# Sources and claim checks

Access date for every record: **2026-09-04**. This ledger distinguishes source facts from this project's calculations. Public observations, an author's post and synthetic AI review do not imply endorsement by the source authors, NOAA, ANM or any scientist.

`verified-direct` means the cited source or downloaded file was inspected. `editor-supplied-extract` means material supplied by the editor. `official-link-unverified` means an official link was retained but its relevant contents could not be checked. No material published claim relies solely on the third category.

## Source register

| ID | Source, edition and usable link | Status | Access method and evidence | Limit |
| --- | --- | --- | --- | --- |
| S01 | [Radu Umbreș's Facebook post](https://www.facebook.com/radu.umbres/posts/pfbid0XJat5DUeyAqVFL7zmuhWEgzAEX6nCYnkdLVXTmV5EnpwSy2GqixDj3gZZ3PbE3EVl) | editor-supplied-extract | The editor supplied both the image and the public URL. The image provides the commission's starting point; an attempted direct web read returned a cache miss. | The URL is linked as supplied. The post's date and live contents were not independently authenticated; quantitative claims rely on the original paper and weather observations. |
| S02 | [Espín-Sánchez, Gil-Guirado and Ryan, *Praying for Rain*, QJE 141(3), DOI 10.1093/qje/qjag026](https://academic.oup.com/qje/article/141/3/2363/8676046) | verified-direct | Publisher HTML, Tables II–III, sections IV.B and V.B.3. | Observational cultural evidence and a theoretical mechanism. The Romanian calculator does not replicate its cultural analysis. |
| S03 | [QJE online appendix, available through the article](https://academic.oup.com/qje/article/141/3/2363/8676046#supplementary-data) | verified-direct | Publisher PDF, B.5 at PDF page 21, printed 83; D11 at PDF page 32, printed 94. | D11 gives rounded group means without subgroup counts. Retrieved D12 contains a placeholder; its numerical robustness table was not checked. |
| S04 | [NOAA GHCN-Daily specification, version 3.34](https://www.ncei.noaa.gov/pub/data/ghcn/daily/readme.txt); [dataset DOI](https://doi.org/10.7289/V5D21VHZ) | verified-direct | Official text and preserved `data/raw/readme.txt`; PRCP definition, measurement, quality and source flags. | PRCP is precipitation, not a diagnosis of agricultural drought. Source S is derived from synoptic reports and carries a caution about daily precipitation. |
| S05 | [NOAA daily station files](https://www.ncei.noaa.gov/pub/data/ghcn/daily/all/); [inventory](https://www.ncei.noaa.gov/pub/data/ghcn/daily/ghcnd-inventory.txt); [station metadata](https://www.ncei.noaa.gov/pub/data/ghcn/daily/ghcnd-stations.txt) | verified-direct | Downloaded snapshot `3.34-upd-2026090318`, accessed 2026-09-04. Every source URL and SHA-256 is in `data/manifest.json`. All 35 stored file hashes matched in this source audit. | Files can change at the same URL. The manifest identifies the actual snapshot. The inventory is not all Romanian meteorological observations. |
| S06 | [RoCliHom, Zenodo record 14880417](https://doi.org/10.5281/zenodo.14880417) | verified-direct | Preserved API metadata and data CSV. Data MD5 `25aabc60a77cfd8d9069d1d5fbd51b4b` matches provider metadata. | The checked public file is monthly. The record's description of daily source observations does not make the distributed CSV daily. |
| S07 | [INP/cIMeC, *Repertoriu național de patrimoniu cultural imaterial*, II](https://cimec.ro/wp-content/uploads/2021/09/Repertoriu-Patrimoniu-Cultural-Imaterial-vol-IIA-romana-si-franceza_0.pdf) | verified-direct | Official PDF downloaded and extracted; chapter 1, printed page 24, notes 112–113. | Scholarly compilation of ethnographic testimony. Historical attestation is not current prevalence. Its underlying cited field records were not all independently checked. |
| S08 | [Basilica, *Procesiune pentru ploaie, la Suceava*](https://basilica.ro/procesiune-pentru-ploaie-la-suceava/) | verified-direct | First-party HTML gives event 9 August 2015 and publication 11 August 2015. | Establishes a reported event. It is not a meteorological measurement or a complete event registry. |
| S09 | [Basilica, *Procesiune și rugăciune pentru ploaie într-o parohie din Călărași*](https://basilica.ro/procesiune-si-rugaciune-pentru-ploaie-intr-o-parohie-din-calarasi/) | verified-direct | Page published 22 July 2024 cites the Prozanic parish's report of the preceding Sunday. | Contextual example retained in the review dossier. Event date 21 July is inferred from the weekday and publication date. |
| S10 | Project outputs, `public/data/index.json`, station JSON files, `public/data/estimates.csv`, and `scripts/analyze.py` | verified-direct | Output values read directly. The original Băneasa verification is documented in `research/verification-review.md`. Current verification scope and results are recorded in `research/independent-verification.json`. | Derived evidence inherits observation, selection and model limits. Source audit is not an independent recalculation of every station. |
| S11 | [WMO/GWP, *Handbook of Drought Indicators and Indices*, 2016, WMO-No. 1173](https://www.droughtmanagement.info/literature/GWP_Handbook_of_Drought_Indicators_and_Indices_2016.pdf) | verified-direct | Official handbook linked from WMO; chapters 1 and 7, especially printed pages 1, 13–14 and 16–17. | Defines drought-monitoring context and distinguishes climate-relative precipitation indices from agricultural moisture measures. It supplies no new Romania-specific drought classification. |

