998 250 60.4, falkor 60.3, World coderoom crates 60.3, World coderoom SBM (after the DAG fix) 60.3, graph SBM force 60.3, graph community force 60.4, graphview 60.3; resident_uploads 1 (graph: 2) across every drag; GPU 39–43 °C. No regression vs the t14s rows (all first frames/settles shorter; Graphs at 60 on the same bytes = lane G's 29 fps was the t14s power-saver profile). Gallery probe on main's four bundles (core 7078870c…): 51/51, wrong-bundle 3/3, universe 100k 61.7 fps, decograph / graphpan 100k resident_uploads 1. Streamlit page (dwarf :8501 through a reverse tunnel, opened and closed by lane L): universe/decograph/graphpan at 100k, manual, simview mount webgpu on the 5080. Table in facett-benchmarks.md ("The final GPU pass … lane L"). Still open: lineage at 100k sits just under vsync (17.1–17.6 ms rAF p50) — fixed by lane M (the per-frame all-tabs state publish, not the hover path); the t14s final pass (Radeon) was not rerun — the 5080 rows stand for the merged code; the Streamlit container still runs the lane-F wheel (core 00a9734b…), not a wheel of main; VK_LOADER_DRIVERS_SELECT='*nvidia*' breaks Chrome's Vulkan instance on loki (-7 → SwiftShader) — leave it unset on NVIDIA; on SwiftShader the demo panics in CloudModelPaintCallback::prepare (createBuffer 1.6 MB mappedAtCreation) instead of saying "no GPU" — fixed by lane M (refused at start, "no GPU").

• Lane G (dwarf, 2026-10-02): falkor, World coderoom, graph, graphview to 100k on the GPU — DONE and LIVE, main 4088fa9. GEN_TABS ceilings 1 000 → 100 000 (FIXTURE_SCALE gone; facet-wrapped's graph2d raised with it, so Streamlit matches). falkor: FalkorBrowser lays out on a non-blocking gpu_layout::LayoutJob on the host's device (grid kernel above 16 384; ≤ 1 000 with the CPU lane's own params, same look), no device → CPU up to CPU_LAYOUT_MAX and "no GPU" above; painted on the resident cloud. World coderoom: DepGraphLayout::build O(V+E) on dense ids, iterative rank (parity with the old recursive one on cyclic graphs), positions once per refresh, resident cloud, camera kept across what-breaks; generated edges point newer → older (Cargo forbids cycles). graph: GraphView scale path above 5 000 — layout once per scene, pan/zoom, resident cloud, FC-4 under a stated budget (≤ 2 000 on-screen AccessKit buttons, a11y in state_json, the canvas label says "K of N"), force-laid on the GPU at scale when a host device exists (facett-graph/gpu-layout). graphview: the same model on the resident cloud above 5 000, L0 CPU below / with no renderer. DecoratedGraphView resident edge = the CPU painter's 1.4 px. t14s (Radeon 890M, headless, drag fps): balanced — falkor 60.4, coderoom (crates) 60.4, graph (force) 59.8, decograph 60.3; power-saver — graphview 52.6, falkor 52.8, coderoom (crates) 52.6; A/B vs the live lane E bundle in the same power-saver hour: no regression (Graphs 29.2/29.7, Universe 51.6/53.0, decograph 50.9/51.5, graphpan 51.8/53.4) — the owner's power profile (power-saver, then balanced) moves every number (facett-benchmarks.md). Tests: 445 passed in 28 binaries on the rebased branch (demo lib 143, four_tabs_at_scale 4, graph_tabs_at_scale 4, navgraph_demo 5, robot_demo 2, deck_no_error_atoms 2, control_matrix 3, demo_frame 2, facet-wrapped lib 66, graphview 139+, graph scale_lane 2, graphgen, coderoom). Pages 3ef70a7, live ?v=4088fa9, served wasm sha256 28016392… == dist (26 924 313 B). Rollback: git push --force origin f0838040c268ec489078560069f88e31a4e359fb:refs/heads/pages. Still open: the final-code t14s pass (all 8 tabs on one bundle, coderoom SBM after the DAG fix, graph SBM with force, graphview balanced) waited on Tctl < 60 °C while the owner was on a video call (76–78 °C) — run on loki's RTX 5080 instead by lane L (above); loki RTX 5080 rows — done by lane L (were refused by the session's permission classifier until the owner's go); Graphs at 100k is 29 fps under power-saver on BOTH bundles (lane E's 58.5 was balanced) — its own lane, not touched here; mega_matrix screenshot and robot_deck ×2 need an adapter (dwarf has none).

