xai-studies: monorepo mirror + catalog (sync)

CHECKLIST.md

@@ -1,36 +1,8 @@
-# Ready-to-study checklist
+# checklist
 
-Use once after clone / pull.
+- [ ] START_HERE · index.html · model packages · dissertation  
+- [ ] `make test` → ✓ test passed  
+- [ ] airplane: Inkling SPEC + modular · Kimi facts + blog  
+- [ ] optional: HF portal · dataset  
 
-## Content
-
-- [ ] `START_HERE.md` · `README.md` · `index.html` · `LICENSE`
-- [ ] Inkling: `book/00-map.md` · `derived/SPEC.md` · `reference/hub/config.json` · `modular_inkling.py`
-- [ ] Kimi K3: `book/00-map.md` · `derived/SPEC.md` · `reference/facts.json` · blog txt
-- [ ] Compare · glossary · `dissertation/main.ipynb`
-
-## Offline tooling
-
-```bash
-cd ~/lyff/xai && make test   # or: python3 studies/run.py test
-```
-
-Expect: `validate OK` ×2 · `package check OK` · `✓ test passed`.
-
-## Airplane
-
-- [ ] Wi‑Fi off · `open index.html` works  
-- [ ] Inkling SPEC + modular sources open  
-- [ ] Kimi SPEC + facts + blog open  
-- [ ] Dissertation notebook opens  
-
-No pip / HF login / API keys needed to **read**.
-
-## Public mirrors (optional)
-
-- [ ] https://lyffseba-xai.static.hf.space  
-- [ ] https://huggingface.co/datasets/lyffseba/xai-studies  
-
-## Out of scope (on purpose)
-
-Product apps (`bet`, `iberian-rugs`, …) · optional Mojo MSc port · old deleted study clones.
+Out of scope: product apps.

CITATION.cff

@@ -1,18 +1,15 @@
 cff-version: 1.2.0
-message: "If you use these study notes, please cite as below."
-title: "xai — offline research monorepo (Inkling, Kimi K3, MSc)"
+message: "If you use these notes, cite as below."
+title: "xai (explainable AI) — offline research notes (Inkling, Kimi K3, MSc)"
 authors:
   - family-names: "lyffseba"
     given-names: "Sebæ"
 version: "2026-07-17"
 date-released: "2026-07-17"
 url: "https://github.com/lyffseba/xai"
-repository-code: "https://github.com/lyffseba/xai"
 license: Apache-2.0
 keywords:
+  - explainable-ai
   - mixture-of-experts
-  - transformers
   - inkling
   - kimi-k3
-  - research-notes
-  - offline

GITHUB_README.md

@@ -1,107 +1,69 @@
 # xai
 
-Personal research monorepo for frontier open-model studies and an MSc dissertation.  
-**Offline-complete.** Python stdlib tooling. No pip. No network required to read or validate.
+**xai** = **explainable AI**.
+
+Personal offline monorepo: study notes on frontier open models + an MSc on interpretable vision.  
+Source-driven. Stdlib validation. No pip. No network required to read or test.
 
 | | |
 |---|---|
-| **Start** | [START_HERE.md](START_HERE.md) · local [index.html](index.html) |
-| **GitHub** | [lyffseba/xai](https://github.com/lyffseba/xai) *(canonical, private)* |
-| **HF portal** | [spaces/lyffseba/xai](https://huggingface.co/spaces/lyffseba/xai) · [static host](https://lyffseba-xai.static.hf.space) |
-| **HF dataset** | [datasets/lyffseba/xai-studies](https://huggingface.co/datasets/lyffseba/xai-studies) |
-
----
+| start | [START_HERE.md](START_HERE.md) · [index.html](index.html) |
+| GitHub | [lyffseba/xai](https://github.com/lyffseba/xai) (canonical) |
+| portal | [spaces/lyffseba/xai](https://huggingface.co/spaces/lyffseba/xai) |
+| dataset | [datasets/lyffseba/xai-studies](https://huggingface.co/datasets/lyffseba/xai-studies) |
 
