DRAFT: LISA ILE driver — portfolio policy + AV state/collapse safety (passes 4–5a, stacked on #99)

.travis/test-lisa.sh

@@ -18,4 +18,6 @@ fi
   MonteCarloMarginalizeCode/Code/test/test_lisa_synthetic_demo.py \
   MonteCarloMarginalizeCode/Code/test/test_lisa_fairdraw_weights.py \
   MonteCarloMarginalizeCode/Code/test/test_lisa_l0_rescue.py \
+  MonteCarloMarginalizeCode/Code/test/test_lisa_sampler_plumbing.py \
+  MonteCarloMarginalizeCode/Code/test/test_lisa_av_state.py \
   MonteCarloMarginalizeCode/Code/test/test_lisa_driver_drift.py

MonteCarloMarginalizeCode/Code/test/expensive_before_merging/integrators/audit_lisa_driver_drift.py

@@ -151,8 +151,17 @@ def compute_gap():
     main = collect(MAIN)
     lisa = collect(LISA)
     gap, extras = [], []
+    # A FUNC is satisfied by its BARE name as well as its qualified one.  The main driver has
+    # ONE analyze_event and nests helpers inside it; this driver has TWO (analyze_event_LISA
+    # and analyze_event), so a helper ported here must be hoisted to module level or else
+    # duplicated -- and duplicating is the failure mode this audit exists to prevent.  Without
+    # this, every correctly-hoisted port would sit in the gap forever as a false positive,
+    # which is how a gate gets trained out of people.
+    _lisa_bare = {n.rsplit(".", 1)[-1] for n in lisa["FUNC"]}
     for cat in ("FUNC", "OPTION", "CONST", "ATTR"):
         for name in sorted(set(main[cat]) - set(lisa[cat])):
+            if cat == "FUNC" and name.rsplit(".", 1)[-1] in _lisa_bare:
+                continue
             gap.append({"category": cat, "name": name,
                         "key": "%s:%s" % (cat, name), "main_line": main[cat][name]})
         for name in sorted(set(lisa[cat]) - set(main[cat])):

MonteCarloMarginalizeCode/Code/test/expensive_before_merging/integrators/lisa_drift_ledger.json

