docs(adr): record octad/verification/justfile decisions, strip empty trees

Adds three ADRs for verisimiser and one for verisimdb-data, plus the
mechanical file deletions they authorise.

- ADR-0001 (octad-ontology): concerns octad is canonical; modalities
  become Tier 2 overlays. Closes #19; sets up #20; closes #21 wontfix.
- ADR-0002 (verification-tree): strip the empty 8-subdirectory tree;
  Idris2 stubs in src/interface/abi/ are unaffected. Closes #15.
- ADR-0003 (justfile-aspirational-recipes): delete recipes that name
  non-existent clap subcommands. Closes #11 (#10 is the mechanical
  follow-up).
- ADR-0001 (verisimdb-data, repo-purpose): repo carries two explicit
  purposes (scan store + ABI dogfood). Lands in the data repo commit.

Deletes:
- examples/SafeDOMExample.res, examples/web-project-deno.json
  (unrelated template flotsam — closes #12)
- root SECURITY.md, root CODE_OF_CONDUCT.md (duplicate; .github/
  versions are canonical — closes #13, #14)
- verification/ subtree (closes #15 via ADR-0002)

Closes #11, #12, #13, #14, #15, #19, #21

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: hyperpolymath

CODE_OF_CONDUCT.md

@@ -0,0 +1,102 @@
+// SPDX-License-Identifier: PMPL-1.0-or-later
+// Copyright (c) 2026 Jonathan D.A. Jewell (hyperpolymath) <j.d.a.jewell@open.ac.uk>
+= ADR-0001: Canonical octad ontology — concerns, not modalities
+:revdate: 2026-05-13
+:status: Accepted
+
+== Status
+
+Accepted — 2026-05-13.
+
+Resolves: https://github.com/hyperpolymath/verisimiser/issues/19[V-L1-A1].
+
+Closes as wontfix: https://github.com/hyperpolymath/verisimiser/issues/21[V-L1-A3] (the modalities-first refactor).
+
+Unblocks: https://github.com/hyperpolymath/verisimiser/issues/20[V-L1-A2] (the README rewrite).
+
+== Context
+
+Two competing ontologies have lived in this repository under the same name
+("octad"):
+
+Modalities octad (README §"VeriSimDB's Octad: Eight Modalities")::
+Graph · Vector · Tensor · Semantic · Document · Temporal · Provenance · Spatial.
+These are *representations* of an entity — how the same data is stored in
+different shapes. The README's eight cross-modal drift categories
+(structural, semantic, temporal, statistical, referential, provenance,
+spatial, embedding) presuppose this ontology.
+
+Concerns octad (`src/abi/mod.rs::OctadDimension`, `src/manifest/mod.rs::OctadConfig`, `src/main.rs::print_octad`)::
+Data · Metadata · Provenance · Lineage · Constraints · AccessControl ·
+Temporal · Simulation. These are *concerns/aspects* of data — what you
+want to know or enforce about it.
+
+The code commits to the concerns octad; the README leads with the modalities
+octad. A user cannot answer "what does an octad-augmented entity look like?"
+without picking one.
+
+== Decision
+
+The *concerns* octad is canonical.
+
+The eight dimensions of the verisimiser octad are:
+
+. **Data** — the original entity as stored in the target database.
+. **Metadata** — schema and type information.
+. **Provenance** — SHA-256 hash-chain tracking of who did what and when.
+. **Lineage** — directed-edge graph of data derivation (target nameschematically a DAG; see ADR-0004 when written).
+. **Constraints** — cross-dimensional invariant enforcement, including
+  drift detection between Data + Metadata + active overlays.
+. **AccessControl** — policy-based row/column-level access permissions.
+. **Temporal** — version history with point-in-time queries and rollback.
+. **Simulation** — what-if branching and sandbox query execution.
+
+Modalities (Graph, Vector, Tensor, Semantic, Document, Spatial) are
+*Tier 2 overlays* — independent representational projections that a user
+can enable per-entity for similarity search, full-text search, geospatial
+indexing, etc. They are not "the octad" and not co-equal with the eight
+concerns. Provenance and Temporal in the modalities list collapse onto
+the same-named concerns.
+
+The eight "cross-modal drift categories" become *symptoms observed by
+the Constraints concern* when Data, Metadata, and the active overlays
+disagree:
+
+[cols="1,2"]
+|===
+| Drift category (legacy framing) | Where it lives in the concerns ontology
+
+| Structural | Constraints (Data vs Metadata schema agreement)
+| Semantic   | Constraints across overlays
+| Temporal   | Constraints between Temporal versions across overlays
+| Statistical | Constraints over Tier 2 vector/tensor overlay drift
+| Referential | Constraints between Tier 2 graph overlay and Data
+| Provenance | Constraints over Provenance chain integrity
+| Spatial    | Constraints over Tier 2 spatial overlay
+| Embedding  | Constraints between Tier 2 vector overlay and source documents
+|===
+
+== Consequences
+
+. README and ROADMAP must be rewritten to drop the modalities octad table
+  and reframe the drift categories under Constraints. Tracked as V-L1-A2.
+. The modalities-first refactor (V-L1-A3) is *not* done. Closed as wontfix.
+. Tier 2 design (V-L1-F, V-L1-G, V-L1-H) continues to use modality terms
+  for overlays — they are just no longer presented as "the octad."
+. `OctadDimension` enum and `OctadConfig` fields are stable; no source-level
+  rename triggered by this ADR.
+
+== Alternatives considered
+
+Modalities octad as canonical::
+Rejected. Would have required rewriting `abi`, `manifest`, `codegen`, tests,
+and example manifests — a multi-week change for pre-alpha framing. The
+codebase had already converged on the concerns ontology; the only cost of
+keeping it is doc updates.
+
+"Both, with one as primary"::
+Considered but rejected. Two ontologies sharing a name is the bug;
+ranking them doesn't fix it.
+
+Renaming "octad" to something else (e.g. "octave")::
+Out of scope here. The brand survives; the contents are defined.

