# Pre-registration: the Paratethys outlines and "your place" (cand-romania-era-mare)

Written 14 Nov 2026, before any outline was digitised and before any fossil site was tested against one.

## What was checked before writing this (counts only, no outlines yet)
- PBDB, collections with cc=RO (API data1.2, fetched 14 Nov 2026, saved in `raw/`): 946 collections, all dated and with coordinates; 151 lie in 16–1.8 Ma, 116 at distinct coordinates (0.01°). Gate "at least 100 dated Romanian sites": **passed**.
- GBIF fossil specimens, country RO: 2,211 records. Overlap with PBDB not yet checked; not added to PBDB counts.
- Shorelines: no openly licensed, stage-by-stage Romania-scale shoreline dataset was found (Codex source hunt + own search). Usable published figures: Jipa & Olariu 2009 (Dacian Basin; Zenodo, licence "other-open", figures adapt Popov, Săulea, Papaianopol) and Palcu et al. 2023 Earth-Science Reviews 246, Figs 12–13 (CC BY 4.0, sketches of the Central Paratethys basins). Popov et al. 2004 is 1:20 million, palinspastic, licence not established; not digitised.

## Method for the outlines
Six snapshots: Badenian 14.5 Ma, Sarmatian 12 Ma, Pannonian/Maeotian 9 Ma, Pontian ~6.5 Ma, Dacian ~4.5 Ma, Romanian ~3 Ma. Each outline is traced by hand from the named figure onto today's coordinates, using the figure's own landmarks (Carpathian arc, Danube, Olt, Black Sea coast, Moesia, labelled basins). Outlines are palinspastic sketches and are drawn on today's relief; the Carpathians rose and the basins subsided since. No flooding of today's elevation model is done and no depth is claimed.

Stated error per outline (distance by which the true line of the source may differ from ours): 25 km for outlines traced from Jipa & Olariu maps that show the Danube or the Olt; 40 km for outlines traced from the Palcu sketches (no rivers or coast on the sketch); 60 km where the source itself marks a margin as uncertain. The outline is shown with this band.

## The fossil test (pre-registered)
Data: PBDB RO collections with latlng_precision in {seconds, minutes, 3, 4, 5, 6 decimals}. A collection belongs to a snapshot if its age interval (max_ma, min_ma) contains the snapshot age and is no wider than 4 Myr.
- Water sites: environment is marine, shallow/deep subtidal, carbonate indet., reef, lagoonal, offshore, basinal, coastal, marginal marine or lacustrine. Land sites: terrestrial, fluvial, floodplain, channel, cave, karst, fissure, loess, pond. Delta plain, empty environment: not tested.
- **Test A:** at least 80% of water sites of a snapshot lie inside that snapshot's outline widened by its stated error.
- **Test B:** at least 60% of land sites of a snapshot lie outside the outline narrowed by its stated error.
- A snapshot with fewer than 5 sites of a kind is reported as "too few", not as pass.
Result of each test per snapshot goes on the page whether it passes or fails. A failed snapshot is drawn as "weak" on the stage.

## Method for "your place"
Places: GeoNames populated places of Romania (shared store). For each snapshot a place is: **under water** if inside the outline narrowed by the error; **at the margin** if inside the outline but within the error of its edge, or outside but within the error; otherwise **land**. Ages are the snapshot's age range from one chart (Palcu et al. 2023 Fig. 1 / Jipa & Olariu Fig 1.5). The card shows the stages the place was under water or at the margin and the youngest of them, as a range. Water type is named per stage (sea; brackish sea; lake) from the sources. **No depth in metres is shown**: no source gives palaeobathymetry at a place.
Nearest fossil site: nearest PBDB collection with a water environment and Neogene age (≤ 23 Ma), precision as above; if none within 100 km, nearest water-environment collection of any age, labelled. Taxa listed come from PBDB occurrences of that collection, genus level, with the reference.

## Iron Gates
Timing is disputed (Messinian ~6–5.3 Ma, Pliocene, 2.0–1.8 Ma; Clauzon 2005, ter Borgh 2013, Andreescu 2009, Jipa et al. 2011). The page gives the range and the views, not one date.

## Addendum, 14 Nov 2026 01:xx UTC, written before any fossil site was tested against any outline
Outcome of the digitising so far: the Dacian Basin outlines of Jipa & Olariu (2009) Figs 2.5, 2.7 and 2.11 were segmented by colour and georeferenced on the cities printed in each figure (similarity fit, residuals 1.4–6.5 km RMS; scale agrees with the printed bar). The regional Maeotian map (Fig 2.8) was registered by overlap of its Dacian Basin with the Fig 2.7 outline (IoU 0.72). The Late Sarmatian regional map (Fig 2.6) and the Badenian map (Fig 2.3) could not be registered this way (IoU below 0.5, rotation unstable) and are **not used**.
For the Transylvanian Basin, the western Romanian (Pannonian) plain, the Moldavian foredeep and Dobrogea no figure could be georeferenced. These parts are drawn as hand polygons following which basins held water at each time in Palcu et al. 2023 Figs 12–13 and Jipa & Olariu 2009 Figs 2.3/2.6, clipped to today's low ground (ETOPO 2022; thresholds written in `scripts/basins.py`). Their stated error is **60 km** and they are labelled on the page as the weakest layer. The Black Sea after the Maeotian map is reused for the Pontian snapshots and drawn for earlier ones only east of 27.9°E; it is labelled as such.
The fossil test above is unchanged and is run once on the outlines as built; outlines are not adjusted afterwards.

## Addendum 2, 14 Nov 2026 (after the first and only test run; disclosed)
Test run on the outlines as built (Transylvanian cutoff 650 m), saved in `data/fossil_test.json`: Test A passed for Badenian (10/10 inside the widened outline) and Sarmatian (31/31; 22/31 inside the plain outline), failed for Early Pontian (3/5) and Romanian (0/12; those are Pliocene river-and-lake sites in the filled Dacian Basin, not sea), and had too few sites for the Maeotian and Late Pontian-Dacian snapshots. Test B had too few land sites in every snapshot (at most one).
Known misses found afterwards by looking at communes (not used to change any outline): Praid and Cacica, both on or beside Badenian salt, lie outside the Transylvanian and foredeep outlines because the cutoff removed the hillier parts of those basin floors. Raising the cutoff to 850 m was tried and did not bring Praid inside, so it was reverted. The outlines stay as pre-registered and the misses are listed on the page.

## Addendum 3, 14 Nov 2026
The Black Sea component for the Badenian and Sarmatian snapshots, announced in Addendum 1, had been left out of the build by mistake. It was added and the test re-run: every number in `data/fossil_test.json` is identical. The numbers on the page are from this final run.

## Addendum 4, 14 Nov 2026
After seeing the first render the hand polygons were smoothed (a 10 km opening and closing, to remove straight cut lines) and the Moldavian foredeep polygon was extended east to the map edge, because its straight eastern edge was an artefact of where the hand polygon stopped, not a shore. Error stays 60 km. The fossil test and the commune table were re-run on the final outlines; the numbers on the page are from this run.

## Addendum 5, 14 Nov 2026: every run, in order, and two corrections to the text above
Addendum 2 calls itself the "first and only" run. That was wrong. The fossil test was run these times, each on the outlines as they stood:
1. First build (Transylvanian cutoff 650 m, no Black Sea in the Badenian and Sarmatian): Badenian 10/10 widened (9/10 plain); Sarmatian 31/31 (22/31 plain); Early Pontian 3/5; Romanian 0/12; Maeotian and Late Pontian–Dacian too few.
2. Cutoff raised to 850 m (tried because Praid fell outside): no meaningful change (Badenian plain 10/10); Praid stayed outside; reverted.
3. Black Sea added to the Badenian and Sarmatian (as Addendum 1 had announced): numbers identical to run 1.
4. Final: polygons smoothed and the foredeep extended east (Addendum 4): Badenian 10/10 widened and plain; Sarmatian 31/31 widened, 23/31 plain; Early Pontian 3/5; Romanian 0/12; others too few. These are the numbers on the page.
Addendum 2 was also rewritten once, minutes after it was first written, before anything was published.
Second correction: the place method on the page uses four levels, not three: "well inside" (inside the outline narrowed by the error, as registered), "inside the drawn water" (inside the plain outline), "within the margin of error" (inside the outline widened by the error) and dry. The "inside the drawn water" level was added so that a commune's card could separate places well inside the water from places near a shore; it is a description of position and changes no outline or test.
Added after the advisor's review: the test's null. The widened outline contains 98% of all precise PBDB collections of any age or setting for the Badenian and 99% for the Sarmatian, and 97% of communes, so the registered rule cannot tell a good outline from a bad one at these errors. The page says so and leads with the plain-outline counts.

## Addendum 6, 14 Nov 2026 (after two independent reviews)
Traced shapes (Dacian Basin, Black Sea, Lake Pannon edge) were smoothed again to remove thin slivers and holes from the colour tracing (opening and closing of about 4-5 km; polygons under 150 km2 dropped), and the fossil test and commune table were re-run; the final numbers on the page are from this run. Collections whose name is a region (for example "North Moldavia") are no longer offered as the nearest site to a commune.
Run 5 (after Addendum 6; the numbers on the page): Badenian 10/10 widened and 10/10 plain; Sarmatian 31/31 widened and 24/31 plain (77%, under the 80% bar); Early Pontian 3/5; Romanian 0/12; Maeotian and Late Pontian-Dacian too few; land sites too few everywhere. Null (all 866 precise collections of any age or setting inside the same outline): Badenian 96% widened and 35% plain; Sarmatian 99% widened and 48% plain; communes inside the widened Badenian outline 97%. Addendum 5's figures (23/31, 98%) describe run 4 and are superseded by these.