• Lane F (dwarf, 2026-10-02): Streamlit parity + the GPU pixel tests — DONE, branch lane-f/streamlit-parity (merged to main, see git log). (1) facett-demo's generators, lowerings, GenTab (summary above 5 000 nodes), per-tab builders and the Universe tab moved to a library, facett-graphgen (the demo re-exports it). facet-wrapped: the generator seam {"generate": {"source","size","seed"}} / "sbm:100000:1" (built in the wasm, never shipped as JSON), Entry::with_generator + manifest generate {max,sizes,sources}, new entries universe (1M) and decograph (100k); graphnav, graph2d (1k), graph3d, graphpan, graphview, lineage take requests (lineage left the fixture-only roster). Streamlit: top tabs = the demo's groups (Grafer, Kartor, Tabeller & diagram, Bygg & jobb, App & UI), each opening on 🎨 Komponenter with the live korp pane beside it; a source/size/seed picker next to every generated component (read off the shipped manifest), ?gen= passthrough. (2) Wheel from all FOUR bundles (core/docs/model3d/sim; the sim group is now required) in the container road, holger index from ~/runs/facett-0929/laneS/holger-run with the ron at ~/runs/facett-0929/laneF/facett/target/pip-build/holger-facett-pip.ron, verify-install 51 components, test-demo 58 passed / 5 skipped. Streamlit RUNNING on dwarf 127.0.0.1:8501 (container draupnir-facet-wrapped-demo, source bind ~/runs/facett-0929/laneF/facett/py ro, installed core wasm sha256 00a9734b… == dist), health ok. Restart: start the index as above, then cargo ops streamlit in ~/runs/facett-0929/laneF/facett. The index was STOPPED after the demo came up. The reverse tunnel to the t14s (ssh -f -N -R 127.0.0.1:8501:127.0.0.1:8501 rickard@100.67.148.64) is UP. (3) GPU: gallery probe 51/51 + wrong-bundle 3/3 on the t14s (amd rdna-3) AND loki (nvidia blackwell); universe 100k 61.5 / 61.8 fps, decograph/graphpan 100k resident (resident_uploads 1), first frames in facett-benchmarks.md; Streamlit page mounts universe/decograph/graphpan/manual/simview webgpu on both. (4) Pixel tests (never run on a GPU before): robot_deck ×2 green on both GPUs; snapshot + mega_matrix were RED on main too (lanes D/E changed the deck; branch render byte-identical to main) — 8 goldens re-blessed on loki's RTX 5080, within 2 on the t14s. Found + fixed: find_cycles unbounded path walk froze state_json on a 120-node generated graph (now 200 000 steps); simview had no gallery payload; inventory.rs missed facett-ops. Still open: decograph/graphpan fps in the Streamlit iframe is not measured by the probe (reactive panes repaint only when nudged; lane E's 60 fps drag is the same code); find_cycles is still a bounded sample, a Johnson walk would make it exact; Look & feel v2.0 not designed (v1 tokens); pages not redeployed (lane G).

• Lane E (dwarf, 2026-10-01): decograph + graphpan on the RESIDENT lane, 4 more tabs on the generator — DONE and LIVE, main f85386c. Above cloud::RESIDENT_ABOVE (5 000) DecoratedGraphView and graphpan (GraphPanView, paint_scene_opts, draw_graph_opts) upload the cloud once per model (cloud::ResidentSlot; decograph also per unfold change) and send the camera per frame; overlays (selection ring, decoration rings/badges/edges, lit trace ≤ 4 096, on-screen labels ≤ 2 000) stay on the CPU. The free paint_scene_opts (dwarves' Universe) fingerprints the borrowed model — no dwarves change, API source-compatible. t14s Radeon 890M headless, 100k SBM drag: decograph 2.8 → 60.4 fps, graphpan 14.8 → 60.3 fps, resident_uploads 1 across the drag; Graphs 58.5, Universe 60.4 (no regression); max Tctl 74 °C (facett-benchmarks.md). Size + generator now also on falkor (a Cypher-shaped answer on the in-memory runner), World coderoom (a crate graph through GenCoderoomData, its data seam), graph (a Scene), graphview (the L0 kernel) — each capped at FIXTURE_SCALE = 1 000, the reason in its LEGEND row (CPU force layout / O(V·E) seam layout / per-node AccessKit buttons / CPU raster). metro (typed lines + stations) and Sverige (its own 1M → 100M control) keep theirs. Pages f083804, live ?v=f85386c, served wasm sha256 1cb72bac… == dist == the measured bytes. Rollback: git push --force origin 69838227e772acee9958b2fb4648a4baa6ea1c84:refs/heads/pages. Still open: (the wgpu kittest binaries — mega_matrix screenshot, robot_deck ×2, snapshot png — were run by lane F on the t14s and loki, goldens re-blessed); in the resident lane edges take their tail node's colour and a selection does not dim the cloud (the trace is drawn over it); the four fixture-scale tabs stop at 1k until their layouts move to the GPU; the RTX 5080 (loki) rows are still owed.