-## Models
+## what is here
 
-| ID | Package | Status | Total | Active | Context | Experts | Notes |
-|---|---|---|---|---|---|---|---|
-| **inkling** | [studies/models/inkling](studies/models/inkling/) | weights public | 975B | 41B | 1M | 6/256+2 shared | relative PE · SWA/global · image+audio · [Hub](https://huggingface.co/thinkingmachines/Inkling) |
-| **kimi-k3** | [studies/models/kimi-k3](studies/models/kimi-k3/) | API live · weights pending | 2.8T | undisclosed | 1M | 16/896 | KDA + AttnRes · Stable LatentMoE · [blog](https://www.kimi.com/blog/kimi-k3) |
+| id | package | status | total | active | ctx | experts |
+|---|---|---|---|---|---|---|
+| inkling | [studies/models/inkling](studies/models/inkling/) | weights public | 975B | 41B | 1M | 6/256+2 |
+| kimi-k3 | [studies/models/kimi-k3](studies/models/kimi-k3/) | API live · weights pending | 2.8T | undisclosed | 1M | 16/896 |
 
-Cross-read: [Inkling vs Kimi K3](studies/shared/comparisons/inkling-vs-kimi-k3.md) · live table: [studies/derived/DASHBOARD.md](studies/derived/DASHBOARD.md)
+[compare](studies/shared/comparisons/inkling-vs-kimi-k3.md) · [dashboard](studies/derived/DASHBOARD.md) · [glossary](studies/shared/GLOSSARY.md)
 
----
+**MSc** — [dissertation/](dissertation/) vinyl-cover genre (CNN + Concept Whitening) · [main.ipynb](dissertation/main.ipynb)
 
-## Dissertation
-
-[dissertation/](dissertation/) — image-based music genre classification from vinyl covers (CNN + Concept Whitening).  
-Notebook: [main.ipynb](dissertation/main.ipynb) · public archive: [msc-dissertation](https://github.com/lyffseba/msc-dissertation)
-
----
-
-## Repository layout
+## quick start
 
 ```
-xai/
-├── START_HERE.md          # offline study launchpad
-├── index.html             # local face (file://)
-├── studies/
-│   ├── models/inkling/    # book · code-study · derived · reference (Hub + transformers)
-│   ├── models/kimi-k3/    # book · code-study · derived · facts.json · blog/docs
-│   ├── shared/            # glossary · comparison
-│   ├── derived/           # monorepo dashboard + STATUS.json
-│   └── run.py             # offline validate (stdlib)
-├── dissertation/          # MSc notebook
-└── hf/                    # HF Space + Dataset package + publish scripts
+make test
+# or
+python3 studies/run.py test
 ```
 
-Each model package: `book/` → narrative · `code-study/` → implementation · `reference/` → primary sources · `derived/` → generated (do not hand-edit).
-
----
+Expect `✓ test passed`. Then open [START_HERE.md](START_HERE.md).
 
-## Commands
+## layout
 
-```bash
-make test                       # offline validate both models
-python3 studies/run.py test     # same, no make
-make check                      # re-derive from local sources
-make refresh MODEL=inkling      # online only — re-fetch Hub/web
-make refresh MODEL=kimi-k3
+```
+studies/models/<id>/{book,code-study,reference,derived,scripts}
+studies/shared/     glossary · comparison
+studies/run.py      offline validate (stdlib)
+dissertation/       MSc notebook
+hf/                 Space + Dataset package
 ```
 
-| Mode | What |
-|---|---|
-| **Offline** | `make test` / `make check` · read all markdown + vendored `reference/` |
-| **Online** | `make refresh MODEL=…` only |
-
-Details: [OFFLINE.md](OFFLINE.md) · curriculum: [studies/STUDY.md](studies/STUDY.md) · glossary: [studies/shared/GLOSSARY.md](studies/shared/GLOSSARY.md)
-
----
-
-## Principles
-
-1. **Source-driven** — numbers from Hub config, official blogs/docs, or papers; never invented.  
-2. **Derived is generated** — `make derive` / `make check` rebuilds SPECs; hand-edit books only.  
-3. **Offline first** — airplane-safe study tree; network is optional refresh.  
-4. **Honest gaps** — K3 active params, layer tables, etc. tracked in `open_questions.md`.
-
----
-
-## Sync mirrors (GitHub + Hugging Face)
+- `reference/` — primary sources (Hub, blogs, papers)  
+- `derived/` — generated SPECs (do not hand-edit)  
+- `book/` / `code-study/` — narrative + walkthrough  
 
-Canonical tree is this repo. One command updates **all** public/private mirrors:
+## commands
 
-```bash
-make sync
-# or: bash hf/sync.sh
+```
+make test                       # offline validate
+make check                      # re-derive from local sources
+make refresh MODEL=inkling      # online re-fetch
+make sync                       # push GitHub + HF
 ```
 
-| Target | URL |
-|---|---|
-| GitHub | https://github.com/lyffseba/xai |
-| HF Space | https://huggingface.co/spaces/lyffseba/xai |
-| Static | https://lyffseba-xai.static.hf.space |
-| Dataset | https://huggingface.co/datasets/lyffseba/xai-studies |
-
-See [hf/README.md](hf/README.md) · [PUSH.md](PUSH.md)
+## what this is not
 
----
+Not a training framework, not a weight host, not affiliated with Thinking Machines or Moonshot.  
+Secondary blogs are never sole authority for architecture numbers.
 
