Regulatory data on disk (`data/regulatory` and cousins)

Status: October 2026 (slices 30–34).

Locus keeps legal text and evidence in three places that must not be confused:

Location	Owner	Purpose
`data/regulatory/fixtures/` (repo root)	Engineering / CI	Tiny synthetic PDFs + `manifest.json` to exercise the platform legal vault pipeline without shipping official gazettes
Postgres + R2 (or local vault dir)	Platform ops (`locus-ops`)	Production instrument / edition / provision / artifact system of record — real PDFs after human review
`locus-web/data/regulatory/`	Dev / analysis engines	Legacy parsed JSON / SQLite corpus used by `catalog.ts` and retrieval until vault provisions are `legal_reviewed`

Tenant deal-room PDFs (term sheets, SYSCOHADA packs) live under the tenant document vault (R2), not under data/regulatory/.

1. `data/regulatory/fixtures/` (what the PDFs are for)

Purpose

These PDFs are CI and local-dev stand-ins for foundational CEMAC instruments that LegalVaultBootstrap creates as stub editions (e.g. cosumaf-rg-2014__edition_bootstrap). They exist so that:

Artifact register / SHA-256 / PdfPig extract / review → publish can be tested without copyrighted full gazettes in the repo.
scripts/attach-vault-fixtures.sh can attach each file to its bootstrap edition via
POST /api/admin/regulatory/editions/{editionId}/artifact (PLATFORM_OPERATOR session).
Contract tests (locus-ops/tests/legal_vault_amendments_applicability_fixtures.test.ts) can assert unique SHA-256 values and a complete instrument list.

They are not the production legal corpus. Digests in manifest.json must never be presented as official source fingerprints.

File	Bootstrap edition	Authority
`syscohada-audcif-2017.pdf`	`syscohada-audcif-2017__edition_bootstrap`	OHADA
`ohada-aus-2010.pdf`	`ohada-aus-2010__edition_bootstrap`	OHADA
`cosumaf-rg-2014.pdf`	`cosumaf-rg-2014__edition_bootstrap`	COSUMAF
`cosumaf-inst-ape-2019.pdf`	`cosumaf-inst-ape-2019__edition_bootstrap`	COSUMAF
`beac-reg-changes-2018.pdf`	`beac-reg-changes-2018__edition_bootstrap`	BEAC
`cobac-r-2016-01.pdf`	`cobac-r-2016-01__edition_bootstrap`	COBAC
`manifest.json`	—	Maps instrument → edition → fixture → SHA-256

Regenerate deterministically:

bash

python3 scripts/generate-regulatory-fixtures.py

How the PDFs are used (pipeline)

text

generate-regulatory-fixtures.py
        │
        ▼
data/regulatory/fixtures/*.pdf  +  manifest.json
        │
        │  attach-vault-fixtures.sh  (or locus-ops "Upload PDF")
        ▼
SoR LegalSourceArtifacts  (storageKey = local:<abs path> or R2 key)
        │
        ├── LegalVaultDocumentExtractor (pages, preview, candidate articles)
        ├── LegalReviewItem → human APPROVE
        └── LegalVaultPublisher → edition legal_reviewed / in_force
                │
                ▼
Tenant analyzeDealWithVault → resolve edition/provision/artifact
        → open-statute BFF streams PDF (hash-verified)
        → DossierRegulatoryFindings stores vault IDs

Staging / production: attach official PDFs in locus-ops Legal Vault (Upload PDF per edition), then approve in the Review Queue. Do not rely on fixture digests in production.

Local attach (dev):

bash

export LOCUS_SESSION_TOKEN=…   # PLATFORM_OPERATOR session
export LOCUS_ORG_ID=…          # that session's org id
# optional: DRY_RUN=1
./scripts/attach-vault-fixtures.sh

The SoR must be able to read local: paths (same machine as the API). For R2-backed deploys, use the ops UI upload instead.

2. `locus-web/data/regulatory/` (engine corpus, not vault)

Path	Role
`parsed/`	JSON article chunks for retrieval / curation stats
`regulations/`, `manifests/`, `checklists/`	Supporting regulatory engine assets
`cosumaf_legal_kb.sqlite`	Local COSUMAF knowledge base used by some auditor paths
`sample_drafts/`, `approved_precedents/`	Sample texts for tooling — not tenant data

catalog.ts still holds stable instrument IDs and evaluation predicates; article bodies gradually move to SoR provisions after ops legal review. Bank UI analyzeDeal can run without vault PDFs; analyzeDealWithVault overlays vault IDs when the session can resolve them.

3. What is not in `data/regulatory`

Tenant deal-room uploads (R2 + Documents table)
Market bulletins (market ingest → MarketBulletins / observations)
Live Watchtower downloads (stored under LEGAL_VAULT_STORAGE_PATH or R2 as LegalSourceArtifacts)

Bank workspace screens: workspace-guide.md
Ops app: ../locus-ops/README.md
Tracker R5/R7: ../IMPLEMENTATION_TRACKER.md
OCR override SOP: ocr-override-sop.md

Regulatory data on disk (data/regulatory and cousins) ​

1. data/regulatory/fixtures/ (what the PDFs are for) ​

Purpose ​

Contents ​

How the PDFs are used (pipeline) ​

2. locus-web/data/regulatory/ (engine corpus, not vault) ​

3. What is not in data/regulatory ​

Related docs ​