@@ -57,10 +57,6 @@
       "decision": "PORT",
       "reason": "Diagnostics for the replica triggers."
     },
-    "FUNC:analyze_event._reject_if_collapsed": {
-      "decision": "PORT",
-      "reason": "Implementation of --reject-collapsed-live-volume."
-    },
     "FUNC:dLofz": {
       "decision": "PHYSICS",
       "reason": "Cosmology helpers behind --d-prior-redshift. Same question: the interpolation range has to be re-chosen for MBHB redshifts."
@@ -311,68 +307,16 @@
     },
     "OPTION:--nf-flow-load": {
       "decision": "PORT",
-      "reason": "Normalizing-flow persistence. Neither driver lists an NF method in ok_lnL_methods (identical lists), so NF is reached only as a portfolio member -- equally available to LISA. Low priority, but not LISA-specific in any way."
+      "reason": "Normalizing-flow persistence is detector-agnostic, but the LISA portfolio factory currently constructs only AV, GMM, and adaptive_cartesian_gpu members. Port the NF member construction and route load/save to that member before exposing these flags; hooks on the portfolio aggregate are a silent no-op because it has no flow API."
     },
     "OPTION:--nf-flow-save": {
       "decision": "PORT",
-      "reason": "Normalizing-flow persistence. Neither driver lists an NF method in ok_lnL_methods (identical lists), so NF is reached only as a portfolio member -- equally available to LISA. Low priority, but not LISA-specific in any way."
-    },
-    "OPTION:--portfolio-adaptive-alloc": {
-      "decision": "PORT",
-      "reason": "mcsamplerPortfolio tuning. LISA wires the portfolio sampler and already accepts --sampler-portfolio-args (an eval-able dict), so these are reachable today via that escape hatch; porting them as first-class flags is pipeline parity, which is what the pipe actually passes. Low risk, no physics."
-    },
-    "OPTION:--portfolio-alloc-exponent": {
-      "decision": "PORT",
-      "reason": "mcsamplerPortfolio tuning. LISA wires the portfolio sampler and already accepts --sampler-portfolio-args (an eval-able dict), so these are reachable today via that escape hatch; porting them as first-class flags is pipeline parity, which is what the pipe actually passes. Low risk, no physics."
-    },
-    "OPTION:--portfolio-freeze-wt": {
-      "decision": "PORT",
-      "reason": "mcsamplerPortfolio tuning. LISA wires the portfolio sampler and already accepts --sampler-portfolio-args (an eval-able dict), so these are reachable today via that escape hatch; porting them as first-class flags is pipeline parity, which is what the pipe actually passes. Low risk, no physics."
-    },
-    "OPTION:--portfolio-grace-iters": {
-      "decision": "PORT",
-      "reason": "mcsamplerPortfolio tuning. LISA wires the portfolio sampler and already accepts --sampler-portfolio-args (an eval-able dict), so these are reachable today via that escape hatch; porting them as first-class flags is pipeline parity, which is what the pipe actually passes. Low risk, no physics."
-    },
-    "OPTION:--portfolio-probe-period": {
-      "decision": "PORT",
-      "reason": "mcsamplerPortfolio tuning. LISA wires the portfolio sampler and already accepts --sampler-portfolio-args (an eval-able dict), so these are reachable today via that escape hatch; porting them as first-class flags is pipeline parity, which is what the pipe actually passes. Low risk, no physics."
-    },
-    "OPTION:--portfolio-quality-signal": {
-      "decision": "PORT",
-      "reason": "mcsamplerPortfolio tuning. LISA wires the portfolio sampler and already accepts --sampler-portfolio-args (an eval-able dict), so these are reachable today via that escape hatch; porting them as first-class flags is pipeline parity, which is what the pipe actually passes. Low risk, no physics."
-    },
-    "OPTION:--portfolio-revive-period": {
-      "decision": "PORT",
-      "reason": "mcsamplerPortfolio tuning. LISA wires the portfolio sampler and already accepts --sampler-portfolio-args (an eval-able dict), so these are reachable today via that escape hatch; porting them as first-class flags is pipeline parity, which is what the pipe actually passes. Low risk, no physics."
-    },
-    "OPTION:--portfolio-varaha-can-freeze": {
-      "decision": "PORT",
-      "reason": "mcsamplerPortfolio tuning. LISA wires the portfolio sampler and already accepts --sampler-portfolio-args (an eval-able dict), so these are reachable today via that escape hatch; porting them as first-class flags is pipeline parity, which is what the pipe actually passes. Low risk, no physics."
-    },
-    "OPTION:--portfolio-varaha-max-frac": {
-      "decision": "PORT",
-      "reason": "mcsamplerPortfolio tuning. LISA wires the portfolio sampler and already accepts --sampler-portfolio-args (an eval-able dict), so these are reachable today via that escape hatch; porting them as first-class flags is pipeline parity, which is what the pipe actually passes. Low risk, no physics."
-    },
-    "OPTION:--portfolio-varaha-min-frac": {
-      "decision": "PORT",
-      "reason": "mcsamplerPortfolio tuning. LISA wires the portfolio sampler and already accepts --sampler-portfolio-args (an eval-able dict), so these are reachable today via that escape hatch; porting them as first-class flags is pipeline parity, which is what the pipe actually passes. Low risk, no physics."
-    },
-    "OPTION:--portfolio-varaha-never-freeze": {
-      "decision": "PORT",
-      "reason": "mcsamplerPortfolio tuning. LISA wires the portfolio sampler and already accepts --sampler-portfolio-args (an eval-able dict), so these are reachable today via that escape hatch; porting them as first-class flags is pipeline parity, which is what the pipe actually passes. Low risk, no physics."
-    },
-    "OPTION:--portfolio-weight-clip": {
-      "decision": "PORT",
-      "reason": "mcsamplerPortfolio tuning. LISA wires the portfolio sampler and already accepts --sampler-portfolio-args (an eval-able dict), so these are reachable today via that escape hatch; porting them as first-class flags is pipeline parity, which is what the pipe actually passes. Low risk, no physics."
+      "reason": "Normalizing-flow persistence is detector-agnostic, but the LISA portfolio factory currently constructs only AV, GMM, and adaptive_cartesian_gpu members. Port the NF member construction and route load/save to that member before exposing these flags; hooks on the portfolio aggregate are a silent no-op because it has no flow API."
     },
     "OPTION:--random-event": {
       "decision": "PORT",
       "reason": "Pick a random event from the input file. Detector-agnostic; flagged dangerous in its own help text for oversampling reasons that apply equally to LISA."
     },
-    "OPTION:--reject-collapsed-live-volume": {
-      "decision": "PORT",
-      "reason": "AV live-volume collapse rejection. AV is wired in the LISA driver identically."
-    },
     "OPTION:--rotation-n-harmonics": {
       "decision": "NA",
       "reason": "Sidereal time-dependence of an EARTH-BASED antenna pattern F(t). The LISA constellation's motion is already carried by the LISA response itself (factored_likelihood_LISA + the h5/TDI frames), so this correction is both unnecessary and wrong there -- it would apply Earth rotation to a heliocentric detector."
@@ -385,18 +329,6 @@
       "decision": "NA",
       "reason": "Sidereal time-dependence of an EARTH-BASED antenna pattern F(t). The LISA constellation's motion is already carried by the LISA response itself (factored_likelihood_LISA + the h5/TDI frames), so this correction is both unnecessary and wrong there -- it would apply Earth rotation to a heliocentric detector."
     },
-    "OPTION:--sampler-anisotropic-bins": {
-      "decision": "PORT",
-      "reason": "AV per-axis bin counts during contraction. AV is wired in LISA, and the argument for it is if anything stronger there: the LISA extrinsic axes are no more isotropic than the ground-based ones, and a sky pair that localizes tightly while distance stays broad is the exact case this exists for."
-    },
-    "OPTION:--sampler-load-state": {
-      "decision": "PORT",
-      "reason": "AV live-volume state serialization. AV is wired in LISA; the state is the sampler's own internal grid, so it carries no LIGO-specific convention."
-    },
-    "OPTION:--sampler-save-state": {
-      "decision": "PORT",
-      "reason": "AV live-volume state serialization. AV is wired in LISA; the state is the sampler's own internal grid, so it carries no LIGO-specific convention."
-    },
     "OPTION:--sampler-sequential-warmstart": {
       "decision": "PORT",
       "reason": "Warm-start each intrinsic point from the previous one's cloud. Applies whenever --n-events-to-analyze>1, which LISA supports. Its snapshot/restore prerequisites (Finding 5) already landed with the L0 rescue, so this is now capture + the event-loop wiring only."

