Import the quickstart guide (curated snapshot of start-training)

[LukasWodka]

Summary

Single-commit import of the start-training quickstart guide into this repo, as part of the repository restructuring. The tree is the curated snapshot of start-training at its develop tip (5af8e509d404), with README and notebook links re-pointed at tracebloc/quickstart and internal references scrubbed (this repo is public).

Included (18 files)

• README.md, LICENSE, .gitignore, .github/pull_request_template.md
• notebooks/traceblocTrainingGuide.ipynb, notebooks/GenerateCheckWeights.ipynb
• notebooks/templates/ — README.md, families.json, verification-dev.json, and the 6 family templates (embeddings, nlp_finetune, nlp_generative, survival, tabular_timeseries, vision_from_scratch)
• scripts/check_templates.py, scripts/check_templates_mutations.py
• .github/workflows/template-rules.yml — the self-contained checker workflow from the snapshot, re-added unchanged so the checker actually runs on PRs and pushes. It calls no shared org workflow; its two action pins (actions/checkout v7.0.1, actions/setup-python v5.6.0) are kept as SHA pins.

Excluded

• The 10 thin caller workflows that reference the shared org workflows (kanban, bugbot gate, code quality, FR gate, stale backlog, etc.) — to be added separately for this repo.
• CLAUDE.md, .cursor/BUGBOT.md — repo-specific, added separately.

Scrub (old → new, per file)

Old values are given in redacted form here on purpose; the point of the scrub is that they do not appear in this public repo.

File	Old	New
README.md (Which Pythons work)	trailing (backend#NNNN) ticket ref	parenthetical removed; sentence ends at "fixed for."
notebooks/templates/README.md title	Family training templates (<RFC id> D9)	Family training templates
notebooks/templates/README.md (validator paragraph)	engine registry (<RFC id> D10)	engine registry
notebooks/templates/README.md (Verification status)	Survival is UNVERIFIED — tracebloc/backend#NNNN.	Survival is UNVERIFIED — tracked internally.
notebooks/templates/README.md (Reaching users note)	still points at (tracebloc/backend#NNNN) — nothing	still points at — nothing
notebooks/templates/families.json _comment	The <RFC id> D9 pre-fill table	The family-template pre-fill table
notebooks/templates/families.json Survival unverified	tracebloc/backend#NNNN	tracked internally: the dev time_to_event_prediction datasets carry time-like labels instead of a 0/1 event indicator
notebooks/templates/verification-dev.json _comment	records for the <RFC id> D9 family templates (tracebloc/backend#NNNN). D9 ships a template only once…	records for the family templates. A template ships only once…
6 template notebooks, notebook-level metadata.tracebloc	"rfc": "<RFC id> D9"	key dropped
scripts/check_templates.py docstring (rule 17)	(where BUGBOT.md permits a bare tracebloc/backend#NNNN)	clause removed
scripts/check_templates.py comment above INTERNAL_REF	where .cursor/BUGBOT.md explicitly permits a bare tracebloc/backend#NNNN	clause removed
scripts/check_templates.py comment above INTERNAL_REF	(`backend#N`, an RFC id)	(one `<repo>#N` reference, an RFC id)

Why the notebook key could be dropped: the checker reads only cell-level metadata.tracebloc (rules 8/9); nothing reads the notebook-level block, so the rfc key was informational. No checker rule or regex was changed — INTERNAL_REF already matches the form (<repo>#N, RFC-…, non-public hosts) rather than a list of spellings, so only its comments needed rewording. The unverified field is free-form (rule 13 tests it for truthiness), so plain text is accepted.

Verification

Run locally exactly as template-rules.yml runs them, from the repo root, with tracebloc installed --no-deps plus numpy psutil requests rich termcolor tqdm (resolved to tracebloc 1.0.11, _SURVIVAL_FRAMEWORKS = ['lifelines', 'scikit_survival', 'sklearn']):

$ python3 scripts/check_templates_mutations.py
OK — 54 of 54 mutations run, every one caught by the rule that owns it.

$ TRACEBLOC_CHECK_STRICT=1 python3 scripts/check_templates.py
note: single_pass_frameworks matches tracebloc 1.0.11 (floor 1.0.7): ['lifelines', 'scikit_survival', 'sklearn'].
OK — 6 families, 16 categories, all D9 rules hold.

Both were also run on the unscrubbed tree first and gave the same result, so the scrub changed nothing the checker measures.

Forbidden-string scan over the whole tree (private ticket prefixes, RFC ids, internal repo paths, tenant names; case-insensitive):

$ grep -rniE '<internal-ref and tenant-name patterns>' --exclude-dir=.git .
(no output — zero hits)

$ gitleaks detect --no-git -s .
no leaks found

Checklist

• Checker + mutation harness green locally, same commands as CI
• Zero internal references / tenant names in the tree
• gitleaks clean
• Targets develop

---

Note

Low Risk
Mostly documentation, notebooks, and validation tooling; no production runtime changes. CI adds third-party tracebloc install on GitHub Actions only.

Overview
This PR imports the curated start-training quickstart snapshot into tracebloc/quickstart: a public-facing README (Colab/local paths), Apache 2.0 license, .gitignore, and a PR template, plus the main traceblocTrainingGuide and GenerateCheckWeights notebooks.

It adds notebooks/templates/ — six family training templates (vision, NLP fine-tune/generative, embeddings, tabular/time series, survival) driven by families.json and dev verification in verification-dev.json, with documentation of the render contract and which families are verified vs Survival (unverified).

Enforcement is new in CI: scripts/check_templates.py (D9 rules), a mutation harness, and .github/workflows/template-rules.yml run both on every PR/push (no path filter), install a minimal tracebloc for SDK mirror checks, and use TRACEBLOC_CHECK_STRICT=1. Internal ticket/RFC references are scrubbed from prose and template metadata for this public repo.

^{Reviewed by Cursor Bugbot for commit be0f7cb. Bugbot is set up for automated code reviews on this repo. Configure here.}

[LukasWodka]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit be0f7cb. Configure here.}