use std::fs;
use std::path::PathBuf;
use anyhow::Result;
use super::fixture::{load_suite, validate_suite, validate_suite_selection};
use super::runner::{run_memory_bench, MemoryBenchOptions};
use super::types::{
MemoryBenchCondition, ADVERSARIAL_POLICY_SUITE, DEFAULT_PUBLIC_ROOT, DEFAULT_SUITE,
};
#[test]
fn remem_code_memory_fixture_covers_required_categories() -> Result<()> {
let fixture = load_suite(DEFAULT_SUITE)?;
let categories = fixture
.tasks
.iter()
.map(|task| task.category.as_str())
.collect::<std::collections::BTreeSet<_>>();
for required in [
"temporal_as_of",
"stale_memory_avoidance",
"conflict_detection",
"workstream_continuity",
"prior_bug_root_cause",
"architecture_constraints",
"file_source_anchors",
"user_context_relevance",
] {
assert!(
categories.contains(required),
"missing required memory bench category {required}"
);
}
assert!(fixture.tasks.iter().all(|task| {
!task.gold_supporting_event_ids.is_empty()
&& task.gold_supporting_event_ids.iter().all(|id| {
task.evidence
.iter()
.any(|evidence| evidence.event_id == *id)
})
}));
Ok(())
}
#[test]
fn memory_bench_conditions_are_supported() {
for condition in MemoryBenchCondition::ALL {
assert_eq!(
MemoryBenchCondition::parse(condition.as_str()),
Some(condition)
);
}
assert_eq!(MemoryBenchCondition::parse("unknown"), None);
}
#[test]
fn memory_bench_fixture_allows_suite_to_differ_from_benchmark_id() -> Result<()> {
let mut fixture = load_suite(DEFAULT_SUITE)?;
fixture.benchmark_id = "independent-benchmark-id".to_string();
validate_suite(&fixture)?;
validate_suite_selection(&fixture, DEFAULT_SUITE)?;
assert_ne!(fixture.suite, fixture.benchmark_id);
Ok(())
}
#[test]
fn memory_bench_fixture_rejects_mismatched_requested_suite() -> Result<()> {
let fixture = load_suite(DEFAULT_SUITE)?;
assert!(validate_suite_selection(&fixture, "misrouted-suite")
.unwrap_err()
.to_string()
.contains("must match requested suite"));
Ok(())
}
#[test]
fn adversarial_policy_fixture_covers_required_categories() -> Result<()> {
let fixture = load_suite(ADVERSARIAL_POLICY_SUITE)?;
let categories = fixture
.tasks
.iter()
.map(|task| task.category.as_str())
.collect::<std::collections::BTreeSet<_>>();
for required in [
"secrets_api_keys",
"credentials",
"payments_accounts",
"unframed_third_party_personal_data",
"jokes_roleplay",
"negation",
"unsupported_assistant_claims",
"unapproved_external_source_claims",
"cross_sentence_splicing",
"same_name_repos",
"multi_task_sessions",
"branch_divergence",
"stale_file_anchors",
"conflicting_memories",
"instruction_injection",
"authority_claim",
"opaque_payload",
"benign_quoted_instruction",
] {
assert!(
categories.contains(required),
"missing required adversarial policy category {required}"
);
}
assert!(fixture.tasks.iter().any(|task| {
task.category == "approved_external_source_claims"
&& task
.policy
.as_ref()
.is_some_and(|policy| policy.explicit_approval)
}));
assert!(fixture.tasks.iter().all(|task| {
task.policy.as_ref().is_some_and(|policy| {
policy.explicit_approval
|| policy.poisoning_quarantine_expected
|| (policy.non_retention_required
&& policy.expected_active_claims == 0
&& policy.expected_candidates == 0
&& policy.expected_summary_inputs == 0
&& policy.expected_policy_abstention)
})
}));
for task in fixture.tasks.iter().filter(|task| {
task.policy
.as_ref()
.is_some_and(|policy| policy.poisoning_quarantine_expected)
}) {
assert!(
