feat(campaign,registry,engine): record widenings + preflight v2 fault set (0108 tasks 1-2)

Task 1: RBootstrap and Generalization gain serde derives;
lineage.rs gains StageBootstrap (per_survivor | pooled_oos) and
CampaignGeneralization; StageRealization.bootstrap and
CampaignRunRecord.generalizations land serde-default sparse (a
pre-0108 campaign_runs.jsonl line still parses — pinned), threaded
None/empty through every literal site. No behaviour change.

Task 2: ExecFault::UnsupportedStage is gone; the preflight now admits
the v2 shape sweep [gate]* [wf]? [mc]? [generalize]? via a monotone
rank walk (precise per-pair PipelineShape prose) and statically
refuses: single-instrument generalize (GeneralizeNeedsInstruments),
non-R generalize metrics via check_r_metric (GeneralizeNonRMetric —
no fourth roster site), zero mc params (ZeroBootstrapParam). CLI
prose swapped accordingly. The loop's E2E phase added a tier-boundary
pin: process validate accepts [sweep, mc, walk_forward] while campaign
run refuses it data-free.

KNOWN RED (documented in the plan, flips in task 5):
crates/aura-cli/tests/research_docs.rs::campaign_run_v1_boundary_refuses_mc_process
still pins the retired 'not executable in v1' prose. Interim gates:
registry 57/0, campaign 33/0, cli --bin 111/0, workspace build clean.

refs #200
This commit is contained in:
2026-07-04 00:38:25 +02:00
parent 3454459e6d
commit ab778f2361
8 changed files with 607 additions and 60 deletions
+46 -4
View File
@@ -44,10 +44,6 @@ pub(crate) fn is_content_id(s: &str) -> bool {
/// (stage index + block id), the `doc_fault_prose`/`ref_fault_prose` register.
fn exec_fault_prose(f: &ExecFault) -> String {
match f {
ExecFault::UnsupportedStage { stage, block } => format!(
"process stage {stage} ({block}) is not executable in v1; executable \
pipeline shape: std::sweep [std::gate]* [std::walk_forward]"
),
ExecFault::PipelineShape { detail } => {
format!("process pipeline is not executable: {detail}")
}
@@ -68,6 +64,16 @@ fn exec_fault_prose(f: &ExecFault) -> String {
ExecFault::DeflatePlateauConflict { stage } => format!(
"process stage {stage}: sweep deflate: true composes only with select \"argmax\""
),
ExecFault::GeneralizeNeedsInstruments { available } => format!(
"campaign has {available} instrument(s); std::generalize needs at least 2"
),
ExecFault::GeneralizeNonRMetric { metric } => format!(
"std::generalize metric \"{metric}\" is not an R metric (gross/net R \
families generalize; pip metrics do not)"
),
ExecFault::ZeroBootstrapParam { stage, field } => format!(
"process stage {stage}: monte_carlo {field} must be > 0"
),
ExecFault::Window { stage, detail } => format!(
"process stage {stage}: walk_forward windows do not fit the campaign \
window: {detail}"
@@ -484,6 +490,42 @@ mod tests {
);
}
#[test]
/// generalize's static instrument-arity refusal is Debug-free and
/// names the available count plus the required floor.
fn generalize_needs_instruments_prose_names_the_shortfall() {
let prose = exec_fault_prose(&ExecFault::GeneralizeNeedsInstruments { available: 1 });
assert_eq!(prose, "campaign has 1 instrument(s); std::generalize needs at least 2");
assert!(!prose.contains("GeneralizeNeedsInstruments"), "Debug leak: {prose}");
}
#[test]
/// generalize's R-metric refusal names the metric and the R-only
/// rationale (pip metrics are not comparable across instruments).
fn generalize_non_r_metric_prose_names_the_metric() {
let prose = exec_fault_prose(&ExecFault::GeneralizeNonRMetric {
metric: "total_pips".into(),
});
assert_eq!(
prose,
"std::generalize metric \"total_pips\" is not an R metric (gross/net R \
families generalize; pip metrics do not)"
);
assert!(!prose.contains("GeneralizeNonRMetric"), "Debug leak: {prose}");
}
#[test]
/// the zero-bootstrap-param refusal is path-addressed (stage index)
/// and names the offending field.