-## License
+## license
 
-Study notes and tooling: [Apache-2.0](LICENSE).  
-Vendored upstream snapshots keep their original licenses — see [NOTICE](NOTICE).
+Study notes: [Apache-2.0](LICENSE). Upstream snapshots keep original licenses — [NOTICE](NOTICE).
 
 ---
 

NOTICE

@@ -1,24 +1,20 @@
-xai — personal research notes (offline monorepo)
-================================================
+xai — personal research notes
+=============================
+
+xai = explainable AI
 
 Copyright 2026 Sebæ / lyffseba
-Study notes & tooling: Apache-2.0 (see LICENSE)
+Study notes & tooling: Apache-2.0 (LICENSE)
 
-Layout
-  studies/       model handbooks + vendored primary sources
-  dissertation/  MSc notebook
-  hf/            Hugging Face Space + Dataset package
+studies/       model handbooks + vendored primary sources
+dissertation/  MSc notebook
+hf/            static Space + dataset package
 
-Commands
-  make test / python3 studies/run.py test  — offline, stdlib only
-  make refresh MODEL=<id>                  — optional, needs network
+make test · python3 studies/run.py test  — offline, stdlib
+make sync                                — GitHub + HF
 
-Upstream
-  Vendored Hub configs, transformers sources, blog/docs extracts, and papers
-  retain their original licenses and copyrights. Do not relicense upstream
-  material. Prefer official URLs when online.
+Vendored Hub configs, transformers, blogs/docs, papers keep original licenses.
 
-Mirrors
-  GitHub:  https://github.com/lyffseba/xai
-  Space:   https://huggingface.co/spaces/lyffseba/xai
-  Dataset: https://huggingface.co/datasets/lyffseba/xai-studies
+GitHub:  https://github.com/lyffseba/xai
+Space:   https://huggingface.co/spaces/lyffseba/xai
+Dataset: https://huggingface.co/datasets/lyffseba/xai-studies

OFFLINE.md

@@ -1,43 +1,33 @@
 # Offline
 
-Airplane mode. No network. No pip.
+Airplane. No network. No pip.
 
-## Read
+## read
 
-1. Open `index.html` in any browser (`file://…/xai/index.html`)
-2. Or open markdown under `studies/` and `dissertation/`
+`index.html` or markdown under `studies/` · `dissertation/`.
 
-## Verify tooling
+## verify
 
-```bash
-cd ~/lyff/xai
+```
 make test
-# or without make:
 python3 studies/run.py test
 ```
 
-Both re-validate local packages and rebuild the dashboard from vendored sources.
-They must not touch the network.
-
-## What is local
+## local
 
-| Path | Contents |
+| path | contents |
 |---|---|
-| `studies/models/inkling/reference/` | Hub config, card, chat template, full transformers sources |
-| `studies/models/kimi-k3/reference/` | facts.json, blog text, API docs, K2.6 config, KDA abstract |
-| `studies/*/derived/` | generated specs (rebuild offline with `make check`) |
-| `dissertation/main.ipynb` | MSc notebook |
+| `studies/models/inkling/reference/` | Hub config + transformers |
+| `studies/models/kimi-k3/reference/` | facts.json + blog/docs |
+| `studies/*/derived/` | generated SPECs |
+| `dissertation/main.ipynb` | MSc (TF/Keras to run) |
 
-## Online only
+## online
 
-```bash
-make refresh MODEL=inkling   # re-pull Hub / web
+```
+make refresh MODEL=inkling
 make refresh MODEL=kimi-k3
+make sync
 ```
 
-Skip these on a plane.
-
-## Requirements
-
-- Python ≥ 3.10 (stdlib: `json`, `pathlib`, `re`, `subprocess`)
-- optional: `make`
+Python ≥ 3.10 stdlib. Optional: `make`.

