Syllabus

A 15-week course meeting twice weekly, ~28 sessions, structured to be taught by an instructor or self-paced by an individual reader. Each session is either a lecture (one chapter) or a lab (one runnable exercise set).

Calendar

Wk	Sessions	Chapters	Deliverable
1	The crisis · Falsifiability	1–2	Reading reflection
2	FRE primer · Canon	3–4	Worked-attestation walk-through
3	Hashing · Lab 5	5	Hand-computed SHA-256 chain
4	Signatures · Lab 6	6	Ed25519 keygen + sign + verify
5	Canonicalization · Lab 7	7	RFC 8785 round-trip; DSSE PAE signing; cross-impl interop
6	Schemas · Lab 8	8	Pydantic Canon validator
7	Embeddings · Hybrid retrieval	9–10	RRF over BM25 + nomic-embed; pgvectorscale diskann; ParadeDB BM25
8	LLM extraction · Adversarial validation	11–12	Tri-model harness over toy corpus; Outlines constrained decoding; Langfuse observability
9	The seven-layer pipeline · Postgres substrate	13–14	Local pgvector + PostGIS instance
10	Idempotent ingest · Procedural primitives	15–16	Matter / parties / acquisition / hold
11	TARG · Epistemic Neutrality Masking	17–18	Identity resolution exercise; Presidio PII masking lab
12	Five challenges · Four attestation kinds	19–20	EnrichmentAttestation over a real email
13	Schemas/migrations · Local-first chunking · Keys	21–23	Unstructured.io section-aware partitioning; production-grade key rotation; cross-platform keyring (keyrings.alt for headless)
14	Audit log · Reference verifier · Admissibility Auditor	24–26	DSSE verifier; Rekor/Sigstore transparency log; standalone verifier published to PyPI
15	Capstone presentations	27–30	Domain Meridian, end-to-end demo; Dagster pipeline; WeasyPrint/Typst PDF output

Grading (instructor-led variant)

Component	Weight
Weekly labs (12)	30%
Two written midterm exercises	15%
Conformance test suite contribution	10%
Capstone artifact (sealed corpus)	30%
Capstone defense (oral, 30 min)	15%

Prerequisites

One semester of data structures and algorithms.
Comfortable Python (decorators, generators, type hints).
Familiarity with SQL.
Has used git for at least a small project.
Has read and understood at least one RFC end-to-end (any RFC).

A short diagnostic in front/03_prerequisites.md lets the reader self-assess before starting.

Lab environment

macOS, Linux, or WSL2. Windows-native is supported with keyrings.alt as the keyring backend (Chapter 23); macOS Keychain and Linux SecretService are used on their respective platforms.
Python ≥ 3.10 (CI matrix: 3.10, 3.12, 3.13).
Postgres ≥ 16 with pgvector, pg_trgm, and postgis extensions. Optional: pg_search (ParadeDB BM25, Chapter 9–10), pgvectorscale (StreamingDiskANN, Chapter 9–10).
~50 GB free disk for the optional reference corpus.
A GPU or rented GPU access for the enrichment/refutation labs (chapters 11–12, 19); the lab manual documents how to substitute a smaller-model workflow on CPU at a precision cost.
Optional extras install additional tooling per chapter; see front/03_prerequisites.md for the full extras list.

Capstone deliverable

The student submits, by the end of week 15:

A working corpus ingester for a domain other than the reference personal-data corpus. (Examples: open court records, a journalistic FOIA pull, a research archive, a non-profit’s internal records.)
A sealed EnrichmentAttestation and SearchAttestation over that corpus, both passing the standalone verifier.
A 4–8 page design memo identifying the corpus’s domain-specific challenges and how the customizations addressed them.
A 30-minute oral defense.

The capstone is the assessment. Everything before it is preparation.

What the course does not attempt to teach

The internal mathematics of cryptographic primitives. Pointers to primary sources are provided; this course teaches their use.
Trial advocacy. Counsel will have a different course for that.
Machine-learning research. The models are components, not subjects of research, in this course.
Any specific jurisdiction’s evidentiary doctrine beyond the federal baseline and a Wisconsin-specific appendix.

