Stonehenge | v1.39 | 1
A fixed angular model of Stonehenge counts recovers Cygnus near 
2965 BCE
Lucas Giovani Ribeiro * and Marcos André Simonssini
Independent Researcher, Uruana de Minas, Minas Gerais 38630-000, Brazil
Lucas Giovani Ribeiro: coin26@mail.com; ORCID 0009-0002-1417-0062
Marcos André Simonssini: thecygnuscode@gmail.com; ORCID 0009-0000-9076-6410
* Correspondence: coin26@mail.com
Abstract
One exterior calibration recovers four Stonehenge interior dimensions with 2.02% root-mean-square 
error. A specified count map supplies the angular target (7°,18°,4°). A census of star-centred circular 
fields identifies 19,939 distinct five-brightest-star groups without constellation boundaries. Geometry 
assigns roles before temporal fitting. Under the fixed angular target, Cygnus ranks first near 2965 
BCE, with 0.0975° RMS and a runner-up error 2.20 times larger, without requiring four-arm topology 
or local visibility. Reoptimization under a uniform necessary horizon condition retains first place 
among 17,860 groups. Removing angular scale gives a nearby Cygnus optimum near 2985 BCE but 
ranks it 68th in the complete census; four-arm subsets retain first place. The result separates 
boundary-independent recovery under an explicit angular scale from scale-free epoch compatibility. 
These retrospective comparisons require archaeological evidence for the count map and its 
transmission before historical attribution.
Keywords: Stonehenge; archaeoastronomy; archaeological reconstruction; geometric modelling; 
Cygnus; precession; sensitivity analysis; reproducibility
1. Introduction
Stonehenge, on Salisbury Plain in Wiltshire, southern Britain, combines an earthwork enclosure, 
buried settings and standing stones constructed in several episodes from about 3000 BCE onward 
[1–3]. The Aubrey Holes are 56 positions around the early enclosure. Sarsens are the large silcrete 
stones; a trilithon consists of two uprights supporting a lintel. The outer sarsen circle, inner trilithon 
horseshoe and four Station positions represent different architectural roles, while Y and Z are later 
concentric hole circuits. Present standing stones, missing settings and excavated traces therefore 
preserve different parts of the site's history (Figure 1).
The mean short side of the Station rectangle is 33.435 m, close to the approximately 33 m outer 
diameter of the sarsen circle [1–2,4]. This suggests a testable reconstruction question: can a scale 
estimated from exterior features reproduce interior dimensions without refitting? We also ask 
whether a specified map of the architectural counts yields angular intervals compatible with stellar 
groups selected by an explicit geometric rule. The reconstruction builds on established accounts of 
the Station rectangle, its astronomical discussions and its approximate Pythagorean proportions. 
The temporal word is the ordered numerical triple W=(28,72,16). Its fixed interpretation assigns one 
quarter degree per integer, giving (7°,18°,4°); a separate dimensionless comparison retains only 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37

Stonehenge | v1.39 | 2
7:18:4. A census generates candidate groups from circular fields and brightness, without consulting 
constellation boundaries or requiring four-arm topology.
The central archaeological question is whether a monument built in several phases can preserve a 
common design logic and an earlier reference epoch. The geometric hypothesis predicts that 
specified interior dimensions and directions remain related to the exterior frame under one 
calibration. The astronomical calculation supplies a separate temporal output for comparison with 
the construction sequence. Together, these components make continuity of design and an 
intentionally retained foundation reference distinct hypotheses with observable consequences and 
different evidential requirements. We evaluate the temporal model with a fixed angular scale and 
with scale removed. Both use geometric role selection; separate horizon-constrained branches 
impose a necessary condition for local observation. Regional assignments, complete field censuses 
and four-arm subsets retain their distinct comparison scopes.
The Station rectangle has long been discussed in relation to solstitial and lunar-extreme directions, 
horizon phenomena and latitude [5–8]. We examine its reduced side ratio and its relation to the 56-
position Aubrey circuit. A 5:12:13 rectangle closely represents the published sides; its nearest 
discrete rectangular form on the ideal circuit is 7|21|7|21. This provides a compact starting geometry 
whose consequences can be compared with other components of the plan.
Recent provenance, excavation and landscape research has substantially enlarged the 
archaeological account of Stonehenge [9–20]. It also establishes why a geometric reconstruction 
must identify the feature, boundary and source behind each input. We therefore distinguish 
published measurements, approximate plan readings and derived quantities, following the principle 
that reconstruction rules should remain traceable to their evidence [21]. The construction sequence 
is retained when interpreting relations between the early circuit and later settings.
The celestial target has a separate documented history. Ribeiro and Simonssini’s January 2026 
book describes a Cygnus grammar with primitive values 16, 28, 34 and 62, assigning a longitudinal 
role to 34 and a latitudinal role to 62 [22]. The public archive dated 26 August 2026 contains the 
ordered word W=(28,72,16), the five-star temporal operator and a nominated clock minimum of 
2965.2 BCE in its Calanais subtree. The word and epoch are also explicit in the public Calanais 
account dated 10 September 2026. S5 documents the exact files, recorded timestamps and verified 
archive checksum. The present Stonehenge analysis asks whether its material inputs regenerate 
that prior word and evaluates the same clock over a full 10,000-year window. Publication order 
documents the antecedent; it does not establish blinded discovery of the Stonehenge application.
Using published measurements, drawing proxies and astronomical catalogues, we test dimensional 
reconstruction, stellar selection, phase discrimination and sensitivity to metric and scale choices. S1 
preserves the original analyses; S5 supplies additional controls and adverse outcomes. S6 tests 
geometric role selection and persistence, and S7 combines that rule with dimensionless timing and 
uniform horizon optimization. S8 evaluates the fixed angular target across the complete nominal 
Hipparcos field census, followed by its four-arm subsets and free-scale controls. It also retains the 
earlier four-arm ratio panels, targeted kinematic variants, coordinate, loss and geometric-ratio audits. 
Prospective tests of geometric inheritance and intentional transmission are distinguished from these 
retrospective calculations.
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78

Stonehenge | v1.39 | 3
2. Materials and methods
2.1. Material inputs and the outer calibration
The primary objects are the Aubrey circuit, the four Station positions, the Sarsen outer envelope, the 
transverse Trilithon-horseshoe span and the Y- and Z-hole circuits. The Sarsen outer envelope is 
distinguished from the published inner-face diameter [23]. The four inventory counts are A=56 
Aubrey positions, B=30 outer sarsen uprights, C=10 trilithon uprights and D=4 Station positions. 
These refer to the reconstructed inventory across the monument's history, including absent settings; 
the four counts do not describe an inventory present at one date. Table 1 defines the notation and 
distinguishes the evidential status of the inputs.
Table 1. Operational notation and evidential status. Published measurements, approximate 
summaries, digitized proxies and model assumptions enter different operations; their sources and 
boundaries are specified in sections 2.1–2.4. The reconstructed inventory spans construction 
phases.
Symbol or 
input
Meaning and use
Evidential status
N, s
Aubrey and Station outer counts; N=A and 
s=D
Published archaeological reconstruction; 
includes absent settings.
a, b, c
Primitive short leg, long leg and 
hypotenuse
Derived frame selected within the declared 
comparison class.
A, B, C, D
Aubrey, sarsen, trilithon and Station 
inventory
Reconstructed counts across phases, not a 
simultaneous surviving inventory.
r, g, j, v
Radial, axial, longitudinal and latitudinal 
seeds
Derived outputs of the specified count map.
Q
Ordered eight-value register
Deterministic expansion of the seeds.
W
Ordered temporal triple
Derived numerical target; its timing 
interpretation is stipulated.
u
Common terrestrial scale
Parameter fitted only to the declared exterior 
dimensions.
Exterior 
measures
Published Station sides and Aubrey 
diameter
Published measurements with stated feature 
boundaries and source conventions.
Interior 
measures
Trilithon span, Sarsen outer envelope, Y/Z 
diameters
Approximate published dimensional 
summaries; excluded from scale fitting.
Plan 
coordinates
Cleal symbols and Rees labels or hatch 
centroids
Digitized drawing proxies; no measured per-
feature survey uncertainty.
Roles and 
timing scale
Outer 10:24:26; inner 4,10,12,16; minute-
per-numerical-unit convention
Declared model choices. The geometric 
roles were recognized with the plan 
available.
79
80
81
82
83
84
85
86
87
88
89
90
91