README.md

@@ -5,14 +5,11 @@ language:
   - en
 tags:
   - research
-  - study-notes
+  - explainable-ai
   - moe
   - inkling
   - kimi-k3
-  - transformers
   - offline
-  - architecture
-  - mixture-of-experts
 task_categories:
   - other
 size_categories:
@@ -20,136 +17,62 @@ size_categories:
 configs:
   - config_name: catalog
     data_files: catalog/models.csv
-dataset_info:
-  features:
-    - name: id
-      dtype: string
-    - name: display_name
-      dtype: string
-    - name: org
-      dtype: string
-    - name: status
-      dtype: string
-    - name: total_params
-      dtype: string
-    - name: active_params
-      dtype: string
-    - name: context_tokens
-      dtype: int64
-    - name: experts
-      dtype: string
-    - name: hub
-      dtype: string
-    - name: api_model_id
-      dtype: string
-    - name: path
-      dtype: string
-    - name: notes
-      dtype: string
-  splits:
-    - name: catalog
-      num_examples: 2
 ---
 
-# Dataset Card: xai-studies
+# xai-studies
 
-**Offline study packages** mirrored from the private research monorepo [lyffseba/xai](https://github.com/lyffseba/xai): Inkling, Kimi K3, and an MSc dissertation notebook.
+**xai** = **explainable AI**. Offline mirror of [lyffseba/xai](https://github.com/lyffseba/xai).
 
-| Resource | URL |
+| | |
 |---|---|
-| **Canonical (private)** | [github.com/lyffseba/xai](https://github.com/lyffseba/xai) |
-| **Browse portal** | [spaces/lyffseba/xai](https://huggingface.co/spaces/lyffseba/xai) |
-| **This dataset** | full `studies/` + `dissertation/` + offline tooling |
+| GitHub | [lyffseba/xai](https://github.com/lyffseba/xai) |
+| portal | [spaces/lyffseba/xai](https://huggingface.co/spaces/lyffseba/xai) |
 
-## Dataset Description
+## use
 
-### Summary
-
-This is a **documentation and source-snapshot dataset**, not a training corpus. It packages:
-
-- Narrative handbooks (`book/`)
-- Implementation walkthroughs (`code-study/`)
-- Generated specs (`derived/`)
-- Vendored primary sources (`reference/`) — Hub configs, transformers modules, official docs/blog extracts
-- Offline validation scripts (`studies/run.py`, package validators)
-
-### Supported tasks
-
-- Architecture study / literature-style notes  
-- Offline reproducibility of claims against vendored sources  
-- Curriculum for comparing MoE / long-context design choices  
-
-### Languages
-
-English (study notes).
+```
+python3 studies/run.py test
+```
 
-## Dataset Structure
+No pip. No network.
 
-### Catalog split (viewer)
+## catalog
 
-[`catalog/models.csv`](catalog/models.csv) — one row per model package (also [`catalog/registry.json`](catalog/registry.json)).
+[`catalog/models.csv`](catalog/models.csv)
 
-| id | display_name | status | total_params | active_params | context_tokens | experts |
+| id | name | status | total | active | ctx | experts |
 |---|---|---|---|---|---|---|
 | inkling | Inkling | weights_public | 975B | 41B | 1048576 | 6/256+2 shared |
-| kimi-k3 | Kimi K3 | api_live_weights_pending | 2.8T | *(empty)* | 1048576 | 16/896 |
+| kimi-k3 | Kimi K3 | api_live_weights_pending | 2.8T | | 1048576 | 16/896 |
 
-### Full tree
+## contents
 
 ```
-catalog/                 machine-readable index
-studies/
-  models/inkling/        book, code-study, derived, Hub + transformers sources
-  models/kimi-k3/        book, code-study, derived, facts.json, blog/docs, papers
-  shared/                glossary, comparison
-  derived/               monorepo dashboard + STATUS.json
-  run.py                 offline validate (stdlib only)
-dissertation/            MSc notebook + README
-START_HERE.md            launchpad
+catalog/
+studies/models/inkling|kimi-k3/   book · code-study · reference · derived
+studies/shared/
+studies/run.py
+dissertation/
+START_HERE.md
 ```
 