task.evidence.iter().any(|evidence| {
crate::memory::poisoning::scan_instruction_pattern(&evidence.content).is_some()
}),
"poisoning fixture {} must match the production pattern scanner",
task.id
);
let outcome = super::diagnostics::score_policy(
MemoryBenchCondition::RememDefault,
task,
&[],
task.abstention_allowed,
None,
);
assert_eq!(
outcome.active_claim_count, 0,
"poisoning fixture {} must never produce an active claim",
task.id
);
assert!(outcome.poisoning_scanner_matched);
assert_eq!(
outcome.policy_failure_count, 0,
"poisoning fixture {} must pass policy scoring",
task.id
);
}
Ok(())
}
#[tokio::test]
async fn remem_default_memory_bench_writes_verifiable_public_artifacts() -> Result<()> {
let root = unique_temp_dir("remem-memory-bench-public")?;
copy_dir_all(std::path::Path::new(DEFAULT_PUBLIC_ROOT), &root)?;
let report_path = root.join("memory/reports/remem-code-memory-v1.json");
let report = run_memory_bench(MemoryBenchOptions {
suite: DEFAULT_SUITE.to_string(),
condition: Some("remem_default".to_string()),
json_out: report_path.to_string_lossy().to_string(),
root: root.to_string_lossy().to_string(),
artifact_prefix: Some("memory/artifacts/remem-code-memory-v1".to_string()),
})
.await?;
assert_eq!(report.conditions, vec!["remem_default"]);
assert_eq!(report.run_artifacts.len(), 8);
let metrics = &report.aggregate_metrics;
assert_eq!(metrics["run_count"], 8);
assert_eq!(metrics["overall"]["tasks"], 8);
assert_eq!(metrics["overall"]["support_coverage"], 1.0);
assert!(metrics["by_category"]["prior_bug_root_cause"].is_object());
let verify = crate::eval::bench_artifact::verify_benchmark_artifacts(
crate::eval::bench_artifact::BenchVerifyOptions { root },
)?;
assert!(verify.passed, "{:#?}", verify.failures);
assert!(verify.run_artifacts_checked >= 10);
Ok(())
}
#[tokio::test]
async fn adversarial_policy_bench_reports_zero_policy_leaks() -> Result<()> {
let root = unique_temp_dir("remem-adversarial-policy-public")?;
copy_dir_all(std::path::Path::new(DEFAULT_PUBLIC_ROOT), &root)?;
let report_path = root.join("memory/reports/adversarial-policy-v2.json");
let report = run_memory_bench(MemoryBenchOptions {
suite: ADVERSARIAL_POLICY_SUITE.to_string(),
condition: Some("remem_default".to_string()),
json_out: report_path.to_string_lossy().to_string(),
root: root.to_string_lossy().to_string(),
artifact_prefix: Some("memory/artifacts/adversarial-policy-v2".to_string()),
})
.await?;
assert_eq!(report.conditions, vec!["remem_default"]);
assert_eq!(report.benchmark_version, "v2");
assert_eq!(report.run_artifacts.len(), 20);
for run_path in &report.run_artifacts {
let run: serde_json::Value =
serde_json::from_str(&fs::read_to_string(root.join(run_path))?)?;
assert_eq!(
run["benchmark_version"], report.benchmark_version,
"generated run {run_path} must inherit the suite version"
);
}
let policy = &report.aggregate_metrics["policy"];
assert_eq!(policy["non_retention_leak_rate"], 0.0);
assert_eq!(policy["false_block_rate"], 0.0);
assert_eq!(policy["suppression_obeyed_rate"], 1.0);
assert_eq!(policy["sensitive_restricted_default_exclusion_rate"], 1.0);
assert_eq!(policy["policy_abstention_accuracy"], 1.0);
assert_eq!(policy["policy_failure_rate"], 0.0);
assert_eq!(
report.aggregate_metrics["verification_paths"],
serde_json::json!(["capture_observation_candidate_promotion"])
);
assert_eq!(
report.aggregate_metrics["measurement_sources"],