## Material claims and arithmetic

Percentages generally use one decimal place; the opening uses rounded whole percentages, while the method also gives exact fractions and two-decimal arithmetic. Claims C05–C09 and C15 preserve the former Băneasa/warm-season example and its original audit scope. Current summer-opening claims appear separately below. Intervals are approximate pointwise 95% intervals. All Romanian calculations refer to 1961–2025 unless a control selects a subperiod.

| Claim ID | Published claim | Primary source and edition | Inputs and formula | Expected display | Limit or caveat |
| --- | --- | --- | --- | --- | --- |
| C01 | 23 eligible stations from 30 inventory candidates | S05, S10 | Country RO and element PRCP; at least 90% overall coverage and 25 years each with at least 90% coverage | 23 of 30; 544,103 valid station-days | Coverage rules are project choices. No population weighting or national probability. |
| C02 | Seven exclusions | S05, S10 | Four source-E series with 71.0% full-period coverage; three mainly synoptic series with 25.6%, 41.0%, 38.4% | Four with large gaps, three synoptic series with inadequate coverage | Excluded for coverage, not an assumption that all synoptic observations are invalid. |
| C03 | PRCP unit and flag handling | S04, S10 | Integer tenths of mm divided by 10; reject missing, nonblank QFLAG and measurement flag P | Strict exceedance `PRCP > threshold` | Equality is a day without exceedance. Source E denotes ECA&D; this is not direct ANM archive access. |
| C04 | Common historical cohort across deadlines | S10 | Known spell start; complete 28 days after the assessment date; count each episode once per duration | Same denominator for 1, 7, 14, 28 days | Full-cohort selection can exclude an early wet outcome followed by missing data. No missing day becomes dry. |
| C05 | Former Băneasa opening | S10, `all/warm/5`, d=14 | k/n = 30/277, 139/277, 197/277, 259/277 | 10.8%, 50.2%, 71.1%, 93.5% at 1, 7, 14, 28 days | April–September assessment dates. These are precipitation frequencies, not ritual success rates. |
| C06 | Former Băneasa uncertainty | S10, same configuration | 1,000 whole-year resamples, seed 20260904 plus the period length in years; percentiles 2.5 and 97.5 | Seven-day interval 45.3%–55.1% | Within-year dependence retained; measurement error and all cross-year dependence are not covered. |
| C07 | Former Băneasa calendar comparison | S10 | Former-example p=0.501805; month reference 0.497738; month-year reference 0.422491; subtract before rounding | 49.8%, difference +0.4 pp, interval −3.9 to +4.6; 42.2%, +7.9 pp, interval +3.3 to +12.2 | Both are descriptive. Round the full-precision difference, not displayed operands. |
| C08 | Former Băneasa daily contrast | S10 | At d=7, 45/537; d=30, 8/70. Difference 0.030487; interval −0.040321 to 0.107010 | 8.4% versus 11.4%; +3.0 pp, interval −4.0 to +10.7 | Different duration cohorts. One contrast does not establish the shape of the whole curve. |
| C09 | Former warm-season atlas example | S10 | d=14, w=7, >5 mm, April–September; min/max across eligible stations | Sulina 52/240=21.7%; Ceahlău Toaca 119/188=63.3% | Unpooled station comparison; mountain precipitation may include snow. |
| C10 | Deadline effect in a hypothetical example | Exact probability identity | `1 − (1 − 0.10)^w`, independent daily outcomes | 52% in 7 days; 77% in 14 | A mathematical illustration, not a fitted weather model. |
| C11 | 44% versus 30% ritual documentation | S03, D11 | Rounded raw group means, increasing versus non-increasing fitted tail hazard | 44% versus 30% | Group-level documentation. Main Table III separately uses n=1,195 and estimates +14 pp with continent controls. |
| C12 | Strict >5 mm and p99 classification | S02 V.B.3; S03 B.5 | Rainfall exceeds 0.5 cm; derivative at local p99 of wet-to-wet intervals | >5 mm; tail-local fitted slope | The calculator uses empirical frequencies and completed dry days. Figure V uses >5 cm for a separate seasonal illustration. |
| C13 | RoCliHom file excluded from daily analysis | S06 | Header `id,year,month,variable,value,flag`; 921,024 rows =156×123×12×4; no day field | Monthly distributed data, unsuitable for daily spell extraction | Exclusion concerns this checked file, not a claim that no daily Romanian archive exists. |
