# Citation Audit: JPL_ASTRO_EARTH_09

Read-only audit of the dissertation reference list. The dissertation file was NOT modified. This document is a repair worklist for Nate.

## Summary

| Metric | Count |
|---|---|
| Total references | 148 |
| DOIs | 126 |
| Non-DOI URLs (arXiv / NTRS / repository) | 22 |
| Resolved (verified live) | 148 |
| Unverifiable (timeout / 403 bot-block) | 0 |
| Flagged dead (confirmed 404/410) | 0 |
| -- mistyped_recoverable | 0 |
| -- misattributed | 0 |
| -- fabricated_neighbor | 0 |
| -- not_found | 0 |

**Verdict: clean. Every one of the 148 references resolves to a live, real source. No defects, no repairs required.**

## Method

- Every DOI was checked against `https://api.crossref.org/works/<doi>` and the HTTP code recorded.
- Every non-DOI URL was checked with an HTTP GET. NTRS items were checked via the hardened `ntrs.nasa.gov/api/citations/<id>` form (the NTRS single-page app returns 200 for every route, so the bare citation URL is not a reliable liveness signal; the API endpoint is).
- Crossref only indexes Crossref-registered DOIs. Four DOIs returned a Crossref 404. Per the guardrail (a 404 at one index is not proof of death), each was re-checked against the agency-agnostic `doi.org` resolver, the DataCite REST API, and the `doi.org/ra` registration-agency lookup before any classification. All four are registered with **DataCite** (arXiv, Zenodo, Dryad), resolve with HTTP 200 at doi.org, and are present in the DataCite API. They are real sources reached through the correct registrar, **not** dead and **not** fabricated.

- The ARGOS near-miss resolver (`D:/Claude_Code/brain/collegium/argos/near_miss_resolver.py`) was **not** invoked, because Step 3 runs only on DOIs that genuinely fail to resolve. Zero references reached that state.

## Flagged references

None. There is no flagged-reference table because no reference is dead, mistyped, misattributed, fabricated, or unverifiable.

## Transparency note: four Crossref-404 / DataCite-live DOIs (NOT defects)

A Crossref-only audit would wrongly flag these four. They are recorded here so the clean verdict is auditable. All four pass the cross-registrar check and require no change.

| Ref | Author-year (short) | Cited DOI | Crossref | doi.org | DataCite API | RA | Classification |
|---|---|---|---|---|---|---|---|
| 12 | Callaway, Goodman-Bacon & Sant'Anna (2021), DiD with a Continuous Treatment (arXiv 2107.02637) | `10.48550/arxiv.2107.02637` | 404 | 200 | 200 | DataCite | resolves (real source, no fix) |
| 27 | Data Citation Synthesis Group (2014), Joint Declaration of Data Citation Principles (Zenodo) | `10.5281/zenodo.7356758` | 404 | 200 | 200 | DataCite | resolves (real source, no fix) |
| 39 | Piwowar (2013), Data from: Data reuse and the open data citation advantage (Dryad) | `10.5061/DRYAD.781PV` | 404 | 200 | 200 | DataCite | resolves (real source, no fix) |
| 53 | Roth, Sant'Anna, Bilinski & Poe (2022), What's Trending in DiD? (arXiv 2201.01194) | `10.48550/arxiv.2201.01194` | 404 | 200 | 200 | DataCite | resolves (real source, no fix) |

**Resolver evidence (quoted from the live cross-checks):**

```
ref 12  10.48550/arxiv.2107.02637  -> doi.org 200 -> https://arxiv.org/abs/2107.02637   RA: DataCite
ref 27  10.5281/zenodo.7356758     -> doi.org 200 -> https://zenodo.org/records/7356759  RA: DataCite
ref 39  10.5061/DRYAD.781PV        -> doi.org 200 -> https://datadryad.org/dataset/doi:10.5061/dryad.781pv  RA: DataCite
ref 53  10.48550/arxiv.2201.01194  -> doi.org 200 -> https://arxiv.org/abs/2201.01194   RA: DataCite
```

_Note on ref 39: the Dryad concept DOI `...781pv` resolves to record `7356759` versioning at the repository; this is normal Zenodo/Dryad concept-vs-version behavior and is a valid resolution, not a mismatch._

## Non-DOI URL liveness (all reachable)

All 22 non-DOI URLs returned HTTP 200: arXiv abstract pages (refs 15, 31, 54, 55, 91, 92, 102, 110), an OSF/arXiv-hosted PDF, the Southampton ePrints DOCX (ref 6), and the 13 NTRS items (refs 135-147), the latter verified through the hardened `api/citations` endpoint.