Artifacts & provenance
What is published, and what is not
Public evaluation artifacts carry metrics, provenance and digests with prompt bodies and raw completions stripped. Private artifacts hold the row-level material and are written at mode 0600; neither their contents nor their digests appear here. Every row below states which side of that line it falls on.
- Published: dataset and manifest digests, adapter digests, public evaluation digests, and the compute receipt that ties them together.
- Withheld: prompt bodies, raw model completions, session content, user identifiers, credentials, and the digests of mode-0600 artifacts. Those digests are retained in the private receipt where the audit chain needs them.
- A digest shown as not published in this record means the artifact is public but its digest was not part of the source record this site was built from — not that it is secret.
Round-1 shadow eval
| Artifact | Path | Visibility | SHA-256 | Note |
|---|---|---|---|---|
| Public site record | data/projects/router-school.json | public | not published in this record | no private prompt text |
| Row-level eval output | tooling/artifacts/router-school-eval.json | private | withheld — private artifact | mode 0600 — body withheld |
| Independent audit transcripts | Grok 4.5 (337 lines) · Kimi K3 (257 lines) | private | withheld — private artifact | dispatch-local, bodies withheld |
Foundation rebuild and bounded GPU matrix
| Artifact | Path | Visibility | SHA-256 | Note |
|---|---|---|---|---|
| V2 baselines | tooling/artifacts/router-school-v2-baselines.json | public | not published in this record | majority + task-class rule |
| V2 TF-IDF logreg baseline | tooling/artifacts/router-school-v2-tfidf-logreg.json | public | not published in this record | balanced word 1–2gram |
| V2 fixtures + manifest | tooling/fixtures/v2/{train,val,test,manifest}.json | public | not published in this record | deterministic rebuild verified by sha256sum -c |
| Preregistration | tooling/ROUTER_SCHOOL_V2_PREREGISTRATION.md | public | not published in this record | P1–P9 gates written before training |
Current-policy distillation
| Artifact | Path | Visibility | SHA-256 | Note |
|---|---|---|---|---|
| V3 train SFT manifest | tooling/fixtures/v3-policy-distill/ | public | b6de0235529d23f47a42d0dcbc6bc368b373b68159e95a0c314738121086c864 | manifest SHA-256 of the train SFT split |
| V3 policy-distill eval (invalidated) | tooling/artifacts/router-school-v3-policy-r16-eval.json | public | not published in this record | retained as the record of the harness defect |
Fairness specialist (3B)
| Artifact | Path | Visibility | SHA-256 | Note |
|---|---|---|---|---|
| V4 dataset manifest | tooling/fixtures/v4-fairness-specialist/manifest.json | public | e082229bab312618f4d4e4c1590dcee3951b03d7734bc45d1c2af8d0e9dab8d7 | schema owlforge.router-school.fairness-specialist.v4 |
| V4 train SFT split | tooling/fixtures/v4-fairness-specialist/train-sft.json | public | 5abcd931c25d8a56ca73cc6cfba4f8f167f8557f8e0fe5436b2fb0e59d86201a | |
| V4 train split | tooling/fixtures/v4-fairness-specialist/train.json | public | d10b425b4bbd706ecf3d8f722ebe4c60c82744987424de42342bb03587107cac | |
| V4 test split | tooling/fixtures/v4-fairness-specialist/test.json | public | 1e9b03dca19423b08ed1396b3b3ae3999813a2b3869edcc48946a1fb9534f037 | 105 rows — the frozen window every later cell is scored on |
| V4 adapter | artifacts/checkpoints/router-v4-fairness-r16-lora | public | b44784f8f0283fcb2169b418de7d10dc186dc80277f663e6bd81c9d466e89837 | adapter SHA pinned in both the train and eval wrappers |
| V4 baselines | tooling/artifacts/router-school-v4-baselines.json | public | not published in this record | written before the GPU cell shipped |
1B distillation
| Artifact | Path | Visibility | SHA-256 | Note |
|---|---|---|---|---|
| V4-1B adapter | artifacts/checkpoints/router-v4-fairness-1b-r16-lora | public | 73fb4805a9ea32aa210bf60074f719317d8f8bfed13517d985ca7cb34e3887f3 | |
| V4-1B public eval | tooling/artifacts/router-school-v4-fairness-1b-r16-eval.json | public | not published in this record | |
