Sync current oshaughnessy-junior rift_O4d after #173 consolidation

.github/workflows/ci.yml

@@ -205,7 +205,170 @@ jobs:
         run: |
           python -m pytest -q \
             MonteCarloMarginalizeCode/Code/test/expensive_before_merging/integrators/test_probe_confirm.py \
-            MonteCarloMarginalizeCode/Code/test/test_cip_priors.py
+            MonteCarloMarginalizeCode/Code/test/test_cip_priors.py \
+            MonteCarloMarginalizeCode/Code/test/test_cosmology_single_source.py
+
+  q-window-stencil-check:
+    needs: install
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.10'
+          cache: 'pip'
+          cache-dependency-path: requirements.txt
+      - name: Enable symlink
+        run: sudo ln -sf $(which python3) /usr/bin/python
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip --break-system-packages
+          python -m pip install -r requirements.txt --break-system-packages
+          python -m pip install coverage pytest --break-system-packages
+          python -m pip install --editable . --break-system-packages
+      - name: Run Q_lm sub-sample stencil accuracy and selection gates
+        # numpy-only, runs in seconds, but it guards a CORE LIKELIHOOD choice that fails
+        # SILENTLY: picking the wrong sub-sample stencil raises nothing, it just makes Q_lm(t)
+        # less accurate, which surfaces only as a slightly wrong likelihood surface.
+        #
+        # test_q_window_interp asserts the cubic/sinc crossover in BOTH directions on purpose.
+        # sinc winning everywhere would mean the Lanczos window had been widened until it was no
+        # longer a local stencil, so neither direction may be relaxed to make a change pass.
+        # test_time_interp_choice pins the pipeline thresholds inside the measured ambiguous
+        # band, and checks the decision uses the sampling rate the run is actually on.
+        #
+        # test_calmarg_stencil_gating runs its CPU arms without a GPU (its GPU arm is additive),
+        # so it belongs here: it is what stops cubic/sinc being routed to the fused calibration
+        # kernel, which is implemented for 'nearest' only.
+        #
+        # test_interpolate_time_cli runs the three scripts as real SUBPROCESSES (~30 s). That
+        # cost is the point: the unit tests exercise the resolver and the gate predicate, but
+        # neither can see whether the SCRIPTS are still wired to them. Reverting a parser to
+        # const=None, or deleting a script's resolver call, leaves every unit test green while
+        # restoring a bare flag that silently does nothing. All three mutations were checked to
+        # fail these tests before they landed.
+        #
+        # The GPU parity files (test_q_window_interp_gpu, test_noloop_gpu_stencils) are
+        # deliberately NOT here -- there is no GPU on these runners, and they would report as
+        # skipped.  They are run by hand on a GPU node; the numbers are in PR #97.
+        #
+        # test_slowrot_* files do NOT belong here.  They are gated by slowrot-check, whose
+        # manifest requires every test_slowrot_*.py in this directory to be listed or
+        # explicitly excluded; a copy in this job's list is invisible to that manifest and
+        # would simply run the file twice (issue #169).
+        run: |
+          python -m pytest -q \
+            MonteCarloMarginalizeCode/Code/RIFT/likelihood/test_q_window_interp.py \
+            MonteCarloMarginalizeCode/Code/RIFT/likelihood/test_time_interp_choice.py \
+            MonteCarloMarginalizeCode/Code/RIFT/likelihood/test_calmarg_stencil_gating.py \
+            MonteCarloMarginalizeCode/Code/RIFT/misc/test_psd_bandwidth.py \
+            MonteCarloMarginalizeCode/Code/RIFT/likelihood/test_interpolate_time_cli.py
+
+  slowrot-check:
+    needs: install
+    runs-on: ubuntu-latest
+    # RIFT/likelihood/test_slowrot_*.py was run by NOTHING in this workflow until this
+    # job landed: `grep -rn slowrot .github/workflows/ci.yml` returned a single hit and
+    # it was a comment (issue #169).  That gap is worse than an ordinary one, because
+    # the two most recent changes to this code are changes whose DELIVERABLE IS THE
+    # GUARD -- #163 (the Nyquist derivative weight at both parities) and #165 (the
+    # Hermitian Nyquist response weight, which provably moves no number).  Neither
+    # leaves anything behind if its guard never runs.
+    #
+    # See .travis/test-slowrot.sh for why the gate counts tests and runs three files
+    # outside pytest: five test_slowrot_*.py files collect ZERO items and exit 5, "no
+    # tests ran", which reads as a pass, and three of those five carry asserts pytest
+    # cannot count.
+    #
+    # SEPARATE FROM jax-ile-check ON PURPOSE.  This suite is numpy + lal: no GPU, no
