Reference
Every TOML key, tagged section, eval category, and
deployment tier defined by the format. Each entry links to
the section of the specification that defines it.
Module metadata · commonsformat.toml
| field | description | defined |
| commonsformat | Format version this module targets, as a string (e.g., "0.2"). Required. | §6 |
| name | Module name. Lower-case ASCII letters, digits, and hyphens; must not begin or end with a hyphen. Required. | §6 |
| version | SemVer-shaped MAJOR.MINOR.PATCH string with no pre-release or build suffix. Required. | §6 |
| description | One-to-three sentence summary, plain text. Required. | §6 |
| license | SPDX license identifier (e.g., MPL-2.0). Required. | §6 |
| authors | Array of inline tables with name (required) and optional email, url. At least one author required. | §6 |
| verifies | Relative path to the eval suite, typically "./evals.toml". Required for D1 or higher. | §6 |
| homepage | URL of a project homepage. Optional. | §6 |
| repository | URL of the module's source repository. Optional. | §6 |
| keywords | Array of strings for indexing. Optional. | §6 |
| depends_on | Array of dependency declarations. See §11. Optional. | §11.2 |
| extends | Single-parent extension declaration; inherits the parent's evals and takes precedence on conflicts. Optional. | §11.6 |
| contributes_only | Array restricting an eval pack to specified categories (adversarial, generator_adversary). Optional. | §11.7 |
Dependency declaration fields
| field | description | defined |
| spec | Git-rooted path of the form host/owner/repo/path. Required. | §11.2 |
| version | Constraint matching a Git tag in the dependency repository. One of version / ref / commit required. | §11.3 |
| ref | Git ref name (branch, tag, or commit reference). | §11.2 |
| commit | Exact Git commit SHA. | §11.2 |
| overrides | Subtable resolving merge conflicts for this dependency. Values may be literals or "ignore". | §11.5 |
Tagged sections · commonsformat.md
| tag | description | defined |
| <intent> | What the generated code should do. At most one per module. | §5.3.2 |
| <constraints> | Hard requirements the implementation must satisfy. Each line name: description. At most one per module. | §5.3.2 / §9.7 |
| <avoid> | Anti-patterns that disqualify an implementation. At most one per module. | §5.3.2 |
| <interface> | Declared shape of the public API. At most one per module. Only the root module's interface is exposed in a merged spec. | §5.3.2 |
| <threat-model> | Adversarial assumptions and attacker capabilities. At most one per module. | §5.3.2 |
| <example name="..."> | Illustrative input/output pair. May appear multiple times with distinct name attribute values. | §5.3.4 |
Data shape · schema.sql
| construct | description | defined |
| schema.sql | Optional file declaring data shape in a pinned SQLite-subset DDL. A shape commitment, not a storage commitment — runtime representation is the generator's choice. Omitted when a module has no shaped data. | §8 |
| header comment | Required leading comment stating the file is shape, not storage; must precede any DDL. | §8.2 |
| CREATE TABLE / CREATE INDEX | The only permitted statements. No views, triggers, procedural SQL, INSERT, or IF NOT EXISTS. | §8.1 |
| types | INTEGER, REAL, TEXT, BLOB, TIMESTAMP, BOOLEAN. No vendor extensions or affinity sugar. | §8.1 |
| constraints | PRIMARY KEY, FOREIGN KEY (same-file target only), NOT NULL, UNIQUE, CHECK, and literal column defaults. | §8.1 |
| composition | Schemas do not merge across modules; each module owns its tables. Cross-module foreign keys are rejected. | §8.3 |
Eval suite · evals.toml
| field | description | defined |
| commonsformat_evals | Eval format version. Required. | §9.1 |
| target | Name of the module these evals verify. Must match the module's name. | §9.1 |
| target_version | Version range these evals are valid for, in the §11.3 constraint syntax. | §9.1 |
| [[cases]] | Array of functional verification cases. | §9.4 |
| [[adversarial]] | Array of cases probing known attack patterns. Required for D1+. | §9.5 |
| [[generator_adversary]] | Array of cases probing generator-introduced flaws. Required for D2+. | §9.6 |
| [adversarial_coverage] | Declares which attack categories the eval suite covers. | §9.5 |
| [properties] | Invariants not easily expressed as input/output cases (advisory unless tooling verifies). | §9.8 |
Eval case fields
| field | description | defined |
| name | Kebab-case identifier, unique within the eval file. | §9.2 |
| description | One-sentence human-readable description. | §9.2 |
| category | One of: functional, input-validation, timing, concurrency, resource, failure-mode, crypto, interface. | §9.3 |
| input | Inline table of inputs to the implementation under test. | §9.2 |
| expect | Inline table of expectations on the implementation's behavior. | §9.2 |
| tags | Optional array of strings for filtering. | §9.2 |
| severity | Optional, one of info, warn, error, critical. Default error. | §9.2 |
| verifies | Optional array of constraint names from the prose (see §9.7). | §9.7 |
| fixture://... | String reference to a file or directory under fixtures/, used in input values. | §9.2.1 |
Eval categories
| category | execution model | defined |
| functional | Invokes implementation with input; checks structural-equality match of expect fields against output. | §9.3 |
| input-validation | Expects implementation to reject input; checks rejects = true and optional error_includes. | §9.3 |
| interface | Examines structural properties of the result (has_fields, has_field) — verifies shape, not content. | §9.3 |
| timing | Sets up a timed scenario; comparison operators in expect express bounds (e.g., max_x_lt = 200). | §9.3 |
| concurrency | Executes multiple operations genuinely concurrently and verifies properties of the combined result. | §9.3 |
| resource | Monitors memory, CPU, file handles, etc.; verifies bounds in expect. | §9.3 |
| failure-mode | Arranges a degraded condition (network down, disk full, dependency timeout) and verifies behavior. | §9.3 |
| crypto | Uses input as a test vector; compares output to expect byte-for-byte, constant-time when possible. | §9.3 |
Deployment tiers
| tier | requirement | defined |
| D0 | Personal / experimental. Functional eval cases pass. | §12 |
| D1 | Internal production. D0 plus adversarial cases pass; every declared constraint verified by at least one case. | §12 |
| D2 | Network-exposed production. D1 plus generator-adversary cases, two independent generator records, and recorded prose review. | §12 |
| D3 | Critical infrastructure. D2 plus two distinct runtime records and any declared formal_artifact_path referenced from the lockfile. | §12 |
Verification axes
| axis | levels | defined |
| eval depth | functional · adversarial · generator-adversary · formal | §13.1 |
| generator diversity | single · same-family · independent-family · independent-provider | §13.2 |
| runtime diversity | single · multi-runtime same family · multi-runtime independent family | §13.3 |
Lockfile · commonsformat.lock
| field | description | defined |
| commonsformat_lock | Lockfile format version. | §14 |
| generated_at | ISO 8601 timestamp of lockfile generation. | §14 |
| generator_tool | Identifier of the tool that produced the lockfile. | §14 |
| [root] | Root module identification, including spec_hash. | §14 |
| [[modules]] | Per-dependency record: spec, version, commit, checksum, fetched_at. | §14 |
| [modules.verification] | Per-dependency verification state: deployment_tier, results by generator, applied_packs. | §14 |
| [modules.verification.prose_review] | D2+ attestation: reviewed, reviewed_at, reviewer, prose_hash. | §14.1 |
describes commons format 0.2 reference · generated from release 0.2 · 2026-05-28