fn zero_bootstrap_param_prose_names_stage_and_field() {
let prose = exec_fault_prose(&ExecFault::ZeroBootstrapParam {
stage: 3,
field: "block_len",
});
assert_eq!(prose, "process stage 3: monte_carlo block_len must be > 0");
assert!(!prose.contains("ZeroBootstrapParam"), "Debug leak: {prose}");
}
#[test]
/// The only shape treated as a direct store address is a bare 64-char
/// lowercase-hex token; anything else (wrong length, uppercase, non-hex)
+147
View File
@@ -794,6 +794,153 @@ fn campaign_run_v1_boundary_refuses_mc_process() {
assert!(!out.contains("UnsupportedStage"), "Debug leak: {out}");
}
/// A v2 process doc with `std::walk_forward` following the annotator
/// `std::monte_carlo` — intrinsically valid (order across population vs.
/// annotator stages is not a document-tier concern), refused only by the
/// executor's v2 preflight shape guard.
const WF_AFTER_MC_PROCESS_DOC: &str = r#"{
"format_version": 1,
"kind": "process",
"name": "wf-after-mc",
"pipeline": [
{ "block": "std::sweep", "metric": "sqn_normalized", "select": "argmax" },
{ "block": "std::monte_carlo", "resamples": 100, "block_len": 5 },
{ "block": "std::walk_forward", "in_sample_ms": 1209600000, "out_of_sample_ms": 604800000,
"step_ms": 604800000, "mode": "rolling", "metric": "net_expectancy_r", "select": "argmax" }
]
}"#;
/// Property (0108 task 2, mirroring aura-research's characterization pin
/// `validate_process_permits_walk_forward_after_an_annotator`): the
/// intrinsic/executor tier boundary is observable across TWO CLI commands.
/// `process validate` accepts `[sweep, monte_carlo, walk_forward]` (exit 0 —
/// the document tier does not order population vs. annotator stages) while
/// `campaign run`'s v2 preflight refuses the very same document, before any
/// member executes (data-free: the [1, 2] 1970-epoch-ms window is never
/// reached), because `std::walk_forward` cannot follow the annotator
/// `std::monte_carlo`. A regression that let the executor silently accept the
/// intrinsic tier's looser order (or that widened the intrinsic tier to
/// match) would collapse this two-command divergence.
#[test]
fn process_validate_permits_wf_after_mc_but_campaign_run_refuses_it() {
// intrinsic tier: no project needed, and it accepts the doc outright.
let vdir = temp_cwd("wf-after-mc-validate");
write_doc(&vdir, "wfmc.process.json", WF_AFTER_MC_PROCESS_DOC);
let (vout, vcode) = run_code_in(&vdir, &["process", "validate", "wfmc.process.json"]);
assert_eq!(vcode, Some(0), "intrinsic validate must accept sweep->mc->walk_forward: {vout}");
assert!(
vout.contains("process document valid (intrinsic): 3 pipeline blocks"),
"stdout/stderr: {vout}"
);
// executor tier: campaign run's v2 preflight refuses the same shape.
let _fixture = project_lock();
let dir = built_project();
let runs_dir = dir.join("runs");
std::fs::remove_dir_all(&runs_dir).ok();
let _cleanup = ScratchGuard(vec![
ScratchPath::Dir(runs_dir.clone()),
ScratchPath::File(dir.join("wfmc.process.json")),
ScratchPath::File(dir.join("wfmc.campaign.json")),
]);
let bp_id = seed_blueprint(dir, "campaign-run-wfmc-seed");
let proc_id = register_process_doc(dir, "wfmc.process.json", WF_AFTER_MC_PROCESS_DOC);
write_doc(dir, "wfmc.campaign.json", &campaign_doc_json(&bp_id, &proc_id, (1, 2), "", ""));
let (out, code) = run_code_in(dir, &["campaign", "run", "wfmc.campaign.json"]);
assert_eq!(code, Some(1), "stdout/stderr: {out}");
assert!(
out.contains("std::walk_forward cannot follow std::monte_carlo"),
"stdout/stderr: {out}"
);
assert!(!out.contains("PipelineShape {"), "Debug leak: {out}");
}
/// A v2 process doc with a static `std::monte_carlo resamples: 0` defect.
const ZERO_RESAMPLES_MC_PROCESS_DOC: &str = r#"{
"format_version": 1,
"kind": "process",
"name": "zero-resamples-mc",
"pipeline": [
{ "block": "std::sweep", "metric": "sqn_normalized", "select": "argmax" },
{ "block": "std::monte_carlo", "resamples": 0, "block_len": 5 }
]
}"#;
/// Property (0108 task 2): `std::monte_carlo`'s `resamples > 0` guard is a
/// STATIC executor-preflight refusal, wired end to end from the on-disk
/// document through to CLI stdout, firing before any member runs (data-free:
/// the [1, 2] 1970-epoch-ms window is never reached) and naming the stage
/// index and the offending field. A regression that dropped the static guard
/// (deferring to the engine's runtime all-zero-resamples degenerate instead)
/// would let this refusal disappear or drift to a different message.