-## How to use
-
-```bash
-# after downloading this dataset
-python3 studies/run.py test
-# stdlib only — no pip, no network
-```
-
-Then read `START_HERE.md` or open the [Space portal](https://huggingface.co/spaces/lyffseba/xai).
-
-### Facts (do not invent beyond sources)
-
-- **Inkling:** ~975B / 41B active · 6/256 + 2 shared · relative PE · hybrid SWA/global · image+audio · 1M ctx  
-- **Kimi K3:** 2.8T · Stable LatentMoE 16/896 · KDA + AttnRes · vision/video · 1M ctx · active params undisclosed · weights ETA 2026-07-27  
-
-## Source data
-
-| Package | Primary sources |
-|---|---|
-| Inkling | [thinkingmachines/Inkling](https://huggingface.co/thinkingmachines/Inkling) Hub config + transformers `models/inkling` |
-| Kimi K3 | [kimi.com/blog/kimi-k3](https://www.kimi.com/blog/kimi-k3), platform.kimi.ai docs, [arXiv:2510.26692](https://arxiv.org/abs/2510.26692), K2.6 Hub config |
-
-## Limitations
+## facts
 
-- Not a training or eval set of model outputs.  
-- Kimi K3 structural detail is incomplete until weights / tech report.  
-- Vendored upstream text may be incomplete extracts; prefer official URLs when online.  
-- Study notes are personal research — not affiliated with Thinking Machines or Moonshot.
+- Inkling: ~975B / 41B · 6/256+2 · relative PE · SWA/global · image+audio · 1M  
+- Kimi K3: 2.8T · 16/896 · KDA+AttnRes · 1M · active undisclosed · weights ETA 2026-07-27  
 
-## License
+## limitations
 
-- **Study notes & tooling in this package:** Apache-2.0  
-- **Vendored upstream snapshots:** retain original licenses (see `NOTICE` and upstream headers)
+Not a training corpus. Not affiliated with Thinking Machines or Moonshot.  
+K3 structure incomplete until weights/tech report. Upstream extracts keep original licenses.
 
-## Citation
+## citation
 
 ```bibtex
 @misc{lyffseba_xai_studies,
-  title        = {xai-studies: Offline research notes on Inkling and Kimi K3},
-  author       = {lyffseba},
-  year         = {2026},
-  howpublished = {\url{https://huggingface.co/datasets/lyffseba/xai-studies}},
-  note         = {Companion Space: https://huggingface.co/spaces/lyffseba/xai}
+  title  = {xai-studies: offline notes on Inkling and Kimi K3},
+  author = {lyffseba},
+  year   = {2026},
+  url    = {https://huggingface.co/datasets/lyffseba/xai-studies}
 }
 ```

START_HERE.md

@@ -1,53 +1,50 @@
 # START HERE
 
-Airplane mode OK. No network. No pip. Python ≥ 3.10 stdlib only.
+**xai** = **explainable AI**. Offline. No pip. Python ≥ 3.10 stdlib.
 
-## 60-second orientation
+## open
 
-| | |
-|---|---|
-| **Local face** | open [`index.html`](index.html) |
-| **This path** | markdown under `studies/` + `dissertation/` |
-| **Public portal** | [HF Space](https://huggingface.co/spaces/lyffseba/xai) · [static](https://lyffseba-xai.static.hf.space) |
-| **Full mirror** | [HF Dataset](https://huggingface.co/datasets/lyffseba/xai-studies) |
-| **Verify once** | `make test` or `python3 studies/run.py test` → expect `✓ test passed` |
-
----
+1. [index.html](index.html)  
+2. markdown under `studies/` · `dissertation/`  
+3. public portal: [spaces/lyffseba/xai](https://huggingface.co/spaces/lyffseba/xai)
 
-## What to study
+## prove it works
 
-| Track | Path | Why first |
-|---|---|---|
-| **Inkling** | [`studies/models/inkling/`](studies/models/inkling/) | Public weights + vendored transformers sources |
-| **Kimi K3** | [`studies/models/kimi-k3/`](studies/models/kimi-k3/) | Live API + KDA theory; weights pending (~July 27) |
-| **Compare** | [`studies/shared/comparisons/inkling-vs-kimi-k3.md`](studies/shared/comparisons/inkling-vs-kimi-k3.md) | Long-ctx / MoE / product bets |
-| **MSc** | [`dissertation/main.ipynb`](dissertation/main.ipynb) | Vinyl-cover genre CNN + Concept Whitening |
+```
+make test
+# or
+python3 studies/run.py test
+```
 
----
+→ `✓ test passed` (no network).
 
