Source of Truth

This project treats source code, tests, and authored notes as durable study material. Generated docs and PDFs must be reproducible from those inputs, and employer-specific interview prep must never enter this tree. This page is the whole contract.

The runnable interview track is currently Python. Cross-language comparisons in reference material are context, not runnable practice tracks.

Authored Material

Path	Purpose
`src/algo/`	Algorithm implementations and docstrings
`tests/`	Examples and property-based correctness checks
`src/concepts/`	Technical concept modules
`src/practice/`	Code-reading and decomposition exercises
`reference-sheets/`	Printable notes, quick-reference sheets, and the practice method (sheets 10–11)
`reference-sheets/appendix-topics.json`	Structured appendix topics rendered into the booklet compile-safely

Generated Material

Output	Command
`booklet.tex` / `booklet.pdf` / `docs/assets/booklet.pdf`	`just packet`
`reference-sheets/pdf/*.pdf`	`just pdf-all`
local MkDocs output in `site/`	`just docs-build`

Production reading pages use a separate static site repository. Its scripts/sync-content.mjs reads a public packet checkout at HEAD, records the packet commit and source paths in src/content/.manifest.json, and prerenders dsa-woodshed.space. This repository owns the content; the site repository owns the shell, navigation, and rendering. The local site/ output is a development artifact, not the production deployment.

Three Layers

Layer	What	Where	Public?
L1: knowledge	algorithms, tests, concepts, reference sheets 01–09	this repo	✅
L2: practice method	sheets 10–11; `just practice-open`, `just practice-study`, `just practice-start`, `just practice-start-tests`, `just practice-next`, `just practice-test`, `just interview`, atomic finish commands, and spaced repetition	this repo	✅
L3: private overlay	employer/panel front & back matter, people, dates, tailored positioning, personal rep logs	a private downstream repo	⛔ never here

The distinction that keeps this publishable: L2 is a method ("how to reason, code, test, and narrate under observation," a general, teachable skill); L3 is a dossier (who is on Tuesday's panel). Method publishes; dossier never touches this tree. If a sentence only makes sense for one employer, it is L3.

The Graph Edge Is One-Way

private superset repo                    THIS REPO (public)
─────────────────────                    ──────────────────
bazel_dep + path/git override   ──────▶  module(name = "dsa_study_packet")
\includepdf{@dsa_study_packet            //:booklet  (neutral PDF)
            //:booklet}
+ private front/back matter (L3)
→ private //:study_packets

The private repo depends on this one; this repo knows nothing about any downstream. Composition is overlay, not bundle: the neutral booklet is compiled here from this tree (so it cannot contain employer text), and a private lane \includepdfs the finished PDF between its own front and back matter. The two bodies of text only ever meet as adjacent pages in a private output, never in source. Delete the private repo and this SSOT is unchanged and complete. examples/overlay-demo/ is a self-contained, fork-me demonstration of the pattern with placeholder content.

The Boundary Is Enforced

scripts/check_public_boundary.py (just public-boundary) fails if any tracked file contains secret-shaped content or secret-file paths: tracked SOPS or dotenv files, age keys, GitHub tokens, private-key blocks, legacy secret tripwires. Name-specific tripwires (employers, panels, private repo names) are deliberately not listed here. The private downstream repo scans this tree with its own marker list, so the public guard cannot itself disclose what it guards against. Neutrality is a machine check, not a promise. Personal rep scores (.challenges/reps.md) are gitignored by default; publish aggregates deliberately, or not at all.

Who Owns The Practice Method

L2 is one loop described from five angles. Each surface owns exactly one question about it; none restates another's answer.

Surface	Owns	Does not own
Sheet 10	the CLARP method + self-review rubric: how to perform	no calendar, no daily loop
Sheet 11	the 14-day editor-first calendar: when. Its prose feeds `scripts/practice_day.py`; edit sheet 11 itself for loop changes	no method detail, no rubric
Evidence page	why: research citations and the video shelf	no calendar, no rubric scoring
`AGENTS.md` persona	how the resident interviewer behaves during a rep; regenerates the `.github` surfaces via `just gen-agents`	no citations, no calendar
Skills (`practice-day`, `interviewer`)	routing: which mode and `just` command runs next	no method content of its own

A change landing on the wrong row is the drift tell: a calendar edit inside sheet 10, a rubric row inside the evidence page, or a citation inside AGENTS.md all mean the edit belongs one row up or down this table instead.

Runbook

I want to…	Do this	Layer
Add/extend a study or practice sheet	`reference-sheets/NN-.md` + one-line `docs/reference/NN-.md` wrapper; nav + PDF pick it up	L1/L2
Add an algorithm	`src/algo/<topic>/<problem>.py` + tests; booklet and site regenerate	L1
Prep for a specific employer/panel	Create a lane in the private downstream repo; author front/back matter there	L3
Keep practice tapes / mock scripts / rep scores	Private repo or gitignored local dir, never tracked here	L3