| C14 | Romanian ritual context | S07 p24; S08 | Caloian/Paparuda documented; Suceava event dated 9 August 2015 | Historical attestation and one dated report | No contemporary prevalence, shared origin, or efficacy inferred. |
| C15 | Original independent synthetic verification | S10 review evidence | Separate parser and implementation checked 27 configurations, 972 Băneasa rows and the default resampling intervals | 972 results reproduced | AI-generated verification. It does not supply external peer review, scientific endorsement or independent all-station replication. |
| C16 | Current Constanța summer opening | S10, station `ROE00108892`, `all/summer/5`, d=30 | 64 episodes across 51 years; k/n = 1/64, 12/64, 28/64, 45/64 | 1.6%, 18.8%, 43.8%, 70.3%; opening rounds seven and fourteen days to 19% and 44% | June–August assessment dates; each preceding day permits up to and including 5 mm. The historical period is 1961–2025. |
| C17 | Current Constanța uncertainty | S10, same configuration | 1,000 whole-year draws; common 64-episode cohort; percentiles 2.5 and 97.5 | Seven-day interval 10.5%–27.8%; fourteen-day interval 32.4%–55.1% | Pointwise, approximate and conditional on observed data. The example was selected editorially and was not preregistered. |
| C18 | Current calendar comparison | S10, same configuration, w=7 | p=0.1875; month reference 0.314595; month-year reference 0.142658 | Observed minus reference: −12.7 pp, interval −21.1 to −3.6; +4.5 pp, interval −2.8 to +12.3 | Distinct descriptive references; no causal interpretation. |
| C19 | Constanța daily contrast | S10, same configuration, w=1 | p at d=30 minus p at d=7; raw difference −0.037007; adjusted difference 0.018261 | Raw −3.7 pp, interval −7.1 to +0.1; adjusted +1.8 pp, interval −1.4 to +5.7 | Neither interval excludes zero; this does not estimate the original paper's spline derivative at p99. |
| C20 | Summer Băneasa estimate suppressed at d=30 | S10, station `ROE00108889`, `all/summer/5`, d=30 | 28 eligible episodes across 24 years; existing display rule requires at least 30 episodes and 10 years | Insufficient data; no percentage | Station-level coverage eligibility does not guarantee support for every configuration. |
| C21 | Individual historical episodes behind the opening | S10, `public/data/history.json` | All 64 eligible Constanța anchors; exact raw-record dates and first >5 mm day within 28 subsequent days | 45 returns within 28 days; 19 bounded non-returns | Fixed opening example, independent of calculator controls. No return within the observed window makes no claim about later rainfall or the end of agricultural drought. |
| C22 | Expanded independent verification | S10, `research/independent-verification.json` | Separate raw parser; all 1,296 rows each for Constanța and Băneasa; two bootstrap cases; 64 historical records | 2,592 rows reproduced; both default and former-example intervals checked; 64 dates and amounts matched | AI-generated executable verification, limited to those two stations; all-station invariants are separate. |
| C23 | New opening source-S sensitivity | S04, S10, `research/source-sensitivity.json` | Remove Constanța's 30 source-S days in June 2024 before history and completeness; 11/62 versus 12/64 | Seven-day 17.74% versus 18.75%, change −1.01 pp; fourteen-day 27/62=43.55% versus 43.75% | Sensitivity reuses production functions; it does not establish instrument homogeneity or validate source S everywhere. |
| C24 | Precipitation spells and drought indicators differ | S11; S10 definitions | Local climate reference and time scale matter for precipitation-deficit indices; agricultural applications consider water availability | The calculator measures a threshold-defined rainfall sequence | No SPI, soil-moisture estimate, drought severity ranking or current drought diagnosis is computed. |
| C25 | Rainfall totals in the current opening spells | S10, `public/data/history.json` | Sum the 30 days ending on each anchor, excluding the preceding wet day; all 64 records | Range 0–25.3 mm, median 4.6 mm; one zero-precipitation episode | These totals describe the selected Constanța cohort and do not constitute a climate-relative drought classification. |