-## Suggested order (~1–2 days)
+## study
 
-1. **Map** — [STUDY.md](studies/STUDY.md) · [DASHBOARD.md](studies/derived/DASHBOARD.md) · [GLOSSARY.md](studies/shared/GLOSSARY.md)
-2. **Inkling deep** — book `00` → `derived/SPEC.md` → arch `01` → code-study with `reference/transformers-inkling/modular_inkling.py`
-3. **Kimi K3 deep** — book `00` → `derived/SPEC.md` + `lineage.md` → KDA notes + API protocol → limitations / open questions
-4. **Compare** — re-read comparison; answer the three questions at the bottom of [STUDY.md](studies/STUDY.md)
-5. **Dissertation** — [dissertation/README.md](dissertation/README.md) then run the notebook when you have data
+| | path | why |
+|---|---|---|
+| Inkling | [studies/models/inkling/](studies/models/inkling/) | public weights + vendored transformers |
+| Kimi K3 | [studies/models/kimi-k3/](studies/models/kimi-k3/) | live API + KDA; weights ~2026-07-27 |
+| compare | [inkling-vs-kimi-k3](studies/shared/comparisons/inkling-vs-kimi-k3.md) | long-ctx / MoE bets |
+| MSc | [dissertation/main.ipynb](dissertation/main.ipynb) | Concept Whitening |
 
----
+## order (~1–2 days)
 
-## Facts (do not invent beyond these)
+1. map — [STUDY.md](studies/STUDY.md) · [DASHBOARD](studies/derived/DASHBOARD.md) · [GLOSSARY](studies/shared/GLOSSARY.md)  
+2. Inkling — book `00` → SPEC → arch → `modular_inkling.py`  
+3. Kimi K3 — book `00` → SPEC → lineage → KDA → API → limits  
+4. compare again  
+5. dissertation when data is ready  
 
-- **Inkling:** ~975B / 41B active · 6/256 + 2 shared · relative PE (not RoPE) · hybrid SWA/global · image+audio · 1M ctx  
-- **Kimi K3:** 2.8T · Stable LatentMoE 16/896 · KDA + AttnRes · vision/video · 1M ctx · active params **undisclosed** · weights ETA **2026-07-27**
+## facts (do not invent)
 
----
+- **Inkling:** ~975B / 41B · 6/256+2 · relative PE · SWA/global · image+audio · 1M  
+- **Kimi K3:** 2.8T · 16/896 · KDA+AttnRes · vision/video · 1M · active undisclosed · weights ETA 2026-07-27  
 
-## Offline vs online
+## offline vs online
 
-| Offline | Online only |
+| offline | online |
 |---|---|
-| `make test` / `python3 studies/run.py test` | `make refresh MODEL=inkling` |
-| `make check` (re-derive from local sources) | `make refresh MODEL=kimi-k3` |
-| Read all markdown + vendored `reference/` | Hub / blog re-fetch |
+| `make test` / `make check` | `make refresh MODEL=…` |
+| read `reference/` | Hub / blog re-fetch |
 
-→ [OFFLINE.md](OFFLINE.md) · keep GH+HF synced → `make sync` ([PUSH.md](PUSH.md)) · checklist → [CHECKLIST.md](CHECKLIST.md)
+[OFFLINE.md](OFFLINE.md) · mirrors: `make sync`

dissertation/README.md

@@ -1,24 +1,17 @@
-# Image-Based Music Genre Classification Using CNNs
+# MSc — genre from vinyl covers
 
-**Study home (this monorepo):** [lyffseba/xai](https://github.com/lyffseba/xai) → `dissertation/`  
-**Public citable archive:** [msc-dissertation](https://github.com/lyffseba/msc-dissertation) (notes + notebook live in xai)
+Part of [lyffseba/xai](https://github.com/lyffseba/xai) (**xai** = explainable AI).
 
-MSc work: classify music genres from **vinyl album covers** with CNNs. Two models — batch-norm baseline and **Concept Whitening** for interpretability. CW improves feature transparency without a large accuracy loss.
+CNNs on album art: baseline + Concept Whitening.
 
