Pre-registration: do the records show a return? (14 Nov 2026)
Honesty note. The gate itself (more records since 2015 than in 1990–2010, corrected for observer effort) was fixed in the 13 Nov 2026 brief. The formal test below was written on 14 Nov 2026 after the first count of GBIF totals, and before any page was built. It is a one-sided exact conditional binomial test, not a Fisher test.
Rule
- Count: GBIF
taxonKey=2480389 (Gyps fulvus), country=RO, occurrenceStatus=PRESENT. Windows 1990–2010 and 2015–2026. - Effort: all birds (
taxonKey=212), country=RO, basisOfRecord=HUMAN_OBSERVATION, same windows. - Only human sightings count. GPS pings (MACHINE_OBSERVATION) are never sightings.
- Pass if the human-sighting rate per 100,000 all-bird records is higher in 2015–26 AND the exact test gives p < 0.05: given the n human sightings in both windows, P(old window gets as few as it did) when sightings fall in proportion to effort.
Result: FAIL (run 14 Nov 2026, scripts/analyse.py, numbers in data/analysis.json)
- GBIF lists 2,658 Romanian records. 50 of them are absences (
NOT_DETECTED, EBCC Atlas of European Breeding Birds, fieldwork 1972–95, no date): squares surveyed where no griffon was found. Presence records: 2,608. - Of those: 2,571 GPS pings from one bird (
1X, dataset "Griffon Vulture movement on the Balkans", FWFF Bulgaria), 21 museum specimens, 16 human sightings (2013: 2, 2023: 2, 2025: 10, 2026: 2). - By window: before 1990: 20; 1990–2010: 0; 2011–14: 2; 2015–26: 2,585 (2,571 pings + 14 human).
- Effort: all-bird human records 28,915 (1990–2010) vs 771,687 (2015–26), 26.7 times more.
- Rate 2015–26: 1.8 griffon sightings per 100,000 bird records; 1990–2010: 0. With 14 sightings, the chance the old window gets 0 at equal rates is 0.60. p = 0.60 > 0.05. Gate not met.
- Reading: the 2,658 are not a return. They are one bird's tag, 50 recorded absences and 16 sightings.
Addendum: series GBIF lacks (looked for 14 Nov 2026 after the fail, "aim higher")
- CRAR (Romanian rarities committee), report 2010–2014, accepted Gyps fulvus: 11 in category A (2010: 2, 2012: 3, 2013: 5, 2014: 1) and 5 in category C* (2013: 2, 2014: 3), 16 reports. Source read from the PDF (rombird.ro, pp. 8 and 27). GBIF holds only 2 of them. This shows GBIF's coverage of Romanian griffons is incomplete.
- Milvus Group, "Vultures in Romania" (Appendix I): 68 observations in total, 24 since 1950; "in the more than 30 year period between 1964 and 1997 there is only one record" (a dead juvenile, 1976); 11 in the last ten years before the document. The document is undated in the PDF; the publisher's file path says 2018 (upload), the text refers to a workshop, so the "last 10 years" are unplaced. The authors say observer growth alone does not explain the rise and that neighbouring populations had not grown then.
- Verdict: these compilations point the same direction as the neighbours' growth (very few records before 1997, many more since 2010), but they have no effort measure, no fixed windows and counts of reports, not birds. They cannot pass the pre-registered test and are not used to claim it passed.
- Not found / not checked: eBird/Observado annual totals; Serbian Iron Gates counts (the Serbian feasibility report says no griffon flights recorded there "in the last decades", no annual series).
What the page claims instead
Romania has no breeding griffons (sources disagree on when they stopped, see SOURCES). Neighbours grew. One tagged bird, of 18 in the dataset, flew into Romania three times in 2018–19 and crossed the Danube six times. Romanian releases are planned, not done.
Post-review notes (14 Nov 2026)
- The 14 human sightings since 2015 fall on 6 different days (several are one flock or one bird recorded by several users), and the two 2013 records are one bird on one day. With effort share 3.6%, the test needs at least 82 sightings, none in the old window, to pass. It could hardly have passed; the page says so. It is a weak test, not proof of no return.
- Serbia: the abstract of Marinković et al. 2020 reads "14 detected pairs in two colonies in 1992 to 262 detected pairs in four colonies in 2018" (found by Codex via the publisher's abstract page).
- Milvus: "Since the late 1960's ... there is no sign of the breeding of the species in the country" (read in the PDF).
Back