SECURITY.md

@@ -0,0 +1,66 @@
+// SPDX-License-Identifier: PMPL-1.0-or-later
+// Copyright (c) 2026 Jonathan D.A. Jewell (hyperpolymath) <j.d.a.jewell@open.ac.uk>
+= ADR-0002: Strip the empty verification/ subtree
+:revdate: 2026-05-13
+:status: Accepted
+
+== Status
+
+Accepted — 2026-05-13.
+
+Resolves: https://github.com/hyperpolymath/verisimiser/issues/15[V-L3-D1].
+
+== Context
+
+The `verification/` tree contained eight subdirectories
+(`benchmarks/`, `coverage/`, `fuzzing/`, `proofs/`, `safety_case/`,
+`simulations/`, `tests/`, `traceability/`), each with a ~20-byte
+`README.adoc` and an a2ml manifest. Zero proofs, zero benchmarks,
+zero fuzzing, zero safety case.
+
+(The actual Idris2 ABI declarations and Zig FFI implementation live in
+`src/interface/abi/{Types,Layout,Foreign}.idr` and `src/interface/ffi/`
+respectively — those are real Phase 0 stubs and are unaffected by this
+ADR. The README's reference to "Idris2 proofs (in `src/interface/abi/`)"
+already points at the correct location for the stubs; only the proofs
+themselves are still future work.)
+
+A reader following the README's verification/-shaped tree was misled
+into thinking benchmarks, fuzzing, safety cases, and traceability
+artifacts existed for this product.
+
+== Decision
+
+Strip the empty `verification/` subtree from the repository.
+
+Reintroduce `verification/` only when there is something to put in it.
+The first content is expected to be a property-test harness landing
+alongside V-L1-C1 (the Tier 1 SQLite piggyback) — at that point a
+single `verification/property-tests/` directory will be added with
+real content. Don't predeclare the rest of the eight subdirs until
+they have artifacts.
+
+== Consequences
+
+. The empty `verification/` subtree is deleted in the same change set
+  that lands this ADR.
+. The README's "Status" / "Architecture" copy that referenced
+  `verification/` is reworded to point at `src/interface/abi/` only
+  (where the Idris2 stubs actually live).
+. Future verification artifacts (proofs, benchmarks, fuzz targets) are
+  added directory-by-directory as they appear, not predeclared. The
+  first such addition is expected alongside V-L1-C1 (Tier 1 SQLite
+  piggyback) as a property-test harness.
+
+== Alternatives considered
+
+Populate the subtree::
+Rejected for now. Writing eight stub READMEs costs effort and doesn't
+move the product forward. The contracts the README cited (drift
+correctness, chain integrity, version ordering, sidecar isolation)
+become provable only once the Tier 1 implementation exists, which is
+V-L1-C1.
+
+Leave the subtree, soften the README::
+Rejected. Empty scaffolding misleads contributors who follow the tree
+looking for examples to extend.

docs/decisions/ADR-0001-octad-ontology.adoc

@@ -0,0 +1,52 @@
+// SPDX-License-Identifier: PMPL-1.0-or-later
+// Copyright (c) 2026 Jonathan D.A. Jewell (hyperpolymath) <j.d.a.jewell@open.ac.uk>
+= ADR-0003: Remove aspirational Justfile recipes
+:revdate: 2026-05-13
+:status: Accepted
+
+== Status
+
+Accepted — 2026-05-13.
+
+Resolves: https://github.com/hyperpolymath/verisimiser/issues/11[V-L3-C2].
+Unblocks: https://github.com/hyperpolymath/verisimiser/issues/10[V-L3-C1] (the
+mechanical fix to the broken recipe block).
+
+== Context
+
+`Justfile` carried a recipe block whose source contained literal `\n`
+characters where newlines were intended. The intent was three recipes
+(`augment`, `check-octad`, `migrate`) calling clap subcommands of the
+same names. The clap CLI in `src/main.rs` exposes
+`init / generate / start / drift / provenance / history / status / octad`
+— there are no `augment`, `check-octad`, or `migrate` subcommands.
+
+So the Justfile advertised commands that don't exist, and did so in a
+syntactically broken way that made the breakage invisible to `just --list`.
+
+== Decision
+
+Delete the three aspirational recipes from `Justfile`.
+
+The Justfile should only reference subcommands that exist. When new
+subcommands are added (per their own design issues), recipe wrappers
+can be added at the same time.
+
+== Consequences
+
+. The broken `\n`-collapsed block is removed cleanly (V-L3-C1).
+. `just --list` no longer shows phantom recipes.
+. Future "convenience wrapper" recipes must be added in the same change
+  set as the underlying subcommand.
+
+== Alternatives considered
+
+Add the missing subcommands::
+Out of scope for V-L3-C2/C1, which are ground-clearing issues. The
+semantics of `augment` / `check-octad` / `migrate` aren't pinned down —
+e.g. `augment DB_URL` overlaps with `generate` (overlay DDL) plus `start`
+(daemon). Designing them is a separate issue, not a one-line fix.
+
+Keep the recipes pointing at not-yet-implemented subcommands::
+Rejected. The recipes would still fail with "Unknown subcommand"; the
+broken UX would just be in a different place.