Consolidate rift_O4c + master + junior/rift_O4d into rift_O4d (and port the open O4c portfolio-driver fix)

[oshaughnessy-junior]

Consolidates rift_O4c (GitHub + GitLab), master, and oshaughnessy-junior/rift_O4d into rift_O4d, and ports the one open O4c fix that O4d still lacked.

What was actually diverged

git.ligo.org/rapidpe-rift/rift and oshaughn/research-projects-RIT are byte-identical on both rift_O4c and rift_O4d — zero commits either way. So there was no GitLab-only history to recover; everything merged there is already on GitHub.

The 29-commit rift_O4c gap is mostly an illusion. The container family, multi-GPU ILE fan-out, srate export fix, Virgo calibration convention and the getenv='*' default had already landed in O4d independently via the junior fork, in a more advanced form. Verified file-by-file rather than by patch-id: for container_manifest.py, test_container_manifest.py and the multi_gpu demo the O4d→O4c diff is pure deletion, zero insertions — O4d is a strict superset. dag_utils_generic.py is actually ahead, carrying an extra and not singularity_runtime_select term in the container-transfer guard that O4c lacks.

Source	Commits absent from rift_O4d	Genuinely new content
ligo/rift_O4c, ligo/rift_O4d	0	— (identical to GitHub)
origin/rift_O4c	29	1 feature (asimov bootstrap source)
origin/master	4	1 CI workflow
junior/rift_O4d	7	all (L0 rescue gate, review checklist)

Contents

1. junior/rift_O4d — L0 rescue reject-gate fix, portfolio warm-pass sample reserve, sampler/estimator pre-merge review checklist. Clean merge.
2. origin/rift_O4c 0.0.17.9→0.0.17.11 — the one new item is the asimov explicit bootstrap source (scheduler: bootstrap file:, rapidpe-rift/rift!53) plus its 204-line test. asimov/rift.py merged as a strict superset of both parents (zero deletions relative to either side): O4c's _resolve_bootstrap_file/_dataset_label and the broadened 'bootstrap upstream' or 'bootstrap file' guard on top of O4d's bootstrap-amplitude, grid-reuse safety, use global priors override and scheduler-environment passthrough.
3. origin/master — private review-dispatch workflow (Add private automated review dispatch for junior RIFT PRs #167). The evidence-calculation fix (BUG: make sure marginal log likelihood calculation uses all available values #156, C. Talbot) was already patch-equivalent in O4d.
4. Port of portfolio: make the driver fail loudly instead of silently substituting a sampler (driver-side complement to #168) #172 — portfolio driver fails loudly instead of silently substituting a sampler.

Conflict resolutions

Eight conflicts. Six were add/add on files proven to be O4d supersets, resolved to O4d. The two judgement calls, both kept on the O4d side because O4c would have regressed them:

• setup.py stays 0.0.18.0rc2. 0.0.17.x is the parallel O4c release line, not O4d's.
• calibration_reweighting.py keeps ast.literal_eval. O4c still has the bare eval there.

CHANGES.rst keeps both release histories, O4d first.

The #172 port

O4d already carried half of defect 4 — it splits --sampler-portfolio on commas, but with no None guard, no empty-member check and no else clause. That hunk conflicted; resolved by keeping O4d's comment and expression and adding the two missing guards. Defects 1, 2 and 3 were entirely absent from O4d and applied clean:

1. elif opts.sampler_method == "portfolio" and mcsampler_Portfolio_ok: made the raise on the next line dead code and sent an unavailable portfolio to the terminal else, which silently integrates with the plain mcsampler.MCSampler.
2. That terminal else dereferenced mcsamplerPortfolio.known_pipelines in its own diagnostic, so a failed import raised NameError from the error message on a torch-free container.
3. The plugin-pipeline elif had the same unguarded dereference.
4. An unrecognized member name left sampler bound to its previous value and appended it silently.

Testing

• Full CI test list (extracted from .github/workflows/ci.yml + .gitlab-ci.yml): 186 passed, 3 skipped.
• Merged-surface tests: 61 passed — including O4c's test_asimov_bootstrap_source.py and test_container_manifest.py running green against O4d's implementation, an independent confirmation of the superset resolutions.
• The hand-merged comma/None/empty parsing checked against 7 cases (None, [], ['AV,GMM'], ['AV','GMM'], mixed, whitespace, [',']).

Two collection errors in test/ (factored_likelihood_test.py, integrators/test_mcsampler_rosenbrock.py) are pre-existing legacy-script issues in files untouched here, and neither is in CI's list.

After this, all seven refs — origin/{master,rift_O4c,rift_O4d}, ligo/{master,rift_O4c,rift_O4d}, junior/rift_O4d — report zero commits missing from the branch.

Note on the other open O4c PRs

#168, #169 and #171 are O4d→O4c backports; their content is already in O4d. Only #172 was missing, and it is included here. Merging this does not close #172 — that PR targets rift_O4c and should still land there.

🤖 Generated with Claude Code

---

Update: also merged O4c 0.0.17.12, and paired with a fork-side PR

Paired PR: oshaughnessy-junior#91 targets oshaughnessy-junior:rift_O4d from the same branch and same commits, so the headline pair stays equivalent. Review changes pushed for one apply to both; they cannot resolve differently. Order does not matter.