| V4-1B private eval | tooling/artifacts/router-school-v4-fairness-1b-r16-eval-private.json | private | withheld — private artifact | raw generations — body withheld |
Tiny numeric MLP
| Artifact | Path | Visibility | SHA-256 | Note |
|---|---|---|---|---|
| V5 public evaluation | tooling/artifacts/router-school-v5-fairness-mlp-eval.json | public | d4f0b8f52d000f8bbb8432a962db3e62cf238cc94e50dc1de94cad6e0f74590d | mode 0664 — public by design |
| V5 private evaluation | tooling/artifacts/router-school-v5-fairness-mlp-eval-private.json | private | withheld — private artifact | mode 0600 — row-level bodies withheld; digest held in the private receipt |
| V5 model | tooling/artifacts/models/router-v5-fairness-mlp.json | private | withheld — private artifact | mode 0600 — digest held in the private receipt |
| V5 OOD window | tooling/artifacts/router-school-v5-ood-window.json | public | not published in this record | a private companion window is held at mode 0600 |
| V5 compute receipt | gpu-lab/receipts/20260804T144750Z-router-school-v5-fairness-mlp.json | public | not published in this record | provenance.artifacts carries the matching digest for every artifact above |
| Experiment-tracker run | mlflow run 7c1d3fdd4470470eba842f7bab883ec2 | public | not published in this record | status success |
This site is a literal port
The visual shell is not inspired by the Clockwork Owl theme; it is the theme. The stylesheets and brand assets below are vendored byte-identical from a pinned commit and re-verified on every test run.
| Vendored file | Canonical source | SHA-256 |
|---|---|---|
css/main.css |
themes/hermanity/assets/css/main.css |
96640ce1e47568ec17a580e139d24bc341be558a258e29a9a2102735be61a3b8 |
css/page-banners.css |
themes/hermanity/assets/css/page-banners.css |
64f61c53cf351bc3a752995d7f6f177a00cccd45228ce981a7d199797f42dd7f |
css/research-bands.css |
themes/hermanity/assets/css/research-bands.css |
50ee01648c68a9e4c66ffcda32b6d130ef7c7b410227bf85c167581379d4acfe |
Source project herman/hermanity-theme-template at commit 9b891c1098f7277bd05e3e75d1e61d78b7ee92aa. The full manifest — including the six favicon and brand assets and the six Clockwork Owl hero renditions — is in THEME_PROVENANCE.json, and the test suite diffs every vendored byte against that commit whenever the canonical checkout is reachable.
The only site-authored stylesheet is css/router-school.css. It defines the programme crest, measurement cells, gate marks, ladder rungs and the progressive ledger filter — and nothing else. A test rejects any palette, font face, token redefinition or shell selector appearing in it.
The site's own source of truth
Everything on this site that is a number comes out of two JSON documents, and the pages are projections of them rather than independent prose. They are served as-is:
| Document | Contains |
|---|---|
data/ledger.json |
The programme record, the current conclusion, the classification ladder, every stage cell with its measurements, gates and artifacts, and the open questions. |
data/methodology.json |
The falsification gate set, the three-baseline requirement, the audit pattern and the standing discipline rules. |
Adding a V6 stage is appending one object to cells and re-running the renderer. The rendered markup is committed, so the ledger reads identically with scripting disabled; scripting only adds the verdict filter at the top of the ledger page.
Where these numbers came from
Router School began as one station of an overnight research campaign and outgrew a single detail page. The campaign index remains the origin record and continues to carry the other stations; this site is the programme's own home.
- nightwatch.hermanity.dev — the origin campaign index.
- hermanity.dev — the family this site belongs to.
Stage records were synthesized from the campaign's own reference documents for the V1 falsification pattern, the V2 foundation and TDD recipe, the V3 policy distillation, the V4 fairness specialist, the V4-1B distillation result, the V5 numeric MLP, and the shared adapter shadow-eval recipe. Two adjacent reference documents — the V2 GPU matrix execution notes and the V3 prompt-template mismatch analysis — supplied the V2 cell results and the V3 harness defect respectively.