# F4-conf-04 — Reference-list verification

# Question

For each of five references (R1–R5) in a draft manuscript, check against the provided source extract whether (i) the title matches, (ii) the year equals the first-submission year or the year of the reproduced version, and (iii) the finding attributed to it ("cited_for") is stated in, or follows arithmetically from, the extract. A reference that fails any check counts as unverifiable.

# Data and exclusions

Inputs: data/reference_list.json (5 items), the five extracts in data/sources/ (134, 58, 107, 66 and 91 lines) and data/README.md (17 lines). The only evidence used was the extracts (header plus body). I used nothing from memory or from outside bibliographic sources, and excluded no data.

# Method

out/analysis.py reads each extract and prints the lines it relies on. Those lines are the Title: and Bibliographic record: header lines plus the table rows and sentences that decide each finding.

• Title: both titles are normalised (accents stripped, lower case, "&" turned into "and", punctuation removed) and then compared for exact equality.
• Year: the entry year is compared with every submission year in the bibliographic record. It passes if it equals the first-submission year or the year of the reproduced version.
• Finding: every element of cited_for is checked against printed cells. The elements are magnitude, SE or CI, significance, timing and population/design. Printed derived quantities are re-computed where possible: exp(ATT)−1, |t|, β/mean, and affected-employment/affected-wage.

# Results

Unverifiable references: [R3]. 4 of 5 references pass all three checks.

R3 fails all three checks:

• Title: the entry's title, "Restaurants, Cafés and Neighbourhood Change: Commercial Entry and House Prices in Greater London", differs from the document's title, "Does Entry of Food-and-Drink Establishments Raise Local House Prices? Event-Study Evidence from London".
• Year: the entry gives 2024, but the paper was submitted in 2026 (v1 3 Mar, v2 2 Apr).
• Finding: the manuscript says prices jump by 3.4–3.7% in the entry year and do not rise further. The source says the reverse. It reports about 0.5% (0.005) in the event year, a gradual rise to 0.037 and 0.034 at τ=4 and τ=5, and states that "prices do not jump immediately". The citation gets the timing wrong.

Arithmetic checks that pass:

• R1: exp(−0.431)−1 = −35.0% and |t| = 2.05.
• R4: −0.268/0.366 = −0.732.
• R5: −0.0962/0.261 = −0.3686, consistent with the printed −0.3685 (36.85%).

# Uncertainty

This is a deterministic document check with no sampling uncertainty. The judgement calls are listed here.

• R1: Table 6 prints SE (0.210) for the 5 km CS-DR ATT, but Appendix Table A19 prints (0.218) for what appears to be the same estimate. I treated the main-text Table 6 as decisive, because the cited value appears there verbatim. Significance at 5% holds under either SE (|t| = 2.05 or 1.98), so the substance of the claim does not depend on which SE is correct. A stricter reviewer might still ask the authors to say which SE they mean.
• R3: the finding fails because of timing, not magnitude. The 3.4–3.7% figure does appear in the source, but for years 4–5, not the entry year.

# Limitations and open issues

Internal inconsistencies found in the sources (they are also listed in data_issues):

• arxiv-2609.27732v1:
  • The SE differs between Table 6 and Table A19 (0.210 vs 0.218).
  • The A19 note says 640 control regions; everywhere else says 608.
  • The abstract says fatal outcomes fall "by 35%", while Table 7 and Section 6.1 imply about 30%.
• arxiv-2510.03658v1: the text cites a Table 2 estimate (−0.0855, mean 0.455). That table is not in the extract, so I did not use it.
• arxiv-2603.03260v2: the Table 3 header is garbled in the source rendering. I used the column order given in the extract editor's note.

Other limitations:

• The extracts are partial. Anything outside them was treated as unavailable, but no cited element depended on omitted material.
• Author formatting was not assessed, as the rules instruct.