Please merge both with a merge commit (or fast-forward), NOT squash. Squashing either side rewrites the commits on that side only, and the next origin <-> junior cross-merge would re-present all of this history as new work.

origin/rift_O4c advanced from 4a8703f3 to ddfc637f during review; that has been merged in too. New content is the log-uniform eccentricity prior (--eccentricity-prior log_uniform, rapidpe-rift/rift!54). O4d had no equivalent, so it is a genuine ingest. Same two resolutions as before (setup.py stays 0.0.18.0rc2; CHANGES.rst keeps both histories). All tests re-run after it: 186 passed / 3 skipped, plus 61 on the merged surface.

⚠ Known upstream defect carried across unmodified

log_eccentricity_prior() in util_ConstructIntrinsicPosterior_GenericCoordinates.py calls np.ln, which does not exist in numpy, so --eccentricity-prior log_uniform raises AttributeError as soon as the prior is evaluated. The normalization is independently wrong: a log-uniform density on [ECC_MIN, ECC_MAX] is 1/(x*log(ECC_MAX/ECC_MIN)), not 1/(x*log(ECC_MAX-ECC_MIN)).

This is upstream's bug on rift_O4c, not a merge artifact, and is deliberately not patched here — fixing it in the merge would fork the exact line this consolidation exists to keep collision-free. It should be fixed on rift_O4c, where it originated. No test on either branch exercises this prior, which is why it shipped.

---

Update: the eccentricity-prior defect is now FIXED here, not just flagged

The np.ln crash and the wrong normalization described above are fixed on this branch, together with the first test suite the CIP priors have ever had. Same commit is up as oshaughn#174 against rift_O4c, where the code originated, so the two lines converge on this function instead of diverging.

• log_eccentricity_prior now uses np.log(ECC_MAX/ECC_MIN). The old constant ln(ECC_MAX-ECC_MIN) is ln(0.399) = -0.918 for the shipped 0.001/0.4 defaults — negative, so the prior would have been negative everywhere had it evaluated at all.
• test_cip_priors.py extracts the shipped def blocks from CIP with ast rather than transcribing them (a copied reference implementation drifts and then tests nothing), so the functions under test are byte-identical to the ones CIP runs, with no argparse or I/O executed. All 38 priors must evaluate finite and non-negative; the 19 claiming a normalized density must integrate to 1 against an explicitly-stated measure (dx, d(ln x), d(x²)).
• Verified as a regression test: 4 failed / 56 passed against the unfixed CIP, all four naming log_eccentricity_prior; 60 passed with the fix. The extraction finds the same 38 priors in O4d's CIP as in O4c's (identical name sets).

Test totals for this branch are now 246 passed / 3 skipped on the CI list plus the new prior suite, and 61 on the merged surface.

Two things found while writing the test and deliberately left alone, flagged in #174 for a maintainer: s_component_volumetricprior integrates to R/9 rather than 1 (excluded as unnormalized; I have not verified whether that factor is intended), and util_ConstructIntrinsicPosterior_GenericCoordinates.py:1007 contains s_compone+nt_zprior_positive, a typo that raises NameError whenever --aligned-prior alignedspin-zprior-positive is used.

---

Update: brought current with O4c 0.0.17.13, junior/rift_O4d and origin/rift_O4d

Three more merges since the last update. The fork-side companion oshaughnessy-junior#91 is now MERGED — with a merge commit, not a squash, so the histories stay aligned and this PR does not re-present that work as new.

• junior/rift_O4d (30 commits) — merged Calmarg #91 plus the _rvs fair-draw audit, the TEOBResumSDALI global sign fix, the EOS double-log evidence fix, nal_io, the ensemble log-space contract fix, and the container-image selector truncation fix. Clean.
• origin/rift_O4d (4 commits) — plot_posterior_corner.py --disable-special-ranges (PR plot_posterior_corner.py: commented out special range for q, eta and … #159). Clean.
• origin/rift_O4c 0.0.17.13 (3 commits) — --use-eccentricity-ln / --sample-eccentricity-ln top-level access to the eccentricity_ln coordinate (rapidpe-rift/rift!55). Conflicted; resolved.

The 0.0.17.13 conflict, and why it should not recur

O4c 0.0.17.13 reached the byte-identical np.ln → np.log(ECC_MAX/ECC_MIN) fix independently, so the two sides differed only in the trailing comment. Resolved by taking O4c's statement verbatim and keeping this branch's derivation comment above it — so the shared line no longer diverges and a future O4c→O4d merge will not conflict there again. setup.py stays 0.0.18.0rc2; CHANGES.rst keeps both histories, O4d first.

The O4d-side eccentricity work merges cleanly alongside O4c's: --eccentricity-prior keeps its choices= constraint in both drivers, log_eccentricity_squared_prior and the broadened ecc-min correction survive, and all seven of O4c's ln(e) option/forwarding lines are present.

Testing

CI's test list has grown from 9 files to 15. Re-run in full after the merges: 475 passed, 3 skipped.

All seven refs — origin/{master,rift_O4c,rift_O4d}, ligo/{master,rift_O4c,rift_O4d}, junior/rift_O4d — report zero commits missing from this branch.