Repository navigation
JAX ILE: --d-prior cosmo and cosmo_sourceframe - #378
Merged
Merged
Conversation
New RIFT/likelihood/jax_ile/distance_prior.py tabulates the Planck15 ln p(d_L) - 2 ln d_L once and interpolates it with numpy or jax.numpy, so one density serves the distance grids, AV/portfolio, the driver's 6-D prior and traced code. The driver forwards the prior to every marginalized likelihood; it was previously ignored with a note. The per-sample GH distance quadrature and the direct-marginalization policy refuse it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
oshaughnessy-junior
deployed
to
private-review-dispatch-rift
October 1, 2026 10:15 — with 
GitHub Actions
Active
GitHub Actions
Active
Adversarial review: --angle-marg-scheme multipeak integrates its local branch against d^2 and normalized with the cosmo grid, so it now refuses non-volumetric priors. New tests pin the driver's d_prior forwarding (AST), AV prior draws, the table support check and the multipeak refusal; each fails under the corresponding mutation. d_min = 0 is accepted. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
oshaughnessy-junior
deployed
to
private-review-dispatch-rift
October 1, 2026 10:58 — with
GitHub Actions
Active
GitHub Actions
Active
Second review round: the driver now refuses a cosmological --d-prior with the per-sample GH distance quadrature or either multipeak scheme before the precompute, instead of late in the wrapper. The AST wiring test also pins distance_prior= on the AV call. Both fail under their mutations. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
oshaughnessy-junior
deployed
to
private-review-dispatch-rift
October 1, 2026 11:10 — with
GitHub Actions
Active
GitHub Actions
Active
oshaughnessy-junior
marked this pull request as ready for review
October 1, 2026 11:55
oshaughnessy-junior
deployed
to
private-review-dispatch-rift
October 1, 2026 11:55 — with
GitHub Actions
Active
GitHub Actions
Active
![interhost-cross-review-canary-v2[bot]](https://avatars.githubusercontent.com/u/274802661?s=60&v=4){kind=small}
oshaughn
deployed
to
private-review-dispatch-rift
October 1, 2026 18:48 — with
GitHub Actions
Active
GitHub Actions
Active
This was referenced Oct 1, 2026
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The JAX ILE now honors
--d-prior cosmoand--d-prior cosmo_sourceframe. Before this change it printed "accepted but IGNORED" and integrated against d², and AV/portfolio refused both priors.RIFT/likelihood/jax_ile/distance_prior.pytabulates the Planck15ln p(d_L) − 2 ln d_Lonce and interpolates it withnumpyorjax.numpy. It is used by:--limit-distanceThe prior is traceable under
jitandgrad.GH distance quadrature (
--distance-gh-nodes > 0),--angle-marg-scheme multipeak[-jax]and--direct-marginalization-policyassume d², so the driver refuses them with these priors. pseudo_cosmo is unchanged; its pre-existing silent drop under AV +--distance-marginalizationis left for a separate PR.Validation (b4321b5, clean tree):
End-to-end on the A100 (3646c83): zero-noise injection at 2.5 Gpc, d in [10, 10000] Mpc, 3 seeds.
Both paths shift lnZ by 1.07 nats. Reweighting the AV-Euclidean fair draws predicts 1.118 ± 0.013.
🤖 Generated with Claude Code