Stonehenge | v1.39 | 4
Symbol or 
input
Meaning and use
Evidential status
Perturbation 
bounds
Side factors, interior ±20% factors and 
±0.25 m coordinate scenarios
Stipulated sensitivity distributions or bounds, 
not measured archaeological errors.
The published Station centre-to-centre distances are 34.17 and 32.70 m for the short sides and 
79.93 and 80.26 m for the long sides [24–25]. Stations 91 and 93 survive; 92 and 94 are 
archaeologically located absent-stone settings. The primary Aubrey diameter is the 86.44 m Thom 
survey value tabulated by MacKie [25]. Alternative published conventions span 85.832–87.0 m [25–
26]. Inner inputs are the approximately 13.7 m Trilithon span, 33.0 m Sarsen outer diameter, 39.0 m 
Z-hole diameter and 54.0 m Y-hole diameter [1]. Hole diameters pass through hole centres; the 
Sarsen measure concerns its occupied outer envelope. The inner-face convention [23] is retained as 
a sensitivity. The ditch [27] and Heel Stone are excluded from calibration and the inner ladder. 
Supplementary Note 2 records boundaries, sources and precision classes.
For the contextual plan in Figure 1B, 56 Aubrey symbols and four Station proxies were digitized from 
Cleal et al.'s published Plan 1, using a 20 m printed scale bar spanning 357 pixels [1]. One raster 
pixel corresponds to about 0.056 m, which describes sampling rather than archaeological accuracy. 
Stations 91 and 93 use projected standing-footprint centres; 92 and 94 use socket-outline proxies, 
with incomplete excavation at 94. A leaning stone's footprint need not locate its original socket 
centre. The coordinates are local drawing coordinates, not georeferenced survey observations. No 
per-feature survey uncertainty is supplied by this extraction, and scanner precision is not assigned to 
these proxies. The primary dimensional fit uses the separately cited measurements above, not this 
new illustration (S5 Appendix, section 6). Figure 1C separately plots the archived Rees click-plan 
labels used in the interior directional diagnostics, identifying the entrance pair 1–30 and complete 
inner pairs 51–52, 53–54 and 57–58. They are label or hatch-centroid proxies, including separate 
fragment labels, rather than stone outlines or newly surveyed centres. Panels B and C have different 
source coordinates and are not registered together.
We estimate one scale from the exterior and apply it to the four inner roles without refitting. 
Calibration uses only the mean Station short side, mean long side and Aubrey diameter, with roles 
k=(10,24,26). For observed dimensions xᵢ and fixed model lengths kᵢ, the least-squares scale 
through the origin is û=Σ(kᵢxᵢ)/Σ(kᵢ²). The inner roles are k=(4,10,12,16). We report signed relative 
residuals and their root-mean-square value. The inner dimensions are held out of scale estimation; 
their model roles were recognized with the plan available. Recovery denotes retrospective 
agreement under these declared roles. Predicted values elsewhere in the paper are conditional 
model outputs; the holdout concerns scale estimation, not blind selection of the roles. Numerical 
outputs in the narrative are rounded for readability; tables and supplements retain reproduction 
precision. Printed digits do not estimate archaeological uncertainty.
Primitive Pythagorean triples with hypotenuse at most 100 define the metric comparison class. On 
an ideal N=56 circle of diameter Φ, chord k is Φ sin(πk/N). The joint diagnostic compares the 16 
primitive frames with 13 unequal-leg complementary step pairs. Independently perturbing each 
published side by uniform factors of ±1%, ±2% or ±5% tests ranking stability, with 200,000 
deterministic-seed trials per level. The perturbations are stipulated sensitivity scenarios. Complete 
estimators and comparison classes appear in Supplementary Notes 2–4 and 51.
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129

Stonehenge | v1.39 | 5
A separate metric-compression control independently multiplies each of the four interior summaries 
by a uniform factor in [0.8,1.2], holds the exterior scale fixed, and assigns each perturbed dimension 
its nearest integer multiple of that scale. It evaluates 1,000,000 trials with seed 20260916 
(Supplementary Note 9; S2, stonehenge_ladder_null.py). This local perturbation control does not 
repeat the selection of geometric constructions.
2.2. Plan comparisons and constructive operations
The archived digitization of Cleal et al.'s Plan 1 uses its printed scale bar [1]. A registered Rees-
based click-plan provides alternative proxies [28]. We distinguish continuous angular gaps between 
Stations, assignment to a numbered Aubrey-anchored grid, and shape comparison after fitting one 
common rotation about the Aubrey centre. All 14 discrete rectangular templates k|(28−k)|k|(28−k), 
k=1,…,14, receive the same rotation fit. A continuous 5:12:13 rectangle and a freely proportioned 
rectangle are comparators. A separate free circle through the four Station proxies tests centre and 
scale compatibility (Supplementary Notes 20–23).
Retrospective constructive tests ask how the frame can supply the proposed lengths and axes 
through explicit operations. For a right frame with short side x, long side y and diagonal z, copying x, 
halving y, subtracting x from z and constructing the inradius (x+y−z)/2 supply the four inner lengths. 
The pivot transformation is T(x,y,z)=(z−x,(x+y+z)/2,x+y). We test its algebra, circle consequences 
and candidate physical realizations. An opening axis is the normal to the line between support 
proxies; axial similarity is the squared inner product of unit directions. Path resultants, pivot rays, all 
220 trios of a declared 12-proxy set and a 14-shape orientation census are evaluated under 
disclosed rules. These later diagnostics use drawing proxies and do not replace the primary survey 
inputs (Supplementary Notes 31–43 and 51). A separate solar audit compares the ideal long chord 
with obliquity, including a Laskar-model sensitivity [29].
For the path decoder, a cord with equal terminal tensions defines a resultant from the sum of its two 
inward-pointing terminal unit tangents. The path length defines its numerical channel, while the 
normalized resultant defines its axis. We compare that axis with each opening normal using 
K(u,n)=(u·n)². The directional score gives equal weight to J against the entrance normal 1–30 and V 
against the mean of the three complete inner opening normals; exchanging J and V leaves the 
targets and weights fixed but supplies a complementary description for orthogonal outputs, not an 
independent control.
We examine two retrospective specifications of the route choice. A joint enumeration applies the 
34/62 nearest-integer length labels and orthogonal output axes to four symmetric route classes 
(Supplementary Note 54). An alternative entrance-based selector chooses between the two contour 
families by alignment with the entrance normal and between the two crossed families by alignment 
with its perpendicular. It then inspects mean lengths across each selected pair of mirrored routes; 
the inner-opening normals do not enter this selector (Supplementary Note 40).
2.3. Inventory map and stellar reference
The material map converts the architectural counts into four ordered seeds: r=A−B−C, g=C, j=B+D 
and v=A+C−D. A seed is a role-specific output of this map; the register is their deterministic 
expansion. With q=j−g and d=v−j, the expansion is Q=(r,q,d,j,d+q,v,3q,v+j), and the temporal word is 
W=(d,3q,r). The map is inverted without changing coefficients when the longitudinal and latitudinal 
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170

Stonehenge | v1.39 | 6
roles are exchanged. A bounded census compares alternative maps; a separate algebraic analysis 
solves r=D², d=A/2 and p²=g²+q² jointly, with p=A−B, over all strictly positive integer inventories 
(Supplementary Notes 5, 28–29).
The stellar module defines g as the Deneb–Sadr longitude step, j as the Deneb–Albireo longitude 
span and v as the mean latitude of Deneb and Fawaris. Coordinates are expressed in a fixed mean 
J2000 ecliptic frame. One nearest-integer rule is applied to each descriptor before differences are 
formed; r=16 is supplied by the material map. Historical catalogues [30], the ecliptic of the ancient 
date and continuous contrast-based recovery provide separate comparisons. Longitude span and 
mean latitude are retained as different types throughout inversion (Supplementary Notes 7–8, 24 
and 27–34).
2.4. Temporal evaluation and free stellar assignments
The fixed angular benchmark compares the role-ordered absolute right-ascension intervals Deneb–
Sadr, Sadr–Albireo and Aljanah–Fawaris, in the equatorial frame of date, with θ=(7°,18°,4°). This is 
exactly the earlier calculation against W=(28,72,16) in sidereal minutes: multiplying degree intervals 
by four multiplies every RMS by four and leaves its minimizers, ranking and runner-up ratio 
unchanged. Thus the time-unit expression is an equivalent coordinate conversion. The fixed target 
still assumes one architectural numerical unit corresponds to one quarter degree. Since D=4 in the 
reconstructed inventory, θ can also be written as (W/D) degrees; this is a retrospective material 
parametrization of the chosen scale, not an independent derivation of its ancient use. The primary 
archived astrometry is propagated with the long-term Vondrák–Capitaine–Wallace model and 
corrigendum [31–34]. We minimize the unweighted angular RMS across astronomical years −8000 
to +2000, refining every bracketed local minimum. A six-dimensional space-motion variant and 
channel omissions assess sensitivity. Earlier tables retain their explicitly labelled sidereal-minute 
scores for reproducibility. Ecliptic descriptors do not redefine the target during optimization.
A material word library is formed from N=A, s=D and the selected primitive frame. Its eight distinct 
values, {4,16,28,32,40,68,72,100}, generate 512 ordered triples. Each receives the same full-
window evaluation. Archaeological intervals enter afterward, as comparisons or expressly identified 
census filters. The supplementary solar audit instead compares a dimensionless ideal chord value 
with obliquity expressed in degrees, under two disclosed models [29,32–33]; that numerical angle 
convention is an additional assumption (Supplementary Notes 9–10, 25–26 and 51).
Retrospective stellar censuses test whether the stated rules recover the proposed configuration 
when the stellar identities are no longer supplied. The within-region stellar search assigns all five 
roles to distinct Hipparcos entries within each modern IAU constellation region. It uses one uniform 
catalogue-epoch convention, V≤5 as the principal magnitude limit, and nested V≤4 and V≤6 
comparisons. Candidates first satisfy the fixed spatial tuple, then receive full-window temporal 
optimization. A temporal pass has RMS no greater than the nominal graph under the same inputs. 
Ancient contrasts (24,28) and the 3000–2920 BCE initial-enclosure interval within Stage 1 are 
evaluated separately and jointly after fitting. Role-by-role brightness is compared without fitted 
weights. A three-star ablation removes the latitude descriptor and third temporal channel. A flat-
horizon calculation at 51.18° N tests whether all selected stars can be above the horizon during 
astronomical night, allowing season and sidereal time to vary (Supplementary Notes 44–48). This 
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211

Stonehenge | v1.39 | 7
construction interval, summarized in Darvill et al.’s construction chronology [2], is the window 
labelled “Stage 1” in the archived filters; it is narrower than the full duration of that stage.
2.5. Crossed selection and a free timing scale
A further retrospective panel selects the five brightest catalogue entries in each of 88 modern 
constellation regions from the archived V≤6 table, breaking magnitude ties by HIP identifier. Each 
set receives all 120 temporal-role assignments and all 512 library words. The resulting 5,406,720 
labelled descriptions reduce to 2,703,360 classes after exchanging the final unsigned pair. No 
spatial-tuple, chronology, visibility or shape filter precedes fitting. Full-window searches use 1-year 
and 0.5-year grids with local refinement and endpoints; the nominal threshold is recalculated from 
the same catalogue. These modern groupings delimit a reproducible panel, not an exhaustive set of 
prehistoric asterisms. The earlier unrestricted within-region stellar census is broader in star subsets 
but conditions on the spatial tuple (S5 Appendix, sections 1–3).
Two controls preserve the original material assumptions within this same bright-star panel. First, an 
independently implemented absolute-RMS search holds W=(28,72,16) fixed across all 5,280 role 
classes, using a 1-year grid, local refinement and endpoint evaluation. Second, the 512 words are 
classified by whether any strictly positive integer inventory generates them under the unchanged 
compiler. For W=(d,m,r), let q=m/3 and h=(d−r)/2. Feasibility requires integer q and h and B=q+h>0; 
then D=t>max(0,−h), C=t+h and A=d+q+t provide every positive integer solution. This is weaker than 
imposing all three original closure equations. Both controls were formulated retrospectively after 
examining the unrestricted panel; their constraints come from the original map, and they do not test 
alternative compilers (S5 Appendix, section 3.1).
A separate test retains W=(28,72,16) but removes its prescribed minute-per-integer timing scale. For 
temporal widths Tᵢ(t), set zᵢ=Tᵢ/Wᵢ and E(t)={mean[(zᵢ/mean(z)−1)²]}½. E is invariant under a common 
unit conversion; mean(z) is reported as the fitted common scale. This relative-dispersion objective 
changes channel weighting and is not the original absolute-RMS objective with its original ranks. It is 
evaluated for the primary and six-dimensional inputs and every bright-panel assignment. It frees only 
the temporal scale; the fixed-J2000 spatial tuple and the auxiliary degree-based solar relation 
remain separate conventions (S5 Appendix, section 4).
In an additional retrospective diagnostic, we retain W, the five stellar roles, timing scale, astrometry 
and equal channel weights, and minimize the same absolute RMS separately inside 3000–2920, 
2980–2720, 2940–2800 and 2620–2480 BCE. The first three intervals were already used as 
archaeological comparisons; the last represents the Stage 2 construction interval. Each bounded 
search samples one-year intervals, refines bracketed local minima and evaluates both boundaries. 
We distinguish membership of the unconstrained optimum in an interval from a new optimization 
restricted to that interval. Restricted ranking is checked against the archived global minima of all 
5,280 fixed-word classes. Error ratios are descriptive model contrasts, not likelihood ratios or 
chronological confidence levels (S5 Appendix, section 3.3).
A further retrospective check minimizes each absolute channel residual separately over 2620–2480 
BCE on 1-year and 0.25-year grids, refines bracketed local minima and evaluates both boundaries. 
We compare these minima with the three residuals at the nominated early optimum. Componentwise 
inequalities then determine whether the phase preference depends on channel weights (S5 
Appendix, section 3.4).
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253

Stonehenge | v1.39 | 8
2.6. Continuous dimensions and joint robustness
A retrospective metric-word diagnostic applies the existing constructive relations g=x, J=x+y, V=2z+x 
and r=z−x to the published exterior measures. It evaluates Wmetric=(2z−y,3y,z−x)/u, retaining 
u=(10x+24y+26z)/1352. Here x=33.435 m and y=80.095 m are the reported side means. Four 
previously declared Aubrey diameter conventions supply z=86.44,86.87,87.00 and 85.83168 m; a 
fifth branch uses the computed Euclidean diagonal √(x²+y²), not a measured diagonal. Each target 
receives the same fixed-scale RMS and free-scale E comparisons in the common catalogue. No 
branch is selected by its result. Target-change bounds applied to the archived global minima identify 
every class that could equal or improve the new Cygnus score; all eligible classes are refitted on 1-
year and 0.5-year grids with local refinement and endpoints. The ranking therefore inherits the 
numerical scope of the archived minima (S5, section 12).
A further diagnostic preserves S=(10,34,62) and W=(28,72,16) but permits separate positive spatial 
and temporal scales. It evaluates both assignments of the final temporal pair to the spatial latitude 
role: 10,560 typed assignments corresponding to 5,280 temporal classes. Spatial error is assessed 
separately by relative dispersion and by the minimum largest fractional residual; temporal error is E. 
A candidate dominates Cygnus only if it does not worsen either error and improves at least one. No 
weighted spatial–temporal sum or chronological filter enters the global comparison. All classes, sign 
conventions and complete frontiers are retained (S5, section 11). The additional phase audit also 
profiles a common timing scale under least squares and a minimax fractional loss. S5, sections 9–
10, reports those sensitivities and the distinct consequences of imposing the original spatial 
constraints.
2.7. Geometric selection of temporal roles
A supplementary diagnostic transfers the existing role selector from a separate stellar study [47, 
section 2.9] to the same five-brightest-star sets used here. Apparent V magnitude defines 
membership and brightness order. For each quintet, angular distances select the unique four-arm 
minimum spanning tree hub when present; otherwise the angular medoid is used. The brightest non-
hub star defines the first endpoint, and the remaining star with the most opposite tangent bearing at 
the hub defines the third role. The remaining unordered pair supplies the third interval. The rule does 
not use W, temporal residuals, archaeological dates, the ecliptic tuple or a Galactic target. S6 
specifies tie handling, the strict tree-margin test and the independent implementation.
The rule is applied to all 88 Hipparcos quintets at J1991.25 and J2000, and to all 88 archived HYG 
quintets. The selected Cygnus clock is then recomputed with the original word, scale and search 
window. Six archived kinematic scenarios test the common trio/quintet median, tree and role 
persistence at the existing clock epoch and at annual samples from 3000 to 1500 BCE. Catalogue 
membership and brightness ranks remain fixed. A subsequent scope check evaluates all 88 
Hipparcos groups at three specified historical epochs. These are retrospective model checks; S6 
records their order and complete results.
2.8. Dimensionless timing and a local horizon test
Let w=(7,18,4) denote the ratios of W=(28,72,16). For each of the 88 J2000 geometric assignments 
from section 2.7, let dᵢ(t) be the three absolute wrapped right-ascension differences relative to the 
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293

Stonehenge | v1.39 | 9
equator of date. Define qᵢ(t)=dᵢ(t)/wᵢ and E(t)=sd(q₁,q₂,q₃)/mean(q₁,q₂,q₃), using population standard 
deviation. E vanishes exactly when the three intervals share the target proportion. Multiplying every 
interval by any positive unit-conversion factor, or every target component by a common positive 
factor, leaves E unchanged. The metric is the existing S5 free-scale relative dispersion, now applied 
to one geometrically selected assignment per group. It requires neither the fixed-J2000 ecliptic tuple 
nor a minute-per-count convention. The count map, regional membership, apparent-brightness 
selection, interval topology and relative-error loss remain specified model choices.
We search astronomical years −8000 to +2000 without archaeological chronology constraints, using 
0.5-year and 1-year grids with refinement of every bracketed minimum and both endpoints. Every 
selected role is checked against S6 and every unrestricted score against S5. A common positive 
rescaling of the angular units or target is checked over all sampled configurations. Removing scale 
leaves two independent interval ratios; the fitted epoch is still a model output. This calculation is a 
recombination of existing components, not an independent replication of their discovery.
The local-observation branch imposes the same necessary horizon condition on every group. At 
latitude φ=51.18° N, a star of declination δ has geometric upper culmination h=90°−|φ−δ|. All five 
stars must have h>0 at the candidate epoch. We identify every admissible epoch segment on both 
grids, refine its boundary crossings and reoptimize E within those segments, including boundary 
infima. Thus a group is not discarded merely because its unrestricted optimum is invisible. This 
condition is necessary for observing the complete configuration locally; it does not model darkness, 
extinction, terrain or transmission of knowledge from another latitude. We separately test 
simultaneous altitude feasibility for Cygnus and Tucana at their unrestricted optima. S7 records the 
complete search, input provenance and numerical scope.
2.9. A field census without constellation boundaries
For each of the 5,041 archived Hipparcos entries with apparent V≤6, we centre a spherical cap on 
that star and increase its radius through every stellar entry event. We record each distinct set of the 
five brightest enclosed stars while the centre remains among them. Ties in apparent magnitude are 
ordered by catalogue identifier. The primary census retains every distinct quintet, without a tree or 
horizon filter, and assigns one role ordering by section 2.7: unique four-arm minimum spanning tree 
hub when present, otherwise angular medoid; brightest non-hub endpoint; most opposite tangent-
bearing endpoint; and remaining unordered pair. We separately identify the four-arm subset with any 
member as hub and the stricter subset whose hub is also a generating cap centre. The generator 
reads neither constellation regions nor the timing target, horizon or archaeological dates. All cap 
radii are enumerated; none is fitted to the temporal score.
The complete census uses nominal Hipparcos states propagated to J2000 for group and role 
selection, held fixed during every epoch search. For absolute wrapped right-ascension intervals dᵢ(t), 
measured in degrees, the fixed objective is R(t)=√{Σᵢ[dᵢ(t)−θᵢ]²/3}, with θ=(7°,18°,4°). Multiplication by 
four expresses the same residual in quarter-degree units, equivalent numerically to the original 
sidereal-minute representation; this conversion changes neither epoch nor rank. The assumed scale 
itself remains substantive. We minimize R and the dimensionless E separately over astronomical 
years −8000 to +2000 on 1-year and 0.5-year grids, refining bracketed minima and endpoints. For 
the fixed-scale local branch, R is reoptimized on every admissible horizon segment from section 2.8, 
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334

Stonehenge | v1.39 | 10
including boundary infima. Complete free-scale rankings use E without a horizon filter. S8, section 9, 
supplies every candidate, score and numerical comparison.
For the earlier, stricter four-arm subset with hub equal to cap centre, we additionally construct the 
fields separately at J2000 and 3000 BCE, using both the Hipparcos table and the archived 5,070-
entry HYG V≤6 table. These epochs govern group construction only. Each retained assignment in 
these ratio panels receives the same full-window E optimization and separate horizon reoptimization 
in section 2.8. The grouping rule is held fixed during each temporal search. Hipparcos uses the 
archived angular proper-motion propagation from J1991.25; HYG uses linear spatial motion from 
J2000, with missing radial velocity set to zero. Nominal catalogue magnitudes are held fixed. Nested 
V≤4 and V≤5 panels restrict the faintest retained member. Six further sensitivity catalogues replace 
the five Cygnus states with the S6 variants, repeat the complete J2000 field census and refit every 
affected or newly generated field. Unaffected scores are reused only for identical states and roles. 
These ratio-panel variants share the HYG background and are not independent surveys. Their 
robustness results are not transferred to the new complete fixed-scale census.
The field family is neither an exhaustive search of all five-star subsets nor a superset of the earlier 
88 regional assignments or 5,280 free-role classes. Star-centred circular caps, five members, 
brightness order and the geometric role rule were chosen retrospectively with Cygnus known. Four-
arm topology defines reported subsets rather than entry to the complete fixed-scale comparison. 
The strict tree condition is checked by d(i,j)>max[d(h,i),d(h,j)] for every leaf pair about hub h, with 
tolerance 10 ¹  degree. The complete enumeration independently reproduces the 256 cap-centred 
⁻ ⁰
four-arm assignments archived in the earlier census. Search-grid agreement supports numerical 
stability without certifying continuous-time global minima. S8 retains the complete scores and 
controls for coordinate invariance, losses, target permutations, continuous dimensions and spherical 
axis ratios.
2.10. Information order
The prior public record fixes the stellar grammar, role order, rounding rule, W and temporal operator. 
The Stonehenge map, dimensional ladder and subsequent diagnostics were developed with relevant 
data available. Archaeological dates do not constrain the original optimization or fixed-word ranking; 
explicitly labelled phase tests use them. S5 fixed-word and integrality checks followed its 
unrestricted panel, followed by phase and robustness diagnostics. S6 transferred the geometric 
selector after both stellar and Stonehenge results were known. S7 combined geometric roles, ratios 
and uniform horizon reoptimization after preliminary rankings and Tucana’s horizon failure were 
known. S8 invariance and loss checks preceded the spherical axis audit, then the field census and 
kinematic recensus. The complete fixed-scale census and its matching free-scale comparison were 
executed after those four-arm results were known. Both field definitions and the angular target 
remain retrospective choices in this application. These are retrospective tests, not preregistered 
discoveries. Supplementary Notes 1, 14 and 50–54 and the S5–S8 protocols retain their execution 
order and adverse outcomes, including rank losses and unsuccessful inverse constructions.
OpenAI ChatGPT assisted with literature retrieval, drafting, code construction and the additional S5 
computational audits in September 2026; its exact model version was not exposed in the authors’ 
working record. It also assisted with the October 2026 manuscript revision, S6–S8 computational 
diagnostics and reproduction checks. The code was executed against archived inputs, with checks 
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376

Stonehenge | v1.39 | 11
and outputs supplied in the accompanying package. No archaeological survey or catalogue 
observation was generated by the service. Scientific interpretation and responsibility remain with the 
authors.
3. Results
3.1. The outer frame selects a simple geometry and recovers interior dimensions
The 5:12:13 rectangle ranks first among the 16 primitive candidates, with 1.57% side RMS against 
6.64% for the next candidate, 36:77:85. Combining metric and cyclic comparisons again selects 
5:12:13 with the 7|21|7|21 partition, first among 208 combinations. It remains first in every side-
perturbation trial at ±1% and ±2%; at ±5%, it remains first in 96.72% of metric-only and 97.63% of 
joint trials. Figure 1 locates the archaeological setting and the exterior positions used by the 
reconstruction.
377
378
379
380
381
382
383
384
385
386
387

Stonehenge | v1.39 | 12
Figure 1. Archaeological setting, input proxies and measurement definitions. (A) Philip Henry Sharpe's 
1906 aerial photograph, Royal Engineers; public domain through expired Crown copyright, via 
Wikimedia Commons. (B) Cleal Plan 1 Aubrey symbols and Station proxies [1]; squares are standing 
footprints, diamonds absent-setting proxies. (C) Archived Rees click-plan labels; solid segments join 
supports, dotted lines show their normals, and crosses distinguish fragment labels. These source 
coordinates are separate from B. (D) Boundaries of the published primary dimensions. Neither proxy 
panel supplies new independent measurements, complete stone outlines or all missing settings. 
Sources and reproduction code are in S5.
The discrete construction gives hypotenuse-normalized legs 4.975 and 12.010, with 0.36% RMS 
against 5 and 12. At the independently published 86.44 m diameter, its chords are 33.08 and 79.86 
m, differing from the observed short- and long-side means by −1.06% and −0.29%. The side ratio 
therefore supports 7|21|7|21 as an approximate discrete shape (Figure 2).
388
389
390
391
392
393
394
395
396
397
398
399
400

Stonehenge | v1.39 | 13
Figure 2. Idealized 56-fold Aubrey construction. Model gaps 7|21|7|21 yield chords close to the 
reduced 5:12:13 legs. The labels 0, 7, 28 and 35 are relative indices within the ideal model. 
Supplementary Notes 20–23 give the shape comparisons and the separate anchored-position test.
Among all 14 discrete templates, 7|21|7|21 gives the smallest shape residual in each archived plan 
convention after fitting a common rotation. The Cleal continuous gaps, 7.178|21.064|7.127|20.631 
steps, round to the same sequence. In Cleal its angular RMS is 0.86°, compared with 2.81° for the 
next discrete template. The continuous 5:12:13 comparator gives 0.79°. At common centre, 
orientation and diameter, corresponding vertices of the two ideal models differ by 0.09 m. 
Supplementary Notes 20–23 and 32 distinguish these shape comparisons from registration with 
individual Aubrey positions.
One outer scale recovers the four inner dimensions at 2.02% relative RMS without refitting (Table 2; 
Figures 3–4). The fitted scale is u=3.3314 m, and every inner prediction uses that same value. The 
simplest transfer is the shared width: the outer rectangle's short side and the Sarsen outer diameter 
both represent 10u. Copying that length predicts 33.31 m against the approximate 33.0 m Sarsen 
envelope.
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416

Stonehenge | v1.39 | 14
Figure 3. One outer frame supplies four inner lengths. (A) The idealized Station short side and Sarsen 
outer diameter share the model length 10u; the other circles and central segment show the adopted 
inner roles at schematic positions. (B) Cutting the 10u by 24u rectangle along its 26u diagonal supplies 
four transfers: copy the short side, halve the long side, subtract the short side from the diagonal, and 
construct the inradius 4u. The inradius is transferred as the candidate span. (C) The reported short-
side mean and approximate Sarsen outer diameter differ by about 1.3%. (D) One outer-only calibration 
predicts all four inner dimensions without refitting (Supplementary Note 31).
417
418
419
420
421
422
423
424

Stonehenge | v1.39 | 15
Figure 4. Four inner dimensions predicted from the outer Station–Aubrey calibration. Each prediction 
uses the same scale without refitting; the combined relative RMS is 2.02%.
Table 2. Inner dimensions predicted using the outer-only scale.
Held-out role
k
Observed m
Predicted m
Residual
Trilithon horseshoe 
span
4
13.700
13.326
-2.732%
Sarsen outer 
envelope
10
33.000
33.314
+0.952%
Z-hole circuit 
diameter
12
39.000
39.977
+2.505%
Y-hole circuit 
diameter
16
54.000
53.303
-1.291%
In the declared metric-compression control, 72,285 of 1,000,000 perturbed configurations reach or 
improve the observed 2.02% RMS, a rate of 7.23%; the median simulated RMS is 4.12%. These 
results quantify the gain over the median and the frequency of comparable nearest-integer 
compression in this local ensemble. The frequency describes this stipulated perturbation process, 
not a probability for the full reconstruction or prehistoric intention (Supplementary Note 9).
The same length-transfer operations select 7|21|7|21 from all 14 discrete templates at fixed Aubrey 
diameter, with 2.12% RMS against 8.75% for the next template. Cutting the 10u by 24u rectangle 
along its 26u diagonal gives a right triangle with inradius 4u; half its long side is 12u, and diagonal 
minus short side is 16u. These operations express the existing Trilithon, Z-hole and Y-hole roles, 
respectively (Figure 3). Together with copying the short side, they specify how each of the four inner 
lengths can be constructed from the outer triangle (Supplementary Notes 31–32).
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439

Stonehenge | v1.39 | 16
The free circle fitted to the four Cleal Station proxies approximately matches the Aubrey centre and 
scale: its diameter is 86.49 m, and its centre lies 0.68 m from the separately fitted Aubrey centre. 
The 56 Aubrey proxies have radial RMS 0.63 m against that Station-derived circle. Alternative 
sources, every Station omission and their sensitivity scenarios are retained in Supplementary Note 
21.
3.2. The counts recover an ordered register and a unique conditional inventory
Alongside its measured lengths, the monument supplies four architectural counts. These counts 
generate the ordered seeds through one map:
r=A−B−C=16; g=C=10; j=B+D=34; v=A+C−D=62.
With q=24 and d=28, their expansion is:
Q=(r,q,d,j,d+q,v,3q,v+j)=(16,24,28,34,52,62,72,96).
The inverse recovers the four counts from the ordered seeds:
A=(r+j+v)/2; B=(−r−2g+j+v)/2; C=g; D=(r+2g+j−v)/2.
Exchanging j=34 and v=62 changes the recovered Station count from four to 32. The unchanged 
inverse returns (56,30,10,4) in the specified order and (56,30,10,32) under exchange, leaving the 
other three counts unchanged (Figure 5A). The ordered stellar roles therefore have an explicit 
consequence within the reconstruction.
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456

Stonehenge | v1.39 | 17
Figure 5. Directed recovery and conditional material uniqueness. Panel A shows the observed 
inventory, typed seeds and exact inverse; exchanging j and v changes the recovered Station count 
from 4 to 32. The stellar roles use the fixed J2000 ecliptic frame. Panel B solves three existing closures 
within the same compiler: r=D², d=A/2 and p²=g²+q², with p=A−B, g=C and q=B+D−C. Their unique 
positive integer solution yields the primitive 5:12:13 frame and W=(28,72,16).
The three specified closures admit exactly one strictly positive integer inventory within the fixed 
compiler, with no upper bound imposed on inventory size: (56,30,10,4). The conditions are r=D², 
d=A/2 and p²=g²+q², where p=A−B. The same solution supplies the doubled frame 
(g,q,p)=(10,24,26) and W=(28,72,16). Within these conditions, both the inventory and its overall size 
are fixed. The factorization proof and condition-removal controls are in Supplementary Note 29; the 
frame and word follow from this conditional solution (Figure 5B).
The relation can also be read directly from the outer counts and primitive frame. With N=56, s=4 and 
(a,b,c)=(5,12,13), the identities N=2s(b−a) and N+s²=s(c+a) give W=s(b−a,c+a,s)=(N/2,N+s²,s²). The 
outer inventories yield 56−30=26 and 30−26=4. These equivalent expressions locate the same 
frame and word in the chosen material system; their full algebraic and quadratic Gram forms are 
preserved in Supplementary Notes 17–19 and 52.
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473

Stonehenge | v1.39 | 18
Two of 40,960,000 ordered maps return the target seeds, and both are invertible. The comparison is 
complete and bounded, with each coefficient restricted to −1, 0 or 1. A separate enumeration over 
ordered count quadruples with entries from 1 to 60 finds that 2.55% can express all four targets 
within this coefficient class. These are conditional counts of formula and input flexibility; the classes 
and wider-coefficient sensitivity are specified in Supplementary Notes 5 and 11.
3.3. The fixed stellar module returns the same ordered values
The primary J2000 stellar descriptors are g=10.488°, j=34.078° and v=62.160°, rounding to 
(10,34,62). Propagating stellar positions to 2965.2 BCE while retaining the fixed J2000 ecliptic frame 
gives (10.485°,34.075°,62.141°), with the same rounded tuple. With material r=16, these descriptors 
supply the same Q and inverse. Fixed-plane stability is retained across the declared 10,000-year 
window under both archived motion prescriptions (Supplementary Notes 24 and 33).
A second, explicitly conditioned route uses contrasts. Given D=4, q=24 and d=28, the compiler gives 
A=q+d+D=56 and B−C=q−D=20. Adding r=D²=16 gives B+C=40, hence B=30 and C=10. This 
recovers the inventory for both tested absolute tuples, (10,34,62) and (11,35,63). Without rounding 
the ancient contrasts, the same inventory uniquely minimizes the constrained squared error: RMS 
0.128°, compared with 0.686° for the next inventory (Supplementary Notes 28 and 33).
Historical catalogues provide a further portability comparison. Their Cygnus figures retain an 
approximately 16° minimum-cover radial scale; the Ulugh Beg values also round to (10,34,62), 
whereas Ptolemy's values produce a non-target inverse [30]. The detailed catalogue readings, 
modern coordinate definitions and provenance of the prior grammar are preserved together in 
Supplementary Table S53.1 and Notes 7–8. Here the radial seed remains a material input.
3.4. The material word ranks first and places the temporal minimum near the early monument
The inventory-derived word W=(28,72,16) ranks first by temporal fit among all 512 words in the 
declared material library, before applying an archaeological interval. Across the full 10,000-year 
search, its minimum occurs at 2965.2 BCE, with executed vector (27.785,72.653,16.233) sidereal 
minutes and RMS 0.419 minute, or 0.105°. The runner-up, (28,68,28), gives RMS 0.318° near 
1744.1 BCE, approximately three times the primary RMS.
Three library words have minima in the declared 3000–2920 BCE initial-enclosure window; only W 
also has a fit at least as close as the primary result. The six-dimensional space-motion variant gives 
2965.9 BCE. Figure 6 displays the complete search window and the archaeological comparisons 
(Supplementary Note 9).
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504

Stonehenge | v1.39 | 19
Figure 6. An astronomical minimum compatible with radiocarbon chronology without a chronological 
fitting constraint. (A) The complete temporal score across the search window, with the refined 
minimum marked at 2965.2 BCE. (B) The same minimum compared with Darvill et al.’s initial-enclosure 
interval within Stage 1 and Greaney et al.’s 95% and 68% ditch-construction intervals. It falls within the 
first two intervals and precedes the narrower third interval. Archaeological intervals enter the 
comparison after astronomical ranking; they do not constrain the minimization. Horizontal bars show 
the published intervals, not probability-density curves.
The astronomical minimum falls within published radiocarbon-based intervals for the early 
monument, although those intervals do not enter the numerical minimization. It lies within Darvill et 
al.'s 3000–2920 cal BCE range for initial enclosure construction within Stage 1 [2] and Greaney et 
al.'s revised 2980–2720 cal BCE 95% ditch interval, but precedes the latter's 2940–2800 cal BCE 
68% range [18]. These chronological models provide an external comparison with an early phase, 
rather than a date for the full input inventory. Any material system producing the same word under 
the specified operator would receive the same astronomical phase.
A retrospective solar comparison concerns the ideal cyclic construction. Its normalized long chord, 
26 sin(21π/56)=24.0209, is close numerically to the 24.0163° obliquity at the stellar phase. Under 
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521

Stonehenge | v1.39 | 20
the degree-identification convention, solar inversion gives 3016.7 BCE in the primary model and 
2999.4 BCE in the Laskar sensitivity model. Of 416 disclosed cyclic arms, three intersect the 
obliquity range in the full window; only the selected arm gives a solution near the stellar phase 
(Supplementary Notes 10 and 25).
3.5. Construction-phase dependency audit
Table 3. Construction-phase classification of inputs and held-out dimensions. Darvill et al.'s 
intervals are identified by event and source [2] (Table 3); the broader English Heritage dates 
are contextual summaries. These are external archaeological ranges, not uncertainty 
intervals from this model. Residual partitions retain the same outer-only scale (Table 2).
Feature and model 
role
Archaeological phase
Consequence for the model
Aubrey circuit; A=56, 
diameter
Initial enclosure: 3000–
2920 cal BCE in Darvill 
et al. [2] (Table 3); 
revised ditch ranges 
compared separately 
[18].
Only securely early count among A, B, C and D; 
its diameter alone does not calibrate the Station 
frame.
Station positions; D=4, 
short and long sides
Insecurely dated; 
assigned to Stage 2 
(2620–2480 cal BCE) in 
the reconstruction of 
Darvill et al. [2] (Table 
3).
Required for the 5:12:13 calibration and the four-
count compiler; a date near the initial enclosure is 
unproved.
Sarsen circle and 
trilithons; B=30, C=10; 
two inner dimensions
Stage 2: 2620–2480 cal 
BCE [2] (Table 3); 
commonly summarized 
as c. 2500 BCE.
Two held-out roles: 2.05% RMS under the scale 
calibrated from Station–Aubrey dimensions.
Y and Z hole circuits; 
two inner diameters
Y/Z excavation: 1630–
1520 cal BCE [2] (Table 
3); broader heritage 
summary, 1800–1500 
BCE.
Two held-out roles: 1.99% RMS under that same 
scale; their agreement cannot independently date 
an early plan.
The earthwork and Aubrey circuit belong to the early monument around 3000 BCE. Darvill et al.'s 
chronology assigns the major sarsen settings and Stations to Stage 2, 2620–2480 cal BCE, and Y/Z 
excavation to 1630–1520 cal BCE [2] (Table 3). The Station dating remains insecure, so their 
placement within that reconstruction is not an independent determination for each setting [1–3]. 
English Heritage summarizes the central stones as about 2500 BCE and the late holes more broadly 
as 1800–1500 BCE [35]. Thus A=56 is the only securely early count among the four inputs at the 
nominal astronomical minimum. The Station-dependent calibration and all four inner comparisons 
cross phases (Table 3; Figure 7). These dates constrain structures; they do not date an otherwise 
undocumented design.
The temporal word alone also lacks a unique material inverse. Its map is W=(A−B+C−2D, 
3(B+D−C), A−B−C), a rank-three linear map with kernel generated by (1,0,1,1). Every positive 
integer inventory (56+t,30,10+t,4+t), t≥−3, gives the same W. With W fixed, the additional closure 
r=D² selects D=4 and t=0. This does not contradict the uniqueness theorem under all three closures; 
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543

Stonehenge | v1.39 | 21
it identifies the extra assumptions that provide uniqueness. The theorem supplies conditional 
mathematical identification, while independent archaeological evidence is needed to assign those 
assumptions to an early phase. A free common timing scale further makes proportional words 
indistinguishable under E (S5 Appendix, section 5).
Figure 7. Construction-phase dependency of the model roles. The Aubrey circuit belongs to the 
earliest monument; the sarsen settings and perhaps the Stations are later, and the Y/Z circuits later 
still. The figure locates features used in the mathematical reconstruction and does not imply that they 
were constructed together or that the proposed information was transmitted between phases.
3.6. One construction connects the two primitive triangles
The pivot transformation converts the Station frame into the doubled 8:15:17 triangle through three 
length operations:
T(10,24,26)=(26−10,(10+24+26)/2,10+24)=(16,30,34).
Requiring both input and output of this transformation to be right triangles forces the input ratio 
y/x=12/5. Its operations transfer the diagonal-minus-short-side difference, the semiperimeter and the 
sum of the legs. The input ratio is therefore required by the simultaneous right-triangle conditions, 
linking the two primitive families through one specified construction (Figure 8; Supplementary Note 
36).
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561

Stonehenge | v1.39 | 22
Figure 8. One transformation connects the two primitive triangles. The 10:24:26 right triangle supplies 
three operations that produce 16:30:34. Its pivot angles round to 28° and 62°, and their difference 
rounds to 34°. The inverse exradii recover the initial legs. All dimensions are model units, and both 
diagrams use local schematic axes. Approximate readings for the 7|21|7|21 chord model are retained 
separately.
The resulting acute angles are 28.072° and 61.928°, and their difference is 33.855°. They round to 
28°, 62° and 34° under the original rule. The ideal 7|21|7|21 version retains the 62/34 angular pair 
when its actual transformed triangle is used. A circle-based inverse returns the initial legs: the 
exradii opposite sides 16 and 30 are 10 and 24. Equal-area relations, Gaussian constructions and 
the distinct uses of 34 as a length and an angle label are derived in Supplementary Notes 36 and 
52.
These constructions produce further quantities that can be compared with the plan. The four 
incentres of the Station rectangle's triangular halves have a circumdiameter of 53.72 m at the 
existing scale, close to the published 54 m Y-ring summary. The ordered paths J=x+y and V=2z+x 
realize 34 and 62; their two positive rectangular solutions are (10,24,26) and (78/7,160/7,178/7). 
Only the first returns integer inventories in the fixed module, and its external metric RMS is 0.26% 
against 7.93% for the other solution (Supplementary Notes 37 and 39).
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579

Stonehenge | v1.39 | 23
A separate inventory-angle comparison gives arctan(30/56)=28.179°, only 0.096° from the 
recomputed stellar difference v−j=28.083°. The Aubrey–sarsen pair ranks first among all six 
inventory pairs under this same function. The derived ditch centreline is 32.12–32.19u under its 
declared boundary convention. These auxiliary comparisons, the complete tangent ranking and the 
modern constellation-boundary diagnostics are retained in Supplementary Notes 15–19 and 52.
3.7. Outer-frame paths connect lengths to architectural directions
Under the same equal-tension rule, the J route joins the midpoints of opposite short sides via one 
long edge, giving 5+24+5=34 in the ideal frame. The V route follows diagonal–short side–diagonal, 
giving 26+10+26=62. Under equal terminal tensions, these paths produce resultants along the short 
and long axes, respectively. Each channel thus has a length and a direction supplied by the same 
route. Applied to the Rees proxies, their axial azimuths are 52.15° and 140.04°: oblique axes with 
east–west and north–south predominance in the grid. The two channels therefore have distinct 
directional consequences under one stated operation.
The entrance can select these route families within the alternative directional protocol. J maximizes 
alignment with the normal to the 1–30 opening; V maximizes alignment with that normal’s 
perpendicular. In both Rees conventions, the selected families have ideal-frame lengths 34 and 62 
when their lengths are inspected after selection. This gives the entrance an explicit role in choosing 
which routes are read (Supplementary Note 40).
The two specifications exchange inputs and outputs. Under the entrance-based selector, the J–
entrance alignment is the selection criterion and the ideal-frame lengths 34 and 62 are its outcome; 
under the length-based enumeration, the length labels and orthogonality are inputs and the 
alignments reported below are outcomes. Both read the same plan geometry and are not 
independent confirmations.
Under the length-based enumeration, the J path, labelled 34, gives an axis only 0.45° from the 
entrance normal in the primary digitization (axial Gram 0.99994). The V path, labelled 62, has mean 
Gram 0.93891 with the three available inner opening normals. These axes come from the specified 
outer-frame path decoder and are compared with the plan without fitting a rotation. In the indicative 
Station 91 branch the scores are 0.99988 and 0.93802 (Figure 9; Supplementary Note 54). The 
inner-opening values average the three individual squared alignments; they do not score a single 
averaged normal.
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609

Stonehenge | v1.39 | 24
Figure 9. The same outer-frame routes supply lengths and architectural directions. (A) The contour 
route gives J=5+24+5=34; the crossed route gives V=26+10+26=62. Arrows show their equal-tension 
terminal resultants in local schematic axes. (B) The corresponding axes are compared with the 
entrance and inner openings in the archived Rees grid. Dashed axes pass through the Station centroid 
for direction only; solid coloured segments mark opening normals. Their positions are not predicted 
force lines. (C) Squared cross-Gram entries for the entrance and the mean across three inner 
openings: 1 denotes parallel axes and 0 perpendicular axes. Near-orthogonal channels give 
approximately complementary entries within each column. (D) Assigned alignments in both Station 91 
conventions, without a fitted rotation. The 34/62 labels come from the specified paths; the direction-
only scores do not identify their numerical labels.
The compact cross-Gram retains scores 0.0684 for J against the inner normals and 0.00084 for V 
against the entrance. The assigned and exchanged aggregate scores are 0.9694 and 0.0346. For 
exactly orthogonal channels, the exchanged score is identically 1 minus the assigned score. The 
digitized axes are nearly orthogonal: the two scores sum to 1.0040 in the primary branch and 1.0014 
in the indicative branch. Their contrast therefore restates the alignment rather than providing an 
independent control (Supplementary Note 54).
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626

Stonehenge | v1.39 | 25
The two channels also restrict each other within the declared route family. Two of its 12 ordered 
assignments recover the 34/62 length labels; only the short-side-midpoint J route paired with the 
diagonal–short-side–diagonal V route also produces orthogonal resultants. The alternative route of 
length 34 produces an axis parallel to V. In the digitized primary branch, the J–V axial Gram is 
0.00136 for the selected pair and 0.99995 for that alternative. The ideal 7|21|7|21 form retains the 
selected pair, giving lengths 33.971 and 61.950, which round to 34 and 62 (Supplementary Note 54).
The three complete trilithon openings have normals closer to the Station long-side family than to its 
short-side family. Squared axial similarities are 0.984, 0.952 and 0.883 for the long sides, against 
0.029, 0.032 and 0.146 for the short sides. No rotation is fitted, and the preference survives both 
Station 91 conventions and the stipulated ±0.25 m coordinate perturbations (Supplementary Note 
35).
The Station-only construction predicts an axis 0.61° from the entrance normal, placing that pair first 
among 19 consecutive proxy pairs. With grid east fixed, the pivot rays approximate two trilithon-
opening normals at 1.18° RMS; the discrete model gives 1.13°. References, reflections and 
assignments were selected retrospectively and are retained in full (Supplementary Notes 38–41).
The exhaustive 220-triangle proxy census finds 52–54–60 closest to 16:30:34 at 0.69% RMS with its 
own fitted scale. Across seven primitive comparators, 5:12:13 and 8:15:17 give the two smallest 
best-fit discrepancies, 0.56% and 0.69%. A different census retains 541 of 252,000 shape–
orientation configurations under all six criteria, every survivor having the 7|21|7|21 form. The 
criterion-removal analysis shows that four restrictions select the same shape; direction adds a 
restriction, while the horizontal/vertical label adds none after it (Supplementary Notes 42–43).
These material and astronomical relations raise a further question: which parts of the reconstruction 
survive alternative assignments, inputs and scales? The next sections test stellar selection, early-
versus-later phase preference, continuous metric targets and joint spatial–temporal performance. 
The closing sensitivity analysis identifies which physical readings depend on the source or 
convention.
3.8. Spatially filtered stellar assignments concentrate in Cygnus
All seven temporal survivors in the principal V≤5 census lie in Cygnus. The expanded Hipparcos 
search supplies neither a named anchor nor Cygnus as input: 1,625 eligible stars across 77 regions 
yield 2,990 spatial candidates, of which seven reach or improve on the nominal temporal 
benchmark. Applying the ancient-contrast and Stage 1 filters jointly retains only Deneb–Sadr–
Albireo–Aljanah–Fawaris in the declared roles. At V≤6, 1,072,375 spatial candidates yield 903 
temporal passes and six joint passes, again all in Cygnus and all retaining Deneb and Sadr (Figure 
10).
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660

Stonehenge | v1.39 | 26
Figure 10. The joint stellar solutions concentrate in Cygnus and the nominal configuration is 
geometrically observable. Panel A reports the nested magnitude censuses after spatial and temporal 
filtering and after the additional joint contrast and Stage 1 filters. Panel B shows the full interval of 
magnitude limits for which the nominal quintet or ablated core is uniquely retained under its 
respective test. Panel C shows simultaneous stellar altitudes in an admissible astronomical-night 
configuration. The census uses uniform Hipparcos astrometry, and the night test allows season and 
sidereal time to vary.
The nominal quintet is brighter than every other joint survivor in at least one differing role and is 
never fainter in a corresponding role. Its role-ordered magnitudes are (1.25,2.23,3.05,2.48,2.86). It 
remains the unique assignment satisfying the unchanged joint criteria throughout the magnitude-limit 
interval 3.05≤V<5.06 (Supplementary Notes 45–46).
Removing the latitude descriptor and third temporal channel leaves the Deneb–Sadr–Albireo core as 
the only temporal survivor at V≤4. This core uniqueness spans limits 3.05≤V<4.59 before ancient-
contrast or horizon filters. There are five temporal survivors at V≤5 and 131 at V≤6. The core 
minimum is near 2848.4 BCE, compared with 2965.4 BCE for the full graph in the uniform catalogue 
convention, showing the third channel's contribution to the full phase (Supplementary Note 47).
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677

Stonehenge | v1.39 | 27
All five selected stars can be simultaneously above 63° during astronomical night at the nominal 
uniform-catalogue phase. The flat-horizon annual-night calculation finds this configuration while the 
solar centre is below −18°. The same rule excludes two V≤5 core competitors containing stars that 
never rise at the site at their fitted epochs. This establishes a geometrically possible observing 
configuration and a physical distinction among candidates (Supplementary Note 48).
Earlier anchored censuses, both historical membership lists and every alternative assignment 
remain in the supplement. The expanded comparisons use Hipparcos positions propagated from 
catalogue epoch J1991.25, with nominal minimum 2965.4 BCE and RMS 0.390 minute. These are 
common-input comparisons; the primary archived specification remains 2965.2 BCE and 0.419 
minute (Supplementary Notes 18, 27, 34 and 44–48).
3.9. Fixed-word selection and alternatives under released assumptions
With W=(28,72,16) and its original timing scale fixed, nominal Cygnus ranks first among all 5,280 
role classes in the declared five-brightest regional panel, without spatial-tuple or archaeological 
chronology filtering. Its RMS is 0.390 minute; runner-up Reticulum gives 0.554 minute at 2848.6 
BCE, 1.42 times the Cygnus RMS. This is sole first place within that panel, not a pass against an 
independently specified absolute threshold or a search of all stellar subsets (S5 Appendix, section 
3.1). Expressed directly in degrees, the same Cygnus minimum has RMS 0.097532° against 
θ=(7°,18°,4°). The first-place ranking and the 1.42 runner-up ratio are identical under the angular 
formulation.
The crossed bright-star panel yields 336 qualifying classes after exchanging the final pair, equivalent 
to 672 labelled role-word descriptions across 57 regions. The common-catalogue Cygnus 
benchmark is 0.390 sidereal minute near 2965.4 BCE. Six classes have their best qualifying word-
specific minima in 3000–2920 BCE: four Cygnus descriptions and two first/third-role reversals in 
Lacerta. The Lacerta example has W=(32,32,28), RMS 0.138 minute and minimum near 2921.5 
BCE. The best timing fit anywhere in the panel is a Scutum description with W=(16,4,4), RMS 0.019 
minute near 3237.2 BCE (Table 4). These alternatives are numerical comparators, not proposed 
ancient identifications. Their words, however, cannot arise from the original integer compiler: neither 
32 nor 4 in the second channel can equal 3q for integer q. They are alternatives under word 
freedom, not counterexamples under the unchanged material map.
Of the 512 words, 55 admit some positive integer inventory under that map, defining 290,400 role-
word classes. Algebraic feasibility retains 26 of the 336 timing passes. Only nominal Cygnus has its 
best qualifying minimum in 3000–2920 BCE. In the broader 2980–2720 cal BCE 95% ditch interval 
[18], three feasible classes remain: Cygnus and two Vela assignments with words (32,72,68) and 
(40,72,68), minima at 2905.8 and 2884.3 BCE, and RMS 0.365 and 0.276 minute. Both Vela 
assignments lie within the 2940–2800 cal BCE 68% interval, which excludes the unconstrained 
Cygnus optimum. Vela's r=68 cannot satisfy r=D² for integer D, so these are not full three-closure 
alternatives. All three chronology windows are reported rather than choosing the one with the 
strongest selectivity.
The free-scale primary Cygnus solution occurs near 2986.8 BCE with relative dispersion E=0.682%; 
the six-dimensional variant gives 2987.5 BCE and E=0.682%. In the common-catalogue 
comparison, Cygnus gives E=0.649% near 2985.5 BCE. Thirty-eight classes improve that value, 
placing Cygnus 39th; the 39 matching-or-better classes span 25 regions; Draco gives E=0.0206% 
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719

Stonehenge | v1.39 | 28
near 3052.0 BCE. Both grid resolutions recover identical qualifying candidate sets. The primary 
branch survives this scale change, but its exclusivity does not. Neither this panel nor the earlier 
separately conditioned censuses supplies a global chance probability or an equal-freedom 
comparison of complete archaeological models (S5 Appendix, sections 3–4).
Table 4. Comparison universes, fixed assumptions and released choices. All searches retain their 
stated stellar definitions and propagation model. Global optimization covers astronomical years 
−8000 to +2000. Absolute RMS and relative dispersion E are distinct objectives; their scores and 
ranks are not interchangeable. A “pass” uses the corresponding Cygnus benchmark, not an 
independent significance threshold. Interval membership after fitting and optimization within an 
interval are distinguished below.
The three constrained comparison windows are 3000–2920, 2980–2720 and 2940–2800 BCE, in 
that order; Stage 2 is 2620–2480 BCE. “Global” results do not use those intervals as fitting bounds. 
The complete panel definitions and adverse outcomes remain in S4 and S5.
Comparison 
and universe
Held fixed and varied
Objective and 
chronology
Result and location
S4 principal 
census
Within-region 
subsets, V≤5
Fixed spatial tuple, W and 
timing scale; stellar identities 
and role assignments vary.
RMS; global epochs first. 
Ancient-contrast and 
initial-window filters 
applied afterward.
2,990 spatial candidates; seven 
temporal passes, all in Cygnus. 
Joint filters retain the nominal 
class (§3.8).
Fixed-word panel
5,280 role 
classes
Fixed W=(28,72,16) and 
timing scale; roles vary among 
five brightest stars in each of 
88 regions.
RMS; no spatial-tuple or 
archaeological 
chronology filter.
Cygnus rank 1: 0.390126 min, 
2965.4 BCE. Reticulum: 0.554285 
min, 1.42× (§3.9).
Integer-feasible 
words
290,400 classes
Same bright-star panel and 
timing scale; 55 words vary 
under unchanged compiler 
feasibility, without all three 
closures.
RMS; global fitting, then 
interval membership. No 
spatial-tuple filter.
26 passes; one in 3000–2920 
BCE and three in 2980–2720 BCE 
(§3.9).
Unrestricted 
library
2,703,360 
classes
Fixed bright-star panel and 
timing scale; roles and all 512 
library words vary. Compiler 
feasibility not imposed.
RMS; global fitting, then 
interval membership. No 
spatial-tuple filter.
336 passes in 57 regions; six 
classes in 3000–2920 BCE (§3.9).
Free timing scale
5,280 role 
classes
Fixed W and bright-star panel; 
role assignments vary and the 
common timing scale is fitted.
E; no spatial filter. Global 
search and separate 
interval-constrained 
searches.
Global: 39 classes match or 
improve Cygnus in 25 regions. 
Cygnus ranks 2, 2 and 4 in the 
three constrained windows (§§3.9
–3.10).
Nominated-input 
sensitivity
Three 
astrometric 
branches
Fixed Cygnus identities, roles 
and W; timing scale fitted. 
Primary, six-dimensional and 
common-catalogue inputs.
E; global search without 
archaeological 
chronology bounds.
Optima 2986.8, 2987.5 and 
2985.5 BCE; E=0.6819%, 
0.6817% and 0.6490%. These 
inputs do not rank a stellar panel.
720
721
722
723
724
725
726
727
728
729
730
731
732