+    # jax, no numpyro.  Folding it into the jax gate would put a 1-2 minute numpy
+    # regression behind that job's jax/numpyro install and its ~5-15 minute run, and
+    # would couple a numpy failure's diagnosis to an unpinned upstream jax release.
+    # Python 3.10 to match the sibling numpy jobs, NOT the 3.11 jax-ile-check needs.
+    #
+    # timeout-minutes is a runaway backstop, not a budget: it is more than an order of
+    # magnitude above the observed job wall.  Measured cost, and the host-by-host numbers
+    # behind it, are in PR #172 (2026-08) -- deliberately not restated here, because a
+    # wall-clock table in a comment rots silently and nothing imports it.
+    timeout-minutes: 30
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.10'
+          cache: 'pip'
+          cache-dependency-path: requirements.txt
+      - name: Enable symlink
+        run: sudo ln -sf $(which python3) /usr/bin/python
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip --break-system-packages
+          python -m pip install -r requirements.txt --break-system-packages
+          python -m pip install coverage pytest --break-system-packages
+          python -m pip install --editable . --break-system-packages
+      - name: Run slow-rotation / finite-size CPU regression gate
+        env:
+          OMP_NUM_THREADS: 1
+        run: bash .travis/test-slowrot.sh
+
+  jax-ile-check:
+    needs: install
+    runs-on: ubuntu-latest
+    # test/jax/ was run by NOTHING in this workflow until this job landed (the file had
+    # zero matches for "jax"), and two real defects survived a month each behind that
+    # gap. See .travis/test-jax.sh for why the gate counts tests instead of just
+    # invoking pytest: most files in test/jax/ are __main__ scripts, and pytest exits 5
+    # ("no tests ran") on those -- a green tick over an empty run.
+    #
+    # Python 3.11 rather than the 3.10 used by the sibling jobs: current jax wheels
+    # require >=3.11. jax/numpyro are installed UNPINNED and deliberately so (CI should
+    # see what a user gets).  NOTE: RIFT.likelihood.jax_ile declares its dependencies
+    # NOWHERE in setup.py -- extras_require['jax-apps'] is for RIFT.interpolators.jax_gp
+    # and lists a different set -- so this job is their de-facto declaration.  An
+    # extras_require['jax-ile'] would be the right home for them
+    # and means a breaking upstream jax release can redden this job outside any PR's
+    # control. If that becomes noisy, pin here rather than deleting the job.
+    #
+    # Cost.  CURRENT (EXPECTED_TESTS=27 in .travis/test-jax.sh): 27 tests, measured
+    # 578-834 s of pytest across repeat runs on quiet CPU nodes (ldas-pcdev11/13, jax
+    # 0.9.2, JAX_PLATFORMS=cpu, JAX_ENABLE_X64=1, OMP_NUM_THREADS=1).
+    # test_jax_slowrot.py dominates (the p_max=0/p_max=1 rotation ladders and
+    # freqresponse, each followed by the AD/jit/vmap/hessian checks); it is the first
+    # thing to trim if CI minutes ever bite.  timeout-minutes is generous so a slower
+    # runner does not flake, but a hang still ends.
+    #
+    # HISTORICAL, and left here only as a runner-vs-local calibration: when this job
+    # collected 14 tests it measured 964 s wall locally and ran 14 passed in 285.96 s on
+    # the actual runner (job 95750869285, python 3.11.15, jax 0.10.2, numpyro 0.21.0) --
+    # ~10x headroom, with the unpinned install already a jax minor version ahead of the
+    # 0.9.2 measured locally.  Do NOT read the 14 as a current count; the file list has
+    # grown since and the gate asserts 27.
+    timeout-minutes: 60
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.11'
+          cache: 'pip'
+          cache-dependency-path: requirements.txt
+      - name: Enable symlink
+        run: sudo ln -sf $(which python3) /usr/bin/python
+      - name: Install dependencies and the CPU JAX stack
+        # ONE pip invocation on purpose: pip does not co-resolve across invocations, so
+        # installing jax separately lets a jax-driven numpy bump past numba's ceiling land
+        # as a dependency-conflict WARNING with exit 0, and surface later as a collection
+        # error (factored_likelihood imports numba) that looks like a RIFT bug.
+        # numpyro is needed by test_nuts_phimarg (the NUTS phase-marginalized sampler).
+        # flowMC is deliberately NOT installed -- see test-jax.sh for the exclusions.
+        run: |
+          python -m pip install --upgrade pip --break-system-packages
+          python -m pip install -r requirements.txt coverage pytest "jax[cpu]" numpyro --break-system-packages
+          python -m pip install --editable . --break-system-packages
+      - name: Run jax_ile CPU regression gate
+        env:
+          JAX_PLATFORMS: cpu
+          OMP_NUM_THREADS: 1
+        run: bash .travis/test-jax.sh
 
   lisa-check:
     needs: install
@@ -279,6 +442,8 @@ jobs:
         include:
           - asimov-series: '0.5'
             asimov-spec: 'asimov>=0.5,<0.6'
+          - asimov-series: '0.7'
+            asimov-spec: 'asimov>=0.7,<0.8'
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-python@v5
@@ -299,6 +464,9 @@ jobs:
           python -m pip install htcondor --only-binary=:all: --break-system-packages
       - name: Install Asimov
         run: python -m pip install '${{ matrix.asimov-spec }}' 'asimov-gwdata>=0.4,<0.5' --break-system-packages
+      - name: Install Asimov 0.7 pipeline plugins
+        if: matrix.asimov-series == '0.7'
+        run: python -m pip install 'asimov-bayeswave>=0.2,<0.3' 'pe-configurator>=1,<2' --break-system-packages
       - name: Run Asimov integration test
         run: bash .travis/test-asimov.sh
 
@@ -370,7 +538,8 @@ jobs:
             MonteCarloMarginalizeCode/Code/test/test_l0_rescue_seed.py \
             MonteCarloMarginalizeCode/Code/test/test_seq_warmstart_seed.py \
             MonteCarloMarginalizeCode/Code/test/test_fairdraw_double_weighting.py \
-            MonteCarloMarginalizeCode/Code/test/test_portfolio_fairdraw_backend.py
+            MonteCarloMarginalizeCode/Code/test/test_portfolio_fairdraw_backend.py \
+            MonteCarloMarginalizeCode/Code/test/test_rvs_record.py
       - name: Audit _rvs consumers against the fair-draw rebind
         # sampler._rvs is rebound to an EXPORT resample at the end of integrate_log, and five
         # separate defects have come from a consumer reading it afterwards as though it were
@@ -380,6 +549,15 @@ jobs:
         # whoever happens to edit the surrounding code.
         run: |
           python MonteCarloMarginalizeCode/Code/test/expensive_before_merging/integrators/audit_rvs_fairdraw.py --check
+      - name: Audit sampler backend contracts
+        # (DESIGN_rvs_naming.md)  The backends are structurally different in ways
+        # nothing states -- _rvs['integrand'] holds lnL on three of them, linear L on two, and
+        # EITHER on mcsamplerEnsemble depending on a kwarg -- and a consumer that guesses wrong
+        # gets a plausible number rather than an error.  This does not forbid the differences;
+        # it fails when one CHANGES without the recorded table changing with it, so the next
+        # developer meets a diff instead of a landmine.
+        run: |
+          python MonteCarloMarginalizeCode/Code/test/expensive_before_merging/integrators/audit_backend_contracts.py --check
       - name: Run test scripts
         run: |
           . .travis/test-coord.sh

.travis/test-asimov.sh

@@ -1,15 +1,14 @@
 #! /bin/bash
 set -euo pipefail
 
-# The RIFT Asimov plugin is currently developed and validated against the
-# Asimov 0.5 series.  This test skips cleanly for unsupported/future series
-# from inside pytest, so developers can preflight 0.6/0.7 environments without
-# editing the test.
+# The RIFT Asimov plugin is validated against the legacy 0.5 series and the
+# plugin-based 0.7 series. Unsupported API series skip cleanly in pytest.
 # Bootstrap-source selection ("scheduler: bootstrap file:") is driven against a stub
 # production rather than a project on disk, so it lives outside asimov_integration/.
 # It still needs asimov importable, and this is the only lane that installs it, so run
 # it here: otherwise the suite is never invoked and the behaviour can regress while the
 # required checks stay green.
 python -m pytest -q \
     MonteCarloMarginalizeCode/Code/test/asimov_integration \
+    MonteCarloMarginalizeCode/Code/test/test_asimov_compatibility.py \
     MonteCarloMarginalizeCode/Code/test/test_asimov_bootstrap_source.py