#[test]
fn campaign_run_refuses_zero_resamples_monte_carlo_before_any_member_runs() {
let _fixture = project_lock();
let dir = built_project();
let runs_dir = dir.join("runs");
std::fs::remove_dir_all(&runs_dir).ok();
let _cleanup = ScratchGuard(vec![
ScratchPath::Dir(runs_dir.clone()),
ScratchPath::File(dir.join("zeromc.process.json")),
ScratchPath::File(dir.join("zeromc.campaign.json")),
]);
let bp_id = seed_blueprint(dir, "campaign-run-zeromc-seed");
let proc_id = register_process_doc(dir, "zeromc.process.json", ZERO_RESAMPLES_MC_PROCESS_DOC);
write_doc(dir, "zeromc.campaign.json", &campaign_doc_json(&bp_id, &proc_id, (1, 2), "", ""));
let (out, code) = run_code_in(dir, &["campaign", "run", "zeromc.campaign.json"]);
assert_eq!(code, Some(1), "stdout/stderr: {out}");
assert!(
out.contains("process stage 1: monte_carlo resamples must be > 0"),
"stdout/stderr: {out}"
);
assert!(!out.contains("ZeroBootstrapParam"), "Debug leak: {out}");
}
/// A v2 process doc that ends in `std::generalize`.
const GENERALIZE_PROCESS_DOC: &str = r#"{
"format_version": 1,
"kind": "process",
"name": "generalize-screen",
"pipeline": [
{ "block": "std::sweep", "metric": "sqn_normalized", "select": "argmax" },
{ "block": "std::generalize", "metric": "net_expectancy_r" }
]
}"#;
/// Property (0108 task 2): `std::generalize`'s campaign-level instrument-
/// arity guard (>= 2 instruments) is a STATIC executor-preflight refusal,
/// wired end to end from the campaign document's `data.instruments` through
/// to CLI stdout, firing before any member runs (data-free) and naming the
/// actual instrument count. `campaign_doc_json`'s fixture campaign carries
/// exactly one instrument (`GER40`), so this exercises the campaign
/// parameter `preflight` newly reads (previously unused in v1) rather than
/// anything reachable from the process document alone. A regression that
/// left the campaign parameter unread would silently accept a single-
/// instrument `std::generalize` and only fail later, opaquely, inside the
/// generalization math.
#[test]
fn campaign_run_refuses_single_instrument_generalize() {
let _fixture = project_lock();
let dir = built_project();
let runs_dir = dir.join("runs");
std::fs::remove_dir_all(&runs_dir).ok();
let _cleanup = ScratchGuard(vec![
ScratchPath::Dir(runs_dir.clone()),
ScratchPath::File(dir.join("gen.process.json")),
ScratchPath::File(dir.join("gen.campaign.json")),
]);
let bp_id = seed_blueprint(dir, "campaign-run-gen-seed");
let proc_id = register_process_doc(dir, "gen.process.json", GENERALIZE_PROCESS_DOC);
write_doc(dir, "gen.campaign.json", &campaign_doc_json(&bp_id, &proc_id, (1, 2), "", ""));
let (out, code) = run_code_in(dir, &["campaign", "run", "gen.campaign.json"]);
assert_eq!(code, Some(1), "stdout/stderr: {out}");
assert!(
out.contains("campaign has 1 instrument(s); std::generalize needs at least 2"),
"stdout/stderr: {out}"
);
assert!(!out.contains("GeneralizeNeedsInstruments"), "Debug leak: {out}");
}
/// Non-empty persist_taps defers LOUDLY, and the note's ORDER is pinned: it
/// prints before member execution, so it is asserted here on a run
/// that refuses at the member-data seam. The [1, 2] epoch-ms window (1970)