• Lane D (dwarf, 2026-10-01): demo graph tabs — DONE and LIVE, main 3f7d99d. Owner order: show the hidden graphs, group the tabs, a size selector + a data generator. (1) Two-level tab bar (FacetDeck::with_groups, the demo's TAB_GROUPS): Grafer 25 · Kartor 5 · Diagram 10 · 3D-scener 4 · Bygg & jobb 9 (wasm; 11 native) · Tabeller 10 · App & UI 16 · Övrigt 2 (9 native). Titles/ids unchanged; navgraph order [groups, tabs, *] → group row 1.2.1, tabs 1.2.2, pane 1.2.3 (snapshot re-blessed, 0 undeclared atoms). (2) Size + generator (facett-graphview::generate: community rings, SBM, Erdős–Rényi, Barabási–Albert, grid, tree, crates; + knut-popsim population) on Graphs, lineage, decograph, graphpan, graph3d (100 → 100k), Universe (→ 1M); deep link ?gen=<src>:<size>:<seed>#<tab> / $FACETT_DEMO_GEN. (3) New Universe tab (WorldGraph galaxy on the resident GPU cloud, labels off, "no GPU" said plainly). GPU lanes turned on: graphnav wgpu + gpu_cloud, graphpan wgpu, graphview gpu-cloud. Measured on the t14s (Radeon 890M, headless): Graphs 100k 57.9 fps (GPU layout + resident cloud), Universe 100k 60.4 fps, Universe 998 250 nodes 60.3 fps — table in facett-benchmarks.md. Pages: 69838227 (rollback git push --force origin 84ef0a636d64571feb138acffa6f96c7bad70bee:refs/heads/pages), live ?v=3f7d99d, served wasm sha256 a380db5e… == dist == the measured bytes. Still open: (decograph / graphpan resident and falkor / World coderoom / graph / graphview wired: done by lane E, below) metro, Sverige and the typed DAG views keep their fixtures (each LEGEND row says why). The Graphs node table is off above 5 000 nodes (exact betweenness is O(V·E) on the CPU) — a GPU or approximate betweenness would bring it back. The RTX 5080 (loki) rows and a headful check are not done (loki asleep; t14s headless only). demo_pages_probe runs one scenario per Chrome: a second iframe navigation in the same Chrome never draws (cause not found).

• Lane X (dwarf, 2026-10-01): DONE, all on main. xtask carries ONE dependency (facett-build-thing). The 8 non-build verbs are cargo ops <verb> (facett-ops). The release pipeline's publish-holger, verify-install and test-demo steps run through alias: "ops". The container road lives once, in edda (nornir_build_thing::container_road); facett's copy is deleted. nornir_pages::demo_wasm(cfg, set, variant) has no host road. Mains: edda e0a2d0b, dwarves 68c28e1, nornir fe406ec, facett — see the lane X commits after 59dd7e6. Measured: facett tests 135/135, the recipe features tree is byte-identical (3167, 044af68e…), and the panel row built through edda's road in 668468e7…. Still open (owner decision, VIKTIGT): nornir and dwarves declare no demo row (their sets are matrix-pinned), so their demo-wasm is refused by name. nornir's xtask/tests/no_silenced_tests.rs is red on main for tests/viz_jobs_headless.rs; it predates lane X and is not ours. Proving nornir and dwarves needs the holger-gaps crates (edda + Skidbladnir) and lane R's gunnar/modgunn clones beside them.

• Pages DEPLOYED (lane P, dwarf, 2026-10-01) — moved out of "Not done". Owner chose clang+llvm in the SHARED cargo-toolchain image: edda 6b5f905 (main; snapshot.debian.org 20260713T000000Z, same SOURCE_DATE_EPOCH; minted twice, second after podman rmi+prune, same manifest digest 668468e775ad…, tag on dwarf nordisk-cargo-toolchain:lanep-clang). facett re-pinned in ad81c64/086c7ba. cargo xtask deploy-pages from main 086c7ba: pages 2822cc5 → f69686f; live https://nordisk.codeberg.page/facett/ serves ?v=086c7ba, served facett-demo_bg.wasm sha256 dde4cc2ea136… == dist (26 467 787 B), facett-demo.js aee5286a… == dist. Rollback: git push --force origin 2822cc5c0a0e0384a954b24b3ae38b31a13d9a8e:refs/heads/pages. Still pinning the OLD 3d5115ba… (valid only where that image already exists — the recipe no longer mints it): knut Skidbladnir draupnir znippy dwarves skade korp-collectors nornir korp znippy-zoomies holger (+ gunnar, holger-ui per ~/git). They can re-pin to 668468e7… when they next touch it.

