feat(membership): refuse episode membership outside the episode

ARCHITECTURE.md

@@ -71,6 +71,8 @@ boundaries above remain the target modular MSA architecture.
 | `location_membership` | location is not entity identity and not a language channel |
 | `validation_core` | RMSE, bias, coverage, graph, Monte Carlo, and exact-head claim-promotion metrics |
 | `tepp_api` | versioned DTO, schema, and export contracts |
+| `episode_membership` | event-time episode membership containment gate |
+| `location_membership` | location is not entity identity and not a language channel |
 | `prompt_source` | prompt boilerplate is not unique latent content and not stopword deletion |
 | `corpus_background` | corpus-background wording is not unique latent content and not stopword deletion |
 | `modality_source` | non-lexical modality is not unique latent content and not stopword deletion |

CHANGELOG.md

@@ -65,6 +65,9 @@ All notable changes to TEPP are documented here. The format follows Keep a Chang
 - `psychometric_core` posterior-aware structural input gates: construct classification, refusal of raw-proportion Pearson/OLS, explicit ALR-versus-ILR geometry boundaries, CPU `f64` OLS recovery, posterior-draw loading point-estimate averaging without Rubin uncertainty claims, invariance-gated latent-mean comparison, and causal-heuristic refusal (ADR 0005 first production slice; no new migration).
 ### Added
 
+- `episode_membership` identity gate: a document's episode membership cannot
+  start before or end after the episode event-time interval; recovered
+  containment flags are compared with known truth (ADR 0003).
 - `tepp_simulation` cutoff eligibility: delayed-reporting documents are excluded when `available_time` exceeds the knowledge cutoff, with recovered eligible counts matching known truth.
 - `topic_measurement` bounded deterministic CPU `f64` TRSL-TM reference estimator: canonical CSR/CSC inputs, cutoff-safe documents, standardized event time, weighted multiple memberships, prevalence covariates, explicit predecessor/successor regularization, multi-seed generalized EM, diagonal Laplace uncertainty, and fitted topic-lineage counts with known-truth RMSE plus exact line/branch coverage (ADR 0012; no persistence or accelerated-backend claim).
 - `topic_measurement` logistic-normal additive log-ratio and sequential Egozcue isometric log-ratio coordinates: fail-closed simplex validation, max-shifted stable ALR/ILR inverses with true-parameter RMSE, pairwise CLR Aitchison distance recovered by ILR Euclidean isometry for valid composition pairs, and refusal of TF-IDF/BM25/keyword scores as inferential topic coordinates (ADR 0012 first production slice; no new migration).

Cargo.toml

@@ -50,6 +50,7 @@ members = [
     "crates/interpretation_gateway",
     "crates/model_selection",
     "crates/checkpoint_authority",
+    "crates/episode_membership",
     "crates/compute_backend",
     "crates/episode_membership",
     "crates/membership_target",
@@ -107,6 +108,7 @@ default-members = [
     "crates/interpretation_gateway",
     "crates/model_selection",
     "crates/checkpoint_authority",
+    "crates/episode_membership",
     "crates/compute_backend",
     "crates/episode_membership",
     "crates/membership_target",

README.md

@@ -34,11 +34,29 @@ evidence, six-clock temporal values, event mentions/instances, relations,
 membership, persistence, splits, simulation, validation, API DTOs, and the
 predicted-versus-observed promotion gate.
 This branch establishes the Rust workspace, quality-gate foundation, and the
-longitudinal within/between decomposition capability. The eleven bounded crates
+longitudinal within/between decomposition capability. The workspace bounded crates
 compile independently. `longitudinal_core` exposes within/between decomposition
 and component RMSE APIs; the remaining crates expose no placeholder production
 APIs, and domain behavior for them begins in Task 2 with immutable evidence
 identifiers and source records.
+This branch establishes the Task 1 Rust workspace and quality-gate foundation.
+The workspace bounded crates compile independently but intentionally expose no
+The workspace bounded crates compile independently; Task 1 includes the
+The twelve bounded crates compile independently but intentionally expose no
+The eleven bounded crates compile independently; Task 1 includes the
+implemented `encrypted_mapping` crate with AES-256-GCM sealing and
+purpose-bound opening, while the remaining domain behavior begins in Task 2
+with immutable evidence identifiers and source records.
+The workspace bounded crates compile independently. `derived_sensitivity` inherits
+source Restricted/Internal classes onto topic, factor, and relation artifacts
+and fails closed on unknown kinds; derivation and blanket PII masking are not
+declassification. Other crates still begin domain behavior in Task 2 with
+immutable evidence identifiers and source records.
+The workspace bounded crates compile independently but intentionally expose no
+
+The eleven bounded crates compile independently but intentionally expose no
+placeholder production APIs. Domain behavior begins in Task 2 with immutable
+evidence identifiers and source records.
 This branch establishes the Rust workspace and quality-gate foundation. The
 bounded crates compile independently. Domain crates expose only validated
 production APIs; placeholder surfaces are prohibited.
@@ -78,6 +96,12 @@ crates/operational_log
 crates/outcome_order
 crates/payload_bound
 crates/persistence_postgres
+crates/corpus_split
+crates/tepp_simulation
+crates/validation_core
+crates/tepp_api
+crates/episode_membership
+crates/location_membership
 crates/prediction_contradiction
 crates/prompt_source
 crates/provider_receipt

crates/episode_membership/src/window.rs

@@ -72,6 +72,12 @@ pub fn identity_recovery_rate(
     if truth.is_empty() || truth.len() != decided.len() {
         return Err(EpisodeMembershipError::InvalidEpisodePayload);
     }
+    let matches = truth
+        .iter()
+        .zip(decided)
+        .filter(|(truth_flag, decided_flag)| truth_flag == decided_flag)
+        .count();
+    Ok(matches as f64 / truth.len() as f64)
     let matches = count_matching_decisions(truth.iter().copied().zip(decided.iter().copied()));
     Ok(recovery_rate_from_tally(matches, truth.len()))
 }