# Syllabus {.unnumbered} A 15-week course meeting twice weekly, ~28 sessions, structured to be taught by an instructor or self-paced by an individual reader. Each session is either a *lecture* (one chapter) or a *lab* (one runnable exercise set). ## Calendar | Wk | Sessions | Chapters | Deliverable | | --- | ----------------------------------------------------- | -------------- | ------------------------------------------- | | 1 | The crisis · Falsifiability | 1–2 | Reading reflection | | 2 | FRE primer · Canon | 3–4 | Worked-attestation walk-through | | 3 | Hashing · Lab 5 | 5 | Hand-computed SHA-256 chain | | 4 | Signatures · Lab 6 | 6 | Ed25519 keygen + sign + verify | | 5 | Canonicalization · Lab 7 | 7 | RFC 8785 round-trip; DSSE PAE signing; cross-impl interop | | 6 | Schemas · Lab 8 | 8 | Pydantic Canon validator | | 7 | Embeddings · Hybrid retrieval | 9–10 | RRF over BM25 + nomic-embed; pgvectorscale diskann; ParadeDB BM25 | | 8 | LLM extraction · Adversarial validation | 11–12 | Tri-model harness over toy corpus; Outlines constrained decoding; Langfuse observability | | 9 | The seven-layer pipeline · Postgres substrate | 13–14 | Local pgvector + PostGIS instance | | 10 | Idempotent ingest · Procedural primitives | 15–16 | Matter / parties / acquisition / hold | | 11 | TARG · Epistemic Neutrality Masking | 17–18 | Identity resolution exercise; Presidio PII masking lab | | 12 | Five challenges · Four attestation kinds | 19–20 | EnrichmentAttestation over a real email | | 13 | Schemas/migrations · Local-first chunking · Keys | 21–23 | Unstructured.io section-aware partitioning; production-grade key rotation; cross-platform keyring (keyrings.alt for headless) | | 14 | Audit log · Reference verifier · Admissibility Auditor| 24–26 | DSSE verifier; Rekor/Sigstore transparency log; standalone verifier published to PyPI | | 15 | Capstone presentations | 27–30 | Domain Meridian, end-to-end demo; Dagster pipeline; WeasyPrint/Typst PDF output | ## Grading (instructor-led variant) | Component | Weight | | -------------------------------------- | ------ | | Weekly labs (12) | 30% | | Two written midterm exercises | 15% | | Conformance test suite contribution | 10% | | Capstone artifact (sealed corpus) | 30% | | Capstone defense (oral, 30 min) | 15% | ## Prerequisites - One semester of data structures and algorithms. - Comfortable Python (decorators, generators, type hints). - Familiarity with SQL. - Has used `git` for at least a small project. - Has read and understood at least one RFC end-to-end (any RFC). A short diagnostic in `front/03_prerequisites.md` lets the reader self-assess before starting. ## Lab environment - macOS, Linux, or WSL2. Windows-native is supported with `keyrings.alt` as the keyring backend (Chapter 23); macOS Keychain and Linux SecretService are used on their respective platforms. - Python ≥ 3.10 (CI matrix: 3.10, 3.12, 3.13). - Postgres ≥ 16 with `pgvector`, `pg_trgm`, and `postgis` extensions. Optional: `pg_search` (ParadeDB BM25, Chapter 9–10), `pgvectorscale` (StreamingDiskANN, Chapter 9–10). - ~50 GB free disk for the optional reference corpus. - A GPU or rented GPU access for the enrichment/refutation labs (chapters 11–12, 19); the lab manual documents how to substitute a smaller-model workflow on CPU at a precision cost. - Optional extras install additional tooling per chapter; see `front/03_prerequisites.md` for the full extras list. ## Capstone deliverable The student submits, by the end of week 15: 1. A working corpus ingester for a domain other than the reference personal-data corpus. (Examples: open court records, a journalistic FOIA pull, a research archive, a non-profit's internal records.) 2. A sealed `EnrichmentAttestation` and `SearchAttestation` over that corpus, both passing the standalone verifier. 3. A 4–8 page design memo identifying the corpus's domain-specific challenges and how the customizations addressed them. 4. A 30-minute oral defense. The capstone is the assessment. Everything before it is preparation. ## What the course does *not* attempt to teach - The internal mathematics of cryptographic primitives. Pointers to primary sources are provided; this course teaches their *use*. - Trial advocacy. Counsel will have a different course for that. - Machine-learning research. The models are components, not subjects of research, in this course. - Any specific jurisdiction's evidentiary doctrine beyond the federal baseline and a Wisconsin-specific appendix.