Struct AreevSubstrate 

pub struct AreevSubstrate { /* private fields */ }

An Areev-backed substrate that owns its facade (CLI path).

Implementations

impl AreevSubstrate

pub fn new(store: Areev, session_ns: Option<String>) -> Self

Wrap an open store, scoping the loop’s own writes to session_ns (defaults to the areev-loop namespace when None).

pub fn into_store(self) -> Areev

Recover the underlying store (e.g. to reuse it for another verb).

pub fn authz(&self) -> AuthzSet

The session’s rights, for the coverage filter (#312).

impl AreevSubstrate

pub fn facade(&self) -> &AreevFacade

Store-level joins the loop vocabulary doesn’t cover (the CLI’s §7.4 blast-radius/provenance reports need runs_touching). Read-only use.

Trait Implementations

impl OmsSubstrate for AreevSubstrate

fn put_grain(&mut self, spec: &GrainSpec) -> WResult<String>

Append a new grain; returns its content address.

fn supersede( &mut self, target_hash: &str, spec: &GrainSpec, _j: &str, ) -> WResult<String>

Supersede target_hash with a new grain carrying justification; returns the new grain’s address. Atomic and distinct from put (OMS §28.4).

fn retract(&mut self, hash: &str, _reason: &str) -> WResult<()>

Index-layer retraction (verification_status = retracted) — the inverse of an applied ADD, used by rollback. Not destructive (the grain stays content-addressed; only the index marks it retracted). The default reports it unsupported so substrates opt in.

fn execute_cal(&mut self, cal: &str) -> WResult<Vec<Value>>

Execute CAL text, returning result rows as JSON. Used to regenerate evidence sets (evidence_query) and to apply proposal_cal. A substrate MAY reject CAL it cannot run with Error::CalUnsupported.

fn validate_cal(&self, cal: &str) -> WResult<()>

Validate a CAL batch without executing it (statement classification, destructive-op detection). Delegated to the substrate — the engine contains a CAL writer, never a parser.

fn definition_inverse(&self, statement: &str) -> WResult<Option<String>>

The CAL that would restore the CURRENT definition named by a DEFINE QUERY / DEFINE TEMPLATE statement — the rollback inverse, captured before the definition is replaced.

fn load_state(&self) -> WResult<Value>

Load the persisted loop state blob (config + watermarks/cooldowns). Returns Value::Null when nothing has been stored yet.

fn store_state(&mut self, state: &Value) -> WResult<()>

Persist the loop state blob (a file-truth, so it travels with the file on sync).

fn put_blob(&mut self, bytes: &[u8]) -> WResult<String>

Store an opaque blob (candidate tool CODE, evalset payloads) in the substrate’s CAS, returning its address. §7.4’s blob seam — CAPABILITY-GATED: the default refuses, so a loop can only carry code on substrates that explicitly opt in. Code enters the substrate only through this seam or an authored add — never from a git mirror.

fn get_blob(&mut self, address: &str) -> WResult<Vec<u8>>

Fetch a blob by the address put_blob returned. Same capability gate.

impl SubstrateRead for AreevSubstrate

fn capabilities(&self) -> Capabilities

Declared optional capabilities.

fn grains_of_type( &self, grain_type: &str, namespace: Option<&str>, opts: ReadOpts, ) -> WResult<Vec<GrainRecord>>

Curated read: all grains of one OMS type, optionally namespace-scoped and watermark/liveness filtered.

fn grain(&self, hash: &str) -> WResult<Option<GrainRecord>>

Fetch one grain by content address.

fn validate_plan(&self, workflow: &Value) -> WResult<()>

Structurally validate a candidate Workflow grain body — the same checks the runtime would run before executing it (unique and reachable nodes, conditions parse, every cycle bounded). The engine calls this before it will stamp a plan_revision as executable, so a proposal that would produce an unrunnable plan never reaches a reviewer as something they could apply.

fn tool_evalset(&self, tool: &str) -> WResult<Option<String>>

The evalset a code revision of tool must be gated against (Rule E1’s pin). Returns Ok(None) when the tool declares none, which makes a code_revision for it advisory — an unpinnable revision is one no gating run could ever satisfy.

fn embed(&self, text: &str) -> WResult<Option<Vec<f32>>>

Embed one text through the substrate’s installed embedder — the T1 leg of “is this the same instruction in different words”. Ok(None) when no embedder is installed (the embeddings capability is off), so the caller falls back to a lexical measure rather than guessing. The default reports none; substrates opt in.

fn address_of(&self, spec: &GrainSpec) -> WResult<Option<String>>

The content address put_grain(spec) WOULD assign, computed without writing — what lets a rehearsal name the exact grain a live pass would have stored. Ok(None) when the substrate cannot say (the default); a replay then reports findings by dedup key and summary.

fn plan_replay( &self, incumbent: &str, candidate: &Value, ) -> WResult<Option<Value>>

Rehearse a candidate Workflow body against the journaled runs of the live plan at incumbent_plan_hash — re-driven through the runtime’s pure scheduler with every effect answered from the journal, nothing dispatched, nothing written (areev run shadow --plan-file). The report is the runtime’s ShadowPlanReport as JSON; the engine reads totals.runs, no_worse, out_of_support_fraction and the per-run rows. Ok(None) when the substrate has no runtime or no journaled runs of that plan — the proposal is then simply unrehearsed, never refused for it. The default reports none.

fn heads(&self, namespace: Option<&str>) -> WResult<Vec<HeadGroup>>

Entities with more than one live head. Requires the forks capability; the default impl reports it missing so non-fork substrates degrade cleanly rather than pretend.

fn telemetry(&self, namespace: Option<&str>) -> WResult<Option<TelemetryView>>

A snapshot of the recall-telemetry rollups (§8). Requires the telemetry capability; the default returns None so substrates without a sidecar degrade cleanly. namespace scopes the snapshot when set.