serde_json::json!(["sqlite_production_tables"])
);
let approved_path = report
.run_artifacts
.iter()
.find(|path| path.contains("remem_default-approved-external-source-001/"))
.expect("approved production-path artifact");
let approved: serde_json::Value =
serde_json::from_str(&fs::read_to_string(root.join(approved_path))?)?;
assert_eq!(approved["metrics"]["policy"]["active_claim_count"], 1);
assert_eq!(approved["metrics"]["policy"]["candidate_count"], 0);
assert_eq!(approved["metrics"]["policy"]["summary_input_count"], 1);
assert_eq!(
approved["metrics"]["policy"]["verification_path"],
"capture_observation_candidate_promotion"
);
let opaque_path = report
.run_artifacts
.iter()
.find(|path| path.contains("opaque-payload-001"))
.expect("opaque production-path artifact");
let opaque: serde_json::Value =
serde_json::from_str(&fs::read_to_string(root.join(opaque_path))?)?;
assert_eq!(
opaque["metrics"]["policy"]["poisoning_source_scanner_matched"],
false
);
assert_eq!(
opaque["metrics"]["policy"]["poisoning_generated_surface_blocked"],
true
);
assert_eq!(opaque["metrics"]["policy"]["policy_failure_count"], 0);
let verify = crate::eval::bench_artifact::verify_benchmark_artifacts(
crate::eval::bench_artifact::BenchVerifyOptions { root },
)?;
assert!(verify.passed, "{:#?}", verify.failures);
assert!(verify.run_artifacts_checked >= 25);
Ok(())
}
#[tokio::test]
async fn write_vs_retrieval_report_includes_diagnostics_baselines_and_performance() -> Result<()> {
let root = unique_temp_dir("remem-write-vs-retrieval-public")?;
copy_dir_all(std::path::Path::new(DEFAULT_PUBLIC_ROOT), &root)?;
let report_path = root.join("memory/reports/write-vs-retrieval.json");
let report = run_memory_bench(MemoryBenchOptions {
suite: DEFAULT_SUITE.to_string(),
condition: None,
json_out: report_path.to_string_lossy().to_string(),
root: root.to_string_lossy().to_string(),
artifact_prefix: Some("memory/artifacts/write-vs-retrieval".to_string()),
})
.await?;
for condition in [
"truncated_full_context",
"oracle_evidence",
"complete_stored_memory",
"retrieved_memory",
"bm25_baseline",
"vector_baseline",
"hybrid_rag_baseline",
"summary_baseline",
] {
assert!(
report.conditions.iter().any(|item| item == condition),
"missing condition {condition}"
);
assert!(
report.aggregate_metrics["failure_decomposition"]["by_condition"][condition]
.is_object(),
"missing failure decomposition for {condition}"
);
assert!(
report.aggregate_metrics["performance"][condition].is_object(),
"missing performance metrics for {condition}"
);
}
assert_eq!(report.aggregate_metrics["run_count"], 80);
assert!(
report.aggregate_metrics["failure_decomposition"]["overall"]["retrieval_miss"]
.as_u64()
.is_some()
);
assert!(report.aggregate_metrics["performance"]["retrieved_memory"]
["retrieval_latency_p95_ms"]
.as_f64()
.is_some());
Ok(())
}
fn unique_temp_dir(prefix: &str) -> Result<PathBuf> {
let root = std::env::temp_dir().join(format!(
"{prefix}-{}-{}",
std::process::id(),
chrono::Utc::now().timestamp_nanos_opt().unwrap_or_default()
));
fs::create_dir_all(&root)?;
Ok(root)
}
fn copy_dir_all(src: &std::path::Path, dst: &std::path::Path) -> Result<()> {
fs::create_dir_all(dst)?;
for entry in fs::read_dir(src)? {
let entry = entry?;
let ty = entry.file_type()?;
let target = dst.join(entry.file_name());
if ty.is_dir() {
copy_dir_all(&entry.path(), &target)?;
} else {
fs::copy(entry.path(), target)?;
}
}
Ok(())
}