# Pre-registration: do the Table, the Gate and the Column line up on a solar axis?

Written 13 Nov 2026 (UTC time in `docs/PREREG-alignment.sha256.txt`), **before any bearing, azimuth or distance between the three works was computed**. What I had seen when I wrote it: the OSM coordinates of the three works as printed by Overpass (not yet converted to a bearing), Wikidata's rounded coordinates, and the fact that Wikipedia and ICOMOS call the axis "west to east". I had not worked out the angle.

## The claim being tested

I found **no published source that states a solar alignment** (searched in English and Romanian on 13 Nov 2026: solstice, equinox, sunrise, "axa", "aliniere"). What the sources do say:
- ICOMOS evaluation of the UNESCO nomination 1473 (2024): the ensemble lies on a "west-east axis" (UNESCO document 208445).
- ro.wikipedia: the three works are on one axis "from sunset toward dawn".
- The brief for this page (Marius, 13 Nov 2026) names the **folk claim**: the Column, the Gate and the Table are aligned on a **solstice axis**.

So the test covers two readings, both fixed now: **(R1)** the claim as the brief words it, a solstice axis; **(R2)** the weaker "west-east" reading, which a solar reading would turn into an equinox axis. The page says that no source for R1 was found.

## Positions

- Primary: OpenStreetMap, read through Overpass on 13 Nov 2026. Column: node 925073636. Table of Silence: node 3477307934. Gate of the Kiss: centroid (mean of the vertices) of way 269780523.
- Cross-check (not "confirmation", Wikidata's coordinates are often copied from OSM): Wikidata P625 for Q2628659, Q9031865, Q2443657, and a measurement on ortho imagery (Esri World Imagery tiles, used to measure only, never reproduced), picking the centre of each work by eye on z19 tiles.
- Error budget: `e` = the larger of 5 m and the biggest OSM-versus-ortho disagreement found for any of the three works.

## Lines measured

1. **Primary axis:** the Table → Column initial bearing (great-circle on WGS84, degrees clockwise from true north).
2. Reported next to it: Table → Gate, and Gate → Column.
3. **Collinearity (test C):** the perpendicular offset of the Gate from the Table–Column line, in metres. Collinear if offset ≤ 2e (10 m at e = 5 m). If it fails, "the axis" is not one number and the page says so.

## Solar events (closed list, no other event may be added after seeing results)

Azimuths from `src/sky.mjs` (Meeus ch. 25), latitude of the Column, flat horizon, upper limb, Saemundsson refraction:
- rise and set at the **summer solstice**, **winter solstice**, **equinox** (declination zero): six azimuths;
- one optional dated event, only because a source dates it: sunrise and sunset on **27 Oct 1938** (completion of the Column, Wikipedia citing the engineer's account), reported as exploratory and never counted as support by itself.

The line is tested **in both directions**: the axis bearing against sunrise azimuths and the reverse bearing (+180°) against sunset azimuths, so an axis at 97° can match a sunrise at 97° and a sunset at 277°.

## Tolerance (fixed before measuring)

Half-width of the sun's disc 0.27° combined in quadrature with the position error of the baseline B (about 1.5 km): `T = sqrt(0.27² + (2e / B · 57.3)²)`. With e = 5 m and B ≈ 1.49 km this is 0.47°, **fixed at T = 0.5°**. If e turns out larger than 5 m, T is recomputed from this formula and published, and nothing else is changed.

## Decision rule

- **Aligned** on event E: |axis − E| ≤ T for the primary axis, and test C passes.
- **Near**: T < miss ≤ 1.5°. Reported with its size, counted as **not aligned**.
- **Not aligned**: anything else.
- Matching an arbitrary day of the year does not count. Over a year the sunrise passes through every azimuth between the solstice extremes, so "some day it matches" is always true. The page prints, from the same code, the chance rate: how many of the 365 days have a flat-horizon sunrise within T of the axis, and the chance that a random line matches one of the six named events (3 distinct orientations mod 180° × 2T / 180° ≈ 1.7 % at T = 0.5°).
- The **real horizon** (Copernicus GLO-30, bare-earth proxy as in `sunrise-alignment`) is secondary. It cannot turn "not aligned" into "aligned" unless the primary also passes; it is reported to show how far the true sunrise moves.
- Obliquity in 1938 differs from today's by less than 0.001°; ignored.

## Not part of this test (stated here so it cannot be added later)

- "The shadow points at the Table on some days" is a separate geometric claim; it is computed afterwards and reported as a result, with the distinction between **pointing** at the Table (a direction) and **reaching** it (needs the sun below about one degree, with buildings and terrain in the way).
- No test of Brâncuși's intent. A match, if any, shows an orientation, not a design aim.

## If the test fails

The page says so in the first screen after the picture and reframes: what the axis does line up with, and what the shadow of the Column actually does through the year.