The source descriptions and quantitative claims appear in the Romanian and English article, method and review pages, with former examples identified as first-edition material. Full raw URLs and checksums belong to the downloadable manifest. Dates in ritual reports, station coverage and calculation windows describe different units and are never pooled into a ritual dataset.


## First-edition verification record

- The original numerical audit covered 23,328 rows in 648 configurations. A separate raw-data implementation recounts all 972 Băneasa rows; production invariants across the other stations are not independent recounts. Archived evidence: `research/first-edition/numerical-checks.json` and `research/first-edition/independent-verification.json`.
- Accepted Băneasa observations in the original example include source S for June 2024. Excluding those 30 days leaves 275 original-example anchors and 139 seven-day successes (50.55%, versus 50.18%). The largest deadline-frequency change is 0.52 percentage points. See the archived `research/first-edition/source-sensitivity.json` and NOAA's source-S warning in the cited README. The current sensitivity program evaluates Constanța instead. This is a sensitivity using production functions, not another independent implementation.
- The measurement-flag audit includes blank, B and D flags by station and year; no accepted trace flag T occurs in this extract. Source IDs cannot establish homogeneous instruments, station locations, observation times, or unbiased missingness.
- Map geometry: Natural Earth 1:110m countries, public domain, accessed 2026-09-04; Romania feature extracted from https://raw.githubusercontent.com/nvkelso/natural-earth-vector/master/geojson/ne_110m_admin_0_countries.geojson . Terms: https://www.naturalearthdata.com/about/terms-of-use/ . Geometry locates measurements and carries no inference about ritual prevalence. Local extracted feature SHA-256: 7b852f92d91c96ca98d82774b36b73b6427591c46e9d8e41558603e23de2cefe.
- Typography: Barlow Condensed and IBM Plex, SIL Open Font License 1.1. License files are distributed with the fonts. https://raw.githubusercontent.com/google/fonts/main/ofl/barlowcondensed/OFL.txt and https://raw.githubusercontent.com/google/fonts/main/ofl/ibmplexsans/OFL.txt .


## Revision 2 verification record

The summer recalculation preserves all 648 former configuration objects, containing 23,328 rows, unchanged. The full output now contains 864 configurations and 31,104 rows across the same 30 input stations. All 10 production boundary tests and whole-output invariants passed. These checks preserve the full-period coverage result of 23 eligible stations and 544,103 accepted daily observations.

A separate raw-record implementation reproduced all 1,296 outputs for Constanța and all 1,296 for Băneasa. It also checked 1,000 annual bootstrap draws for the new Constanța seven-day example and the previous Băneasa warm-season fourteen-day-duration example, and reconstructed every field of the 64-row dated history. The evidence files linked above contain this expanded revision-2 scope. These remain synthetic checks rather than external scientific peer review.

The new opening and all claims here use the frozen NOAA snapshot already present in the first edition. Adding June–August changes the seasonal selection and the opening example; it does not introduce current observations or substitute a national average. The descriptive interval does not account for selecting an illustrative station after inspecting preliminary results.