• Pages no-GPU fix (lane P, 2026-10-01): owner saw "loading, then black" without an adapter. Main 2776a69: wasm_start checks navigator.gpu / isSecureContext / requestAdapter() (facett-core adapter_retry, moved from facet-wrapped) before eframe and writes the plain "no GPU: …" into the page; graph-wgpu puts the 2-D graph on the GPU cloud. t14s headless: renders with GPU, message without. Live: pages 84ef0a6, ?v=2776a69, served wasm sha256 ebd69777… == dist. Previous pages f69686f.

• Lane P done: panel-wasm is the facett-wasm-demo-panel row (edda 2637751 BuildVariant::example

  • wasm::artifact_stem), built in the PINNED image on dwarf (cargo 171 s, panel_bg.wasm 5 620 426 B). Gotcha: a fresh worktree without LFS smudge carries 131-byte pointers; facett-docview then refuses to compile (DejaVuSans.ttf is a Git-LFS pointer) — run git lfs checkout before any wasm build.

• Streamlit: RUNNING on dwarf (lane S, 2026-09-29), facett main fc07b72. Container draupnir-facet-wrapped-demo (image nordisk-streamlit sha256:b84e4494…), 127.0.0.1:8501 only, source bind ~/runs/facett-0929/facett-laneS/py (ro). Wheel facet_wrapped_web-0.1.19 (sha256 2fcc0fa7…) built in the container road from all three bundle groups (core wasm 12 599 322 B 7166e4a4…, docs 9 036 092 B 3fcd3eae…, model3d 7 229 203 B 10179d1a…); verify-install: 11 assets identical, 48 components. Stop: podman rm -f draupnir-facet-wrapped-demo. Restart needs the index up first (the container pip-installs the wheel at start): cd ~/runs/facett-0929/laneS/holger-run && HOLGER_DATA=$PWD/data ./holger-server start --config ~/runs/facett-0929/facett-laneS/target/pip-build/holger-facett-pip.ron, then cargo xtask streamlit in facett-laneS. The index was STOPPED after the demo came up.

  • holger index: holger main 424945a server-local row built in the pinned image (holger xtask build --suite ORIGINAL --variant server-local, binary sha256 a53935cd…), in a private sibling tree ~/runs/facett-0929/laneS/src/ = worktrees on edda holger-gaps d71ade3 (the only edda head carrying appliance-grow + admit + disk-snapshot + estate-clock with estate_clock::poll + estate-front; release-2 lacks poll), Skidbladnir holger-gaps 52b0c07, + clones tunnr, znippy, monetize, monetize-impl, modgunn, holger-ui. holger main's Cargo.lock does NOT match any sibling set: re-resolved LOCALLY (commit 4327e77, not pushed): appliance-pid1 0.1.1→0.1.2 + additive edges (facett-core ahash/getrandom/wasm-bindgen/web-sys, wg-appliance-core tracing). No edda branch merged.
  • Mounts proven on a GPU (t14s, Radeon 890M amd rdna-3, headless flag set above, reverse ssh tunnel closed after): Streamlit page mounts core (binview, geomap, graph2d, boxplot, table, calendar), docs (?pick=manual, ?pick=docview) and model3d (?pick=model3d), each canvas_api=webgpu, software=false — read off the loader's new facet:mounted console line. gallery_probe on the INSTALLED assets: 48/48 PASS (45 core + docview + manual + model3d), wrong-bundle 2/2, every adapter amd rdna-3, attempts 1. loki unreachable (ssh timeout) — the RTX 5080 rows are still owed.

• cargo xtask test-demo: green against the wheel from the private index — 53 passed, 5 skipped (3 need a live korp-server, 2 a saga sink). cargo test -p xtask 111 passed. Still owed: holger main needs a Cargo.lock that resolves against a published edda set (today only a local re-resolution builds it); robot_streamlit.py (playwright) not run — python on the host is out.

• Skidbladnir install road refuses sparse files (korp-server --refresh → bad_entries=["korp-server"]); every container-built refresh hits it until fixed.

• edda: from_wasm must become a list (the wheel has 3 producers); WasmStageRun::ComponentManifest drops the row's features (facett works around it).

• Headful browser rows are owed on both GPUs: loki has no desktop login, the owner stopped headful on oden.

• ahash runtime seeding (crypto.getRandomValues via getrandom wasm_js) not yet exercised in a browser.

• korp debt: tests/robot_shell_mount.rs:234 (tab floor predates 17 → 8 tabs); two lib tests collide in a full run (src/jobs.rs:1183, src/map_networks.rs:2594). dwarves debt: robot_two_doors.rs:217 needs a monetize-impl checkout; dwarves-gui has 46 wasm errors in native-only modules.

Evidence on oden

~/runs/facett-0929/: gate-logs/, groups-evidence/, logs-container/, bench-logs/, deploy-logs/, wasm-proof/, tmp/clean-targets.log (2.74 TB of idle target dirs freed, /home 96 % → 54 %).