ASTR 430 · Lecture 01

Lecture 01: What Makes a Result Reproducible?

Research Methods in Astronomy

Research methods is astronomy's error budget for knowledge: each conclusion needs a traceable path back to observation and assumption.

Objectives and route

Separate a claim, evidence, computation, environment, and provenance.
Define and use claim-evidence chains and FAIR data.
Audit one quantitative or procedural relation with assumptions.
Connect the method to an archive, literature record, or proposal decision.

Terms for This Lecture

Claim-evidence chain

Definition: A traceable sequence connecting a scientific statement to observations, transformations, and assumptions.

Why it matters here: In What Makes a Result Reproducible?, Claim-evidence chain is needed to reason about claim-evidence chains and FAIR data rather than merely label the phenomenon.

Relationship: It is interpreted alongside Reproducibility and the lecture relation below; identify which quantities are observed, calibrated, or model-derived.

Limitation or common confusion: A missing field is a reproducibility defect, not merely a documentation preference.

\(R=(claim,data,code,environment,provenance)\)

Why the method matters

Research methods is astronomy's error budget for knowledge: each conclusion needs a traceable path back to observation and assumption.

Keep observation, calibration, transformation, model, and claim separate so each can be challenged.

claim-evidence chains and FAIR data. The geometry distinguishes inputs, transformations, and inferential limits.

Formal object

\(R=(claim,data,code,environment,provenance)\)

Define fields, units, defaults, version identifiers, and the boundary between measured, adopted, and inferred quantities.

Read the evidence

Identify which part is input, transformation, and claim. A caption should let a reader reproduce the path without asking the author what was omitted.

claim-evidence chains and FAIR data. The geometry distinguishes inputs, transformations, and inferential limits.

Worked audit

R has five required fields: claim, data, code, environment, and provenance.

A missing field is a reproducibility defect, not merely a documentation preference.

Estimate direction first, record units or decision criteria, then test one assumption. This check is generated from shared ASTR430 values.

Failure modes

A query without a release identifier is not a stable input.
A formal uncertainty does not include an unmodeled calibration offset.
A lower residual or p-value does not by itself validate a claim.

Active reasoning

Prompt: A result is not reproducible if its exact input, transformation, and software state cannot be reconstructed.

What would you record so a skeptical collaborator could rerun or challenge the conclusion?

Paired workshop

The workshop turns this method into a reproducible artifact: a query manifest, screening log, uncertainty table, figure critique, time budget, or proposal review. The deliverable includes result and provenance.

Takeaways

Separate a claim, evidence, computation, environment, and provenance.
Every result needs an evidence chain and limitation statement.
Label live-verified documentation, standard literature, and instructor-provided records separately.
Make one hidden assumption visible in the paired assignment.

Exact source anchors

Astropy documentation: Getting Started and Acknowledging or Citing; NASA Open Science Overview.

Archive, software, and ethics citations with verification status are in reference-log.md.