-## Offline study
+- [main.ipynb](main.ipynb)  
+- public: [msc-dissertation](https://github.com/lyffseba/msc-dissertation)  
+- start: [../START_HERE.md](../START_HERE.md)
 
-Part of the research monorepo with Inkling / Kimi K3. Launchpad: [../START_HERE.md](../START_HERE.md). No network required to **read** the notebook.
+## run
 
-## Run
+1. open notebook  
+2. set local image paths  
+3. run (needs TensorFlow/Keras)
 
-1. Open [main.ipynb](main.ipynb)
-2. Point dataset paths at your local image data
-3. Run all cells (needs TensorFlow / Keras when executing — not offline-stdlib like `studies/`)
-
-## Related (not merged)
-
-| Tree | Role |
-|---|---|
-| This folder | Canonical notebook + study notes for the MSc |
-| `~/lyff/mojo-music-genre-classifier/` | Optional Mojo/MAX port of the same task (product/experiment, separate) |
-| GitHub `msc-dissertation` | Public archive; point homepage at xai |
+Optional Mojo port is outside this repo.

index.html

@@ -3,47 +3,41 @@
 <head>
 <meta charset="utf-8">
 <meta name="viewport" content="width=device-width, initial-scale=1">
-<meta name="description" content="xai — offline research monorepo: Inkling, Kimi K3, MSc dissertation (lyffseba)">
 <title>xai</title>
 <style>
-  :root { --fg:#111; --muted:#444; --line:#bbb; --link:#00c; }
-  body { font-family: Times, "Times New Roman", serif; margin: 0; color: var(--fg); line-height: 1.4; }
-  .wrap { max-width: 44em; margin: 0 auto; padding: 1.4em 1.2em 2.5em; }
-  h1 { font-size: 1.35em; font-weight: normal; margin: 0 0 0.35em; }
-  h2 { font-size: 1.05em; font-weight: normal; margin: 1.25em 0 0.4em; border-bottom: 1px solid var(--line); padding-bottom: 0.12em; }
+  body { font-family: Times, "Times New Roman", serif; margin: 1.5em; max-width: 42em; line-height: 1.35; }
+  h1 { font-size: 1.25em; font-weight: normal; margin: 0 0 0.5em 0; }
   p { margin: 0.5em 0; }
-  a { color: var(--link); }
-  hr { border: none; border-top: 1px solid var(--line); margin: 1.1em 0; }
-  .meta { color: var(--muted); font-size: 0.95em; }
-  .nav a { margin-right: 0.75em; white-space: nowrap; }
-  code { font-size: 0.88em; }
-  table { border-collapse: collapse; width: 100%; font-size: 0.93em; margin: 0.5em 0; }
-  th, td { text-align: left; padding: 0.28em 0.5em 0.28em 0; border-bottom: 1px solid #ddd; vertical-align: top; }
-  th { font-weight: normal; color: var(--muted); }
-  ul { margin: 0.4em 0 0.7em 1.15em; padding: 0; }
-  pre { font-family: ui-monospace, Menlo, Consolas, monospace; font-size: 0.82em; background: #f6f6f6; border: 1px solid #e2e2e2; padding: 0.65em 0.75em; overflow-x: auto; }
+  a { color: #00c; }
+  hr { border: none; border-top: 1px solid #999; margin: 1.1em 0; }
+  .meta { color: #333; font-size: 0.95em; }
+  nav a { margin-right: 0.85em; }
+  code { font-size: 0.9em; }
+  table { border-collapse: collapse; width: 100%; font-size: 0.95em; margin: 0.5em 0; }
+  th, td { text-align: left; padding: 0.2em 0.5em 0.2em 0; border-bottom: 1px solid #ddd; vertical-align: top; }
+  th { font-weight: normal; color: #333; }
+  pre { font-family: monospace; font-size: 0.85em; }
 </style>
 </head>
 <body>
-<div class="wrap">
 
 <h1>xai</h1>
-<p class="meta">offline research monorepo · lyffseba</p>
-<p class="nav meta">
+<p class="meta">xai = explainable AI · offline research notes · lyffseba</p>
+<nav class="meta">
   <a href="START_HERE.md">START HERE</a>
   <a href="studies/derived/DASHBOARD.md">status</a>
-  <a href="studies/STUDY.md">study path</a>
+  <a href="studies/STUDY.md">study</a>
   <a href="OFFLINE.md">offline</a>
-  <a href="https://huggingface.co/spaces/lyffseba/xai">HF portal</a>
-  <a href="https://huggingface.co/datasets/lyffseba/xai-studies">HF dataset</a>
-</p>
+  <a href="https://huggingface.co/spaces/lyffseba/xai">HF</a>
+  <a href="https://huggingface.co/datasets/lyffseba/xai-studies">dataset</a>
+</nav>
 
 <hr>
 
 <table>
-  <tr><th>ID</th><th>Name</th><th>Status</th><th>Total</th><th>Active</th><th>Ctx</th><th>Experts</th></tr>
+  <tr><th>id</th><th>name</th><th>status</th><th>total</th><th>active</th><th>ctx</th><th>experts</th></tr>
   <tr>
-    <td><code>inkling</code></td>
+    <td>inkling</td>
     <td><a href="studies/models/inkling/">Inkling</a></td>
     <td>weights public</td>
     <td>975B</td>
@@ -52,7 +46,7 @@
     <td>6/256+2</td>
   </tr>
   <tr>
-    <td><code>kimi-k3</code></td>
+    <td>kimi-k3</td>
     <td><a href="studies/models/kimi-k3/">Kimi K3</a></td>
     <td>API live · weights pending</td>
     <td>2.8T</td>
@@ -64,33 +58,32 @@
 
 <p>
   <a href="studies/models/inkling/">Inkling</a> —
-  Thinking Machines open MoE · relative PE · hybrid SWA/global · image+audio ·
+  Thinking Machines MoE · relative PE · SWA/global · image+audio ·
   Hub config + transformers under <code>reference/</code>.
 </p>
 <p>
   <a href="studies/models/kimi-k3/">Kimi K3</a> —
-  Moonshot 2.8T Stable LatentMoE · KDA + AttnRes · vision/video ·
-  <code>facts.json</code> + blog/docs local · weights ETA 2026-07-27.
+  2.8T LatentMoE · KDA+AttnRes · vision/video ·
+  <code>facts.json</code> local · weights ETA 2026-07-27.
 </p>
 <p>
-  <a href="studies/shared/comparisons/inkling-vs-kimi-k3.md">Compare</a> ·
-  <a href="dissertation/">MSc dissertation</a>
-  (<a href="dissertation/main.ipynb">main.ipynb</a> · vinyl covers · CNN + Concept Whitening)
+  <a href="studies/shared/comparisons/inkling-vs-kimi-k3.md">compare</a> ·
+  <a href="dissertation/">MSc</a>
+  (<a href="dissertation/main.ipynb">main.ipynb</a> · vinyl · CNN + Concept Whitening)
 </p>
 
-<h2>Verify</h2>
+<hr>
+
 <pre>make test
-# or
 python3 studies/run.py test</pre>
-<p class="meta">stdlib only · no pip · no network · expect <code>✓ test passed</code></p>
+<p class="meta">stdlib · no pip · no network · expect ✓ test passed</p>
 
 <p class="meta">
-  <a href="studies/shared/GLOSSARY.md">Glossary</a> ·
-  <a href="CHECKLIST.md">Checklist</a> ·
-  <a href="https://github.com/lyffseba/xai">GitHub</a> ·
+  <a href="studies/shared/GLOSSARY.md">glossary</a> ·
+  <a href="https://github.com/lyffseba/xai">github</a> ·
+  <code>make sync</code> ·
   last update: 2026-07-17
 </p>
 
-</div>
 </body>
 </html>

studies/README.md

@@ -1,27 +1,20 @@
 # studies
 
-Source-driven model notes. Offline. Stdlib Python only.
+Source-driven model notes for **xai** (xai = explainable AI). Offline. Stdlib only.
 
-## Packages
-
-- [inkling](models/inkling/) — ~975B / 41B, 1M, image+audio
-- [kimi-k3](models/kimi-k3/) — 2.8T, 16/896 MoE, KDA, 1M
-- [compare](shared/comparisons/inkling-vs-kimi-k3.md)
-- [status](derived/DASHBOARD.md) · [path](STUDY.md) · [glossary](shared/GLOSSARY.md)
-
-## Commands
+- [inkling](models/inkling/) — 975B/41B · 1M · image+audio  
+- [kimi-k3](models/kimi-k3/) — 2.8T · 16/896 · KDA · 1M  
+- [compare](shared/comparisons/inkling-vs-kimi-k3.md) · [status](derived/DASHBOARD.md) · [path](STUDY.md) · [glossary](shared/GLOSSARY.md)
 
 ```
 make test
 python3 run.py test
 make check
-make refresh MODEL=inkling   # online
+make refresh MODEL=inkling
 ```
 
-## Layout
-
 ```
 models/<id>/{book,code-study,reference,derived,scripts}
 ```
 
-`reference/` = primary sources. `derived/` = generated (do not hand-edit).
+`reference/` = primary sources. `derived/` = generated.