Skip to content

Native CosmoSIS SACC likelihood (sacc_like) #254

Description

@cailmdaley

Let the CosmoSIS inference chain consume the sp_validation SACC file natively through an sp_validation-owned subclass of the upstream SaccClLikelihood, so the fiducial (IA-only, ξ±) pipeline no longer depends on the 2pt-FITS converter.

Desired end state

  • An sp_validation-owned CosmoSIS module, SaccLikeUnions(SaccClLikelihood), reads a {version}.sacc file and produces a ξ± likelihood numerically identical to the existing 2pt-FITS + 2pt_like path. It overrides only what two hazards in the upstream class require: the upstream ξ± path evaluates a radian-θ theory spline at arcmin data tags (theory ≈ 0, χ² silently wrong), and it assumes a theory↔data ordering it never enforces. The subclass evaluates theory on radian θ while keeping self.sacc_data in arcmin, applies scale cuts in arcmin, and adds a loud ordering guard that raises on a misordered file. A third upstream defect (an undefined-name bug in the unused keep_tracers path) is deliberately left untouched.
  • The inference-prep step emits, from one assembled {version}.sacc, two CosmoSIS ini files: the legacy 2pt-FITS + 2pt_like path and a native sacc_like_unions path pointing directly at the SACC file. n(z) is read directly from the SACC NZ tracers — no 2pt-FITS is required on the native path.
  • The shim is self-retiring: a tripwire test fails the day upstream fixes the unit bug, signaling the subclass can be deleted.

Out of scope (dormant, unchanged): the PSF / xi_sys pipeline variant and the glass-mock rules stay on the existing FITS-assembly path.

Acceptance

  • A test suite asserts χ² / theory-vector / post-cut-count equality between the two paths to machine precision, and asserts the ordering guard raises on a misordered file. The suite skips cleanly where CosmoSIS is unavailable.

— Fable on behalf of Cail.

Activity

  1. changed the title [-]PR 7: adopt CosmoSIS native SACC likelihood (sacc_like), validated against the converter path[/-] [+]Native CosmoSIS SACC likelihood (sacc_like)[/+] on Jul 10, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions