Fold a Gliese star into its Gaia source in Gaia's photometry, wherever HYG already put it, and place two naked-eye stars by SIMBAD's name for them

The Gliese fold (030447b) kept HYG's magnitude, band and colour over the Gaia entry's, putting
CNS3's V at Gaia's distance. Of the 49 folded, 15 lost a measured BP-RP and 6 their temperature and
radius (Gl 700.1C, Gl 632.2B, GJ 9800B, Gl 734B, Gl 323B, GJ 3605), and GJ 4285, V 11.45 where
SIMBAD has G 13.05, was drawn five times too luminous. A folded star now keeps the photometry of the
entry already there; HYG's name, id and type still come with the fold. Published catalogue, before
-> after: GJ 4285 V 11.45, 3 850 K, 0.491 R☉, 4.76e-2 L☉ -> G 13.05, BP-RP 2.74, 3 291 K, 0.297 R☉,
9.29e-3 L☉; GJ 3207 0.574 -> 0.278 R☉; Gl 700.1C no temperature, 4.54 L☉ -> 5 155 K, 1.214 R☉,
0.937 L☉; Gl 632.2B -> 9 044 K, 0.018 R☉; Gl 734B -> 3 360 K, 0.381 R☉.

The fold also looked its target up by name, so it saw only Gaia entries still bare. Where a
Hipparcos row of HYG's had already taken the source, the Gliese-only row of the same star stayed
beside it and the check counted none: Gl 251 at 5.76 pc beside HD 265866 (the host of GJ 251 b and
c) at 5.58, Gl 422 beside HD 304043. Gaia entries now carry their DR3 designation through the merge
(ETL only; the assets do not store it) and the fold looks it up; into an entry a HYG star already
describes it folds only where the two V agree within 0.5, so a companion SIMBAD gives its primary's
source stays. 14 more rows fold, 63 in all, each within 0.47 of the star it joins: Gl 251, Gl 422,
Gl 162, Gl 794, Gl 225.2C, Gl 905.2B, GJ 3232, GJ 4046, GJ 9490C, GJ 9608, 69 Tau Oph (Gl 700.1A)
and the second rows of HD 65277, HD 120237 and HD 336196. Stars within 10 pc 367 -> 366, within
25 pc 5 479 -> 5 468; distances without an error 346 -> 332; HYG survivors 11 478 -> 11 465.

validateMerge's count of Gliese rows beside their source took SIMBAD's map from the function the
fold takes it from, so a slip in the map turned both off: with it empty, GJ 2097 and GJ 4285 were
back inside 10 pc and the ETL passed. It now also asks for GJ 2097 and GJ 4285 beyond 20 pc and for
no Gl 251 or Gl 422 by name, and glieseGaiaDesignations refuses fewer than 3 200 HYG rows matched to
SIMBAD (3 352 of HYG's 3 801 measured).

4d47896 left 22 naked-eye HYG rows out as having no distance anywhere; two had one. HD 45951 (K0 III,
V 6.2) was on the map only as "Gaia DR3 3369454521490604416", HYG's declination being 31.7' out,
and θ¹ Ori A (HD 37020, HIP 26220) sits 0.1" from Gaia DR3 3017364132050194688, 2.643 ± 0.072 mas,
which the magnitude test refused because HYG gives V 4.98 against G 6.63. One cached SIMBAD query
now names the Gaia DR3 source of each naked-eye HYG row with no distance (199 of 206 HD numbers),
and a row nothing else places takes that source from the bright list, at its position: HD 45951 at
112.0 pc, under its name, and θ¹ Ori A at 378.4 pc. HD 45951 joins the required names. The twenty
still left out are named in build.ts; none has a bright source with a parallax five times its error
within a minute of arc nor under its SIMBAD name. Naked-eye stars 8 898 -> 8 899; 455 532 -> 455 519
stars.

The fold and the identity lookup are tested in star-merge.spec.ts. Guarded mutants (1 of 833
failed unless noted):
- the Gliese row's photometry kept: caught by "...in Gaia's photometry" (and the Hipparcos-row fold test, 2 failed)
- combine dropping the designation: caught by "folds one into a Gaia entry a HYG star of the same brightness..."
- folding whatever the brightness; tolerance 2 magnitudes: caught by "leaves a star with a Hipparcos error alone..."
- folding into bare entries only: caught by the Hipparcos-row fold test (and the leave-alone test, 2 failed)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

tools/etl/build.ts

@@ -10,6 +10,7 @@ import { fetchSolarSystem } from './fetchSolarSystem';
 import { BYTES_PER_STAR_META, BYTES_PER_STAR_POSITION, decodeStarCatalog, encodeStarCatalog, isDesignation } from '../../src/app/shared/models/star-catalog';
 import { fetchStars, glieseGaiaDesignations } from './fetchStars';
 import { ARCHIVE_EPOCH, archiveStarId, CATALOGUE_EPOCH } from '../../src/app/shared/astro/host-star-matching';
+import { foldsInto } from '../../src/app/shared/astro/star-merge';
 import { propagateProperMotion, raDegDecDistanceToXyz } from '../../src/app/shared/astro/coordinates';
 import { describeSources } from './sources/registry';
 import { dataPath } from './lib/paths';
@@ -24,7 +25,7 @@ function assertCondition(condition: boolean, message: string): void {
 
 /**
  * Stars whose magnitude is a stand-in, and distances published without an error. Measured 309 and
- * 346: the 44 Gaia sources with no G and the 265 archive hosts with neither V nor G; the 264 Gliese
+ * 332: the 44 Gaia sources with no G and the 265 archive hosts with neither V nor G; the 250 Gliese
  * distances (no Hipparcos parallax behind them) and 82 archive hosts the archive gives no error
  * for. Losing either field loses it for hundreds of thousands of stars.
  */
@@ -73,16 +74,23 @@ const MIN_COLOURS_FROM_TEMPERATURE = 40;
  */
 const MAX_ARCHIVE_IDS_OFF_THEIR_NAME = 5;
 /**
- * Every star the naked eye sees, kept at any distance (c64eea0): measured 8 898 of V 6.5 or
- * brighter, 1 663 of them past 250 pc. With no magnitude handed to placementDistancePc, 7 379 are
+ * Every star the naked eye sees, kept at any distance (c64eea0): measured 8 899 of V 6.5 or
+ * brighter, 1 664 of them past 250 pc. With no magnitude handed to placementDistancePc, 7 379 are
  * left and Rigel, Deneb and Alnilam are gone — and every other check passed, the HYG survivors
  * going down rather than up. HD 197770 and HD 45291 are two of the twelve only Gaia's bright
  * sources place, by position; without that lookup they are gone and the count drops by twelve,
- * which the floor alone would not see.
+ * which the floor alone would not see. HD 45951 is one of the two only SIMBAD's name for its Gaia
+ * source places, HYG's declination for it being 31.7′ out; θ¹ Ori A (HD 37020) is the other.
+ *
+ * Twenty HYG rows of V 6.5 or brighter are still left out, for want of a distance: neither HYG nor
+ * Hipparcos gives one, and Gaia DR3 has no source brighter than G 7.5 with a parallax five times
+ * its error within a minute of arc, nor under the name SIMBAD gives. They are β Phe, φ Cas, χ Aur,
+ * ο¹ Cen, Polis, 16 Sgr, ρ Cas, η Car, and HD 47240, 50820, 90772, 97534, 100198, 101205, 101947,
+ * 129092, 151804, 185936, 202214 and 212466.
  */
 const MIN_NAKED_EYE_STARS = 8_800;
 const NAKED_EYE_MAGNITUDE_V = 6.5;
-const REQUIRED_NAKED_EYE_STARS = ['Rigel', 'Deneb', 'Alnilam', 'HD 197770', 'HD 45291'];
+const REQUIRED_NAKED_EYE_STARS = ['Rigel', 'Deneb', 'Alnilam', 'HD 197770', 'HD 45291', 'HD 45951'];
 