MonteCarloMarginalizeCode/Code/test/expensive_before_merging/integrators/make_lisa_drift_ledger.py

@@ -112,23 +112,28 @@
      "with that pass rather than with the sequential warm start it is named for."),
 
     # --------------------------------------------------------------- L0 rescue / warm start
-    (r"^OPTION:--reject-collapsed-live-volume$", "PORT",
-     "AV live-volume collapse rejection. AV is wired in the LISA driver identically."),
-    (r"^FUNC:analyze_event\._reject_if_collapsed$", "PORT",
-     "Implementation of --reject-collapsed-live-volume."),
+    (r"^OPTION:--reject-collapsed-live-volume$", "PORTED",
+     "AV live-volume collapse rejection. AV is wired in the LISA driver identically. "
+     "NOTE the main driver calls its gate TWICE -- first run and replica pool -- and only "
+     "the first call exists here, because there is no pooling yet; the second MUST be "
+     "added with --mc-error-replicas or the flag is bypassed for the case pooling creates."),
+    (r"^FUNC:analyze_event\._reject_if_collapsed$", "PORTED",
+     "Hoisted to module level rather than nested, because this driver has TWO "
+     "analyze_event variants. The audit matches FUNC items on the bare name for exactly "
+     "this reason."),
     (r"^OPTION:--sampler-sequential-warmstart$", "PORT",
      "Warm-start each intrinsic point from the previous one's cloud. Applies whenever "
      "--n-events-to-analyze>1, which LISA supports. Its snapshot/restore prerequisites "
      "(Finding 5) already landed with the L0 rescue, so this is now capture + the "
      "event-loop wiring only."),
     (r"^OPTION:--sampler-sequential-warmstart-cover-frac$", "PORT",
      "Coverage floor for the above; meaningless without it, so they travel together."),
-    (r"^OPTION:--sampler-anisotropic-bins$", "PORT",
+    (r"^OPTION:--sampler-anisotropic-bins$", "PORTED",
      "AV per-axis bin counts during contraction. AV is wired in LISA, and the argument "
      "for it is if anything stronger there: the LISA extrinsic axes are no more "
      "isotropic than the ground-based ones, and a sky pair that localizes tightly "
      "while distance stays broad is the exact case this exists for."),
-    (r"^OPTION:--sampler-(save|load)-state$", "PORT",
+    (r"^OPTION:--sampler-(save|load)-state$", "PORTED",
      "AV live-volume state serialization. AV is wired in LISA; the state is the "
      "sampler's own internal grid, so it carries no LIGO-specific convention."),
     (r"^OPTION:--sampler-warmstart-(cover-frac|inflate)$", "PORT",
@@ -163,17 +168,18 @@
      "is the ecliptic pair -- the grouping still makes sense, the docstring does not."),
 
     # ------------------------------------------------------------------ portfolio plumbing
-    (r"^OPTION:--portfolio-", "PORT",
-     "mcsamplerPortfolio tuning. LISA wires the portfolio sampler and already accepts "
-     "--sampler-portfolio-args (an eval-able dict), so these are reachable today via "
-     "that escape hatch; porting them as first-class flags is pipeline parity, which is "
-     "what the pipe actually passes. Low risk, no physics."),
+    (r"^OPTION:--portfolio-", "PORTED",
+     "mcsamplerPortfolio freeze/allocation policy. Definitions copied verbatim and the "
+     "_freeze_policy_kwargs assembly is textually identical to the main driver's, so "
+     "unset options (None) stay out of the dict and the sampler keeps its own defaults. "
+     "--portfolio-varaha-can-freeze wins over --portfolio-varaha-never-freeze, as there."),
 
     # ------------------------------------------------------------------- NF flow plumbing
     (r"^OPTION:--nf-flow-(load|save)$", "PORT",
-     "Normalizing-flow persistence. Neither driver lists an NF method in "
-     "ok_lnL_methods (identical lists), so NF is reached only as a portfolio member -- "
-     "equally available to LISA. Low priority, but not LISA-specific in any way."),
+     "Normalizing-flow persistence is detector-agnostic, but the LISA portfolio factory "
+     "currently constructs only AV, GMM, and adaptive_cartesian_gpu members. Port the NF "
+     "member construction and route load/save to that member before exposing these flags; "
+     "hooks on the portfolio aggregate are a silent no-op because it has no flow API."),
 
     # --------------------------------------------------------- extrinsic proposal handoff
     (r"^OPTION:--extrinsic-proposal-output$", "PORT",