go.mod declared this module under gitea.d-ma.be, the pre-rename host
(infra ADR-0004). That name still resolves in DNS but no longer serves --
HTTPS to it fails outright -- so the declared path pointed at nothing.
It built anyway, because a Go module does not have to be fetchable from its
own declared path to compile locally. That is exactly what made it durable
rot: nothing failed, so nothing flagged it. What it did break is anyone
trying to depend on this repo -- `go get git.d-ma.be/mathias/<repo>` fails on
the path mismatch, and an importer would have to spell the dead host.
Verified safe before changing: nothing in the estate requires this module
(checked every go.mod across the Go repos) -- these are leaf binaries, so the
rename breaks no importer.
Mechanical: module path plus every internal import rewritten, templ
regenerated where applicable. Verified after -- go build, go vet, go test and
golangci-lint (against a Go 1.26.1 toolchain, matching CI) all pass, and no
gitea.d-ma.be reference remains in any .go/.templ/.mod file.
Found by an estate-wide scan for the old host, prompted by the same rot
turning up in tapir internal/web/views.templ (tapir@4bafd5f).
- train.py build(): year-based split (train≤2021, OOS≥2022) replaces
misleading 70/30 mixed-period split; true OOS val_vol_r2 now ~-0.36
vs previously reported +0.18 (artefact of cross-period data leakage)
- train.py: EXPORT_EMBEDDINGS block now exports both train+OOS embeddings
with dates and HV labels for Go eval harness
- cmd/eval: LinearProbeTrainTest uses train stats for standardisation of
both sets (no leakage); standardiseCompute/applyStandardise helpers
- internal/eval: add LinearProbeTrainTest (fit-on-train, eval-on-OOS)
alongside LinearProbe (same-set); 8/8 tests still green
Phase-0 gate result: val_vol_r2=-0.36, silhouette=0.043, erank=58.9/64.
Backbone produces high-rank embeddings (SIGReg working) but does NOT
generalize across 2021→2022 regime boundary. Gate: INCONCLUSIVE/FAIL.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>