 function validateStars(stars: StarRecord[]): void {
   assertCondition(stars.length > 0, 'No stars were produced.');
@@ -218,37 +226,48 @@ function validateStars(stars: StarRecord[]): void {
  *
  * A star kept twice leaves its two entries near each other on the sky, from *different* sources —
  * one catalogue does not list a star twice. Under an arcsecond that is never two stars at this
- * depth, so every such pair is a miss. Twenty-three survive today; the nineteen first counted were
+ * depth, so every such pair is a miss. Twenty-two survive today; the nineteen first counted were
  * all a second HYG row wanting a Gaia entry that already absorbed one (Gliese lists some doubles
  * twice), and the merge that trusted a Hipparcos parallax over direction left 1 112.
  *
  * The other failure leaves no close pair at all, because proper motion had already carried the
  * two entries tens of arcseconds apart — the 2026-08-24 refresh, where HYG sat at epoch 2000.0
  * and Gaia at J2016.0. What it does leave is HYG rows that found no counterpart: 36 056 of them
- * against the 11 478 today, and no counterpart was possible for most of those. 8 307 of them are
- * every star beyond 250 pc but the archive's planet hosts, which the main query never downloads: 1 660
+ * against the 11 465 today, and no counterpart was possible for most of those. 8 308 of them are
+ * every star beyond 250 pc but the archive's planet hosts, which the main query never downloads: 1 661
  * naked-eye stars kept at any distance, and 6 647 fainter ones that Hipparcos put inside
  * `ETL_STAR_DISTANCE_PC` while Gaia's parallax puts them past `ETL_GAIA_DISTANCE_PC`. The other
- * 3 170 are what Gaia genuinely lacks: 1 204 brighter than V 8, which it saturates on or measures
- * poorly, 1 804 between 8 and 12, and 162 fainter, 112 of them Gliese stars within 50 pc that
+ * 3 156 are what Gaia genuinely lacks: 1 202 brighter than V 8, which it saturates on or measures
+ * poorly, 1 793 between 8 and 12, and 161 fainter, 111 of them Gliese stars within 50 pc that
  * neither of its queries holds. So the headroom left to the ceiling tracks the gap between those two cutoffs as
  * much as Gaia's completeness.
  *
  * This bounds a merge that went wrong, and — loosely — a Gaia download that came back short: a
  * truncated answer leaves the HYG rows whose counterpart it dropped without one, so survivors go
  * *up*, not down. Measured on the main query when it was the only one, with 10 886 survivors
- * against today's 11 478: 11 004 at nine tenths of its rows, 12 711 at half, 16 258 at a third. So
+ * against today's 11 465: 11 004 at nine tenths of its rows, 12 711 at half, 16 258 at a third. So
  * this ceiling only catches a deep truncation, and `fetchGaiaStars` catches the shallower ones
  * with a row floor on each query.
  */
 const MAX_UNMERGED_TWINS = 100;
 /**
  * Gliese-only rows beside the Gaia entry SIMBAD names as the same star, which no geometry saw:
- * 49 before `foldByIdentity`, two of them false stars inside 10 pc (GJ 2097 at 6.41 pc and GJ 4285
- * at 6.80, which Gaia has at 24.47 and 28.25). The twin count above does not see them, being up to
- * minutes of arc apart.
+ * 49 beside a bare one before `foldByIdentity`, two of them false stars inside 10 pc (GJ 2097 at
+ * 6.41 pc and GJ 4285 at 6.80, which Gaia has at 24.47 and 28.25), and 10 beside one a Hipparcos row
+ * already described, Gl 251 at 5.76 pc beside HD 265866. The twin count above does not see them,
+ * being up to minutes of arc apart.
+ *
+ * That count takes SIMBAD's map from the function the fold takes it from, so a slip in the map
+ * turns both off together: with it empty, GJ 2097 and GJ 4285 were back inside 10 pc and the count
+ * read none. These are checked by name and distance instead.
  */
 const MAX_GLIESE_ROWS_BESIDE_THEIR_GAIA_SOURCE = 0;
+const FOLDED_GLIESE_STARS = [
+  { name: 'GJ 2097', beyondPc: 20 },
+  { name: 'GJ 4285', beyondPc: 20 }
+];
+/** Gliese-only rows of stars HYG also lists by their Hipparcos row, HD 265866 and HD 304043. */
+const ABSORBED_GLIESE_ROWS = ['Gl 251', 'Gl 422'];
 const MAX_HYG_SURVIVORS = 15_000;
 const TWIN_TOLERANCE_RAD = (1 / 3600) * (Math.PI / 180);
 
@@ -296,12 +315,26 @@ function validateMerge(stars: StarRecord[], gaiaDesignationById: ReadonlyMap<num
     twins <= MAX_UNMERGED_TWINS,
     `${twins} stars from different catalogues sit within an arcsecond of each other (at most ${MAX_UNMERGED_TWINS} expected), starting with ${example} — the merge is keeping the same star twice.`
   );
-  const bare = new Set(stars.filter((star) => star.source === 'gaia' && isDesignation(star)).map((star) => star.name));
-  const beside = stars.filter((star) => star.source === 'hyg' && star.distanceError === undefined && bare.has(gaiaDesignationById.get(star.id) ?? ''));
+  const byDesignation = new Map(stars.filter((star) => star.gaiaDesignation !== undefined).map((star) => [star.gaiaDesignation!, star]));
+  const beside = stars.filter((star) => {
+    const target = star.source === 'hyg' && star.distanceError === undefined ? byDesignation.get(gaiaDesignationById.get(star.id) ?? '') : undefined;
+    return target !== undefined && foldsInto(target, star);
+  });
   assertCondition(
     beside.length <= MAX_GLIESE_ROWS_BESIDE_THEIR_GAIA_SOURCE,
     `${beside.length} Gliese stars are drawn beside the Gaia source SIMBAD names them as, starting with ${beside[0]?.name} — the identity fold is not being made.`
   );
+  for (const expected of FOLDED_GLIESE_STARS) {
+    const star = stars.find((candidate) => candidate.name === expected.name);
+    const distancePc = star && Math.hypot(star.x, star.y, star.z);
+    assertCondition(
+      distancePc !== undefined && distancePc > expected.beyondPc,
+      `${expected.name} is at ${distancePc?.toFixed(2)} pc, not beyond ${expected.beyondPc} where Gaia measures it — Gliese stars are no longer folded into their Gaia source.`
+    );
+  }
+  for (const name of ABSORBED_GLIESE_ROWS) {
+    assertCondition(!stars.some((star) => star.name === name), `${name} is drawn beside the Hipparcos star it is — Gliese stars are no longer folded into their Gaia source.`);
+  }
   console.log(`  ${survivors} HYG stars have no Gaia counterpart; ${twins} unmerged cross-catalogue pairs within an arcsecond; ${beside.length} Gliese stars beside their own Gaia source.`);
 }