Stonehenge | v1.39 | 29
Comparison 
and universe
Held fixed and varied
Objective and 
chronology
Result and location
Phase 
comparison
Specified 
Cygnus model
Fixed word and roles. Original 
scale with channel-weight 
checks; separate tests fit 
scale under specified losses.
Global versus interval-
constrained minima. 
Chronology explicitly 
bounds the latter.
Stage 2/original global RMS: 
5.33× in the common catalogue. 
Free-E ratios: 15.31–16.06× 
across inputs (§3.10).
Metric-word 
extension
Five targets, 
same panel
Fixed construction and 
calibration coefficients; targets 
derive from all five declared 
exterior conventions. Roles 
vary.
Fixed-scale RMS and 
free-scale E, separately; 
global and constrained 
searches.
Fixed-scale global rank 1 in all 
five variants, with optima in 3000–
2920 BCE. Free-scale global 
ranks 10–49 (§3.11).
Joint space and 
time
5,280 classes; 
10,560 typed 
assignments
Fixed S, W, reference and 
panel. Independent positive 
spatial and temporal scales; 
both final-pair spatial 
orientations.
Spatial CV or minimax 
with temporal E; no 
aggregate weight. Global 
and constrained 
comparisons.
No class dominates Cygnus. 
Global frontiers contain eight or 
six classes; no unique winner 
(§3.12).
3.10. The fixed model discriminates construction phases
The fixed Cygnus model favours the early epoch over the later sarsen phase: the best fit in that 
phase has 5.33 times the RMS of the unrestricted minimum. In the common catalogue, the 
chronology-free optimum is 2965.4 BCE at 0.390 minute RMS. Restricting the same model to 2620–
2480 BCE gives its lowest RMS at the oldest boundary, 2620 BCE: 2.080 minutes. The primary and 
six-dimensional inputs give corresponding ratios of 4.97 and 4.96 (Figure 11A). These input variants 
test astrometric sensitivity of one model, not independent archaeological confirmations.
The early epoch also outperforms Stage 2 channel by channel. At 2965.4 BCE, the common-
catalogue absolute residuals are (0.211,0.605,0.213) minutes; the respective minima found 
anywhere in 2620–2480 BCE are (0.742,0.728,3.449), all at 2620 BCE. Every channel therefore 
favours the early reference date. Given these numerical interval bounds, any nonnegative channel 
weights with positive total preserve its lower weighted RMS, including when a channel is omitted. 
The primary and six-dimensional inputs give the same ordering. This establishes robustness of the 
phase comparison to weights, while allowing the weighted optimum to move to another epoch; it 
does not establish a new stellar ranking under changed weights (S5 Appendix, section 3.4). In the 
common catalogue, the narrowest margin is the Sadr–Albireo channel, 0.605 versus 0.728 minute; 
across all three input branches, the narrowest ratio is 1.040 in the six-dimensional branch.
The nominal optimum remains outside Greaney et al.’s 2940–2800 cal BCE 68% ditch interval. A 
separate constrained-epoch test gives Cygnus RMS 0.418 minute at its oldest boundary, 2940 BCE, 
only 7.20% above the unrestricted score (Figure 11B). Cygnus remains first among the same 5,280 
fixed-word classes: every other class has an archived unrestricted minimum of at least 0.554 minute, 
and restricting its epoch cannot lower that minimum. Reticulum attains this bound at 2848.6 BCE, 
inside the interval, so it remains second with 1.33 times the constrained Cygnus RMS. This follows 
from the archived global scores and the new Cygnus bounded fit; it does not place the unrestricted 
optimum inside the narrower interval.
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758

Stonehenge | v1.39 | 30
Figure 11. Temporal discrimination under the unchanged word and timing scale. Both panels 
use the common-catalogue absolute-RMS objective. (A) The unrestricted optimum falls in the 
initial-enclosure interval; the smallest error permitted in Stage 2 is 5.33 times the unrestricted 
minimum. (B) Restricting the epoch to the 68% ditch interval moves the Cygnus optimum to 
2940 BCE while retaining a score below the best unrestricted score of every competing class. 
Reticulum’s minimum lies within that interval, fixing its second-place score. Shaded bands 
show archaeological intervals, not probability densities. These retrospective constraints do 
not replace the chronology-free search in Figure 6.
The constrained ranking shows that an optimum can fall outside an archaeological interval while 
retaining comparative fit inside it. It does not relocate the original optimum or remove the free-word 
and free-scale alternatives in section 3.9 (S5 Appendix, section 3.3).
The preference for the early branch also survives fitting a common timing scale under the tested 
losses. Across the primary, six-dimensional and common-catalogue inputs, the best Stage 2 relative 
dispersion E is 15.31–16.06 times the global minimum. In the common catalogue, even the best 
Stage 2 epoch and minimax scale leave at least one channel discrepant by 11.06%, whereas an 
epoch near 2998 BCE permits all three channels to agree within 0.794%. These results compare 
phases within the specified Cygnus model. They do not preserve its first-place temporal ranking 
when scale is freed: interval-constrained free-scale searches rank it second, second and fourth in 
the initial-enclosure, 95% and 68% windows, respectively (S5, sections 9–10).
3.11. Continuous exterior measures preserve fixed-scale selection
The fixed-scale selection does not require the target to equal the exact integer word. Applying the 
specified construction to continuous exterior measures leaves Cygnus first among the same 5,280 
role classes in all five declared variants (Table 5). The principal Aubrey convention gives 
Wmetric=(27.852,72.127,15.911), RMS 0.323 minute and an optimum at 2975.9 BCE. The five 
global optima span 2992.6–2960.6 BCE, all within 3000–2920 BCE, without using that interval in the 
global ranking (S5, section 12).
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785

Stonehenge | v1.39 | 31
Table 5. Fixed-scale stellar selection under all five continuous metric conventions. Every row 
uses the common catalogue and the original numerical timing convention. The Aubrey 
diameters serve as diagonal proxies in the declared construction; the Euclidean diagonal is 
calculated from the two reported side means. Rank is within the 5,280-class panel. The range 
across variants is a sensitivity, not a confidence interval.
Exterior convention
Optimum BCE
RMS min
Rank
Aubrey 86.44 m
2975.885
0.322947
1
Aubrey 86.87 m
2964.122
0.424256
1
Aubrey 87.00 m
2960.577
0.459085
1
Aubrey 85.83168 m
2992.625
0.257887
1
Computed diagonal 86.79348 m
2966.211
0.404460
1
The five variants share measured inputs and the original construction. Their free-scale global ranks 
range from 10 to 49. All five fixed-scale global optima lie outside the 2940–2800 BCE interval; 
restricting to that interval reduces the MacKie variant to second place. The result establishes stability 
under the examined metric changes at the original timing scale, not under every timing convention.
3.12. Joint spatial and temporal comparison with separate scales
Cygnus remains nondominated in the joint spatial–temporal comparison when the two positive 
scales are fitted independently. Its spatial relative dispersion is 2.146% and its spatial minimax 
fractional error is 2.267%; its global temporal E is 0.649%. Cygnus belongs to frontiers of eight and 
six classes, respectively. It remains nondominated under both spatial criteria in each of the three 
previously examined archaeological windows (S5, section 11).
The spatial criterion alone ranks Cygnus second: a Vulpecula assignment improves spatial 
dispersion to 1.064% but has temporal E=4.075%. Conversely, candidates with lower temporal E 
worsen the spatial fit. This trade-off is the result of the joint comparison; nondominance does not 
identify a sole winner or supply a historical probability. The targets, stellar roles, J2000 reference 
and regional star selection remain fixed, and the diagnostic was formulated retrospectively after the 
separate rankings were known.
3.13. Source and model sensitivities
The Aubrey-anchored assignment in the primary Cleal branch gives 8|21|7|20; exact registration 
with individual Aubrey positions remains unresolved. The continuous-gap and fitted-shape results 
above concern different observables. The tested eight-sector construction procedure and seven-hole 
residual model also fail to improve cross-validation over the regular-circle baseline (Supplementary 
Notes 20 and 30).
In the ancient mean ecliptic of date, the absolute stellar tuple is (11,35,63), and the unchanged 
inverse with r=16 returns (57,30,11,5). The rounded contrasts remain (24,28) at the tested phase. 
Fixed-J2000 stability therefore does not establish the same absolute register relative to the 
prehistoric ecliptic (Supplementary Notes 24 and 27). Changing the ideal cyclic solar input to the 
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816

Stonehenge | v1.39 | 32
exact 5:12:13 ratio or published mean-side ratio gives 2784.8 or 2713.2 BCE, respectively 
(Supplementary Note 25).
Several physical realizations are source-sensitive or unsupported. The four-incentre assignment to 
the 30 digitized Y symbols has 2.87 m RMS, with 97.13% of tested rotations matching as well or 
better. The path's equivalent force line crosses Stone 1 in the primary branch and clears the 
entrance only with indicative Station 91. The best 16:30:34 point triangle has a predominantly north–
south long side, contrary to the proposed horizontal-long-side assignment. The ditch centreline is 
derived from width summaries; alternative boundaries do not preserve its integer reading 
(Supplementary Notes 15, 37, 40 and 42).
Selection depends on the declared comparison universe. Releasing Deneb in the historical contrast-
only census leaves two assignments common to both lists. Among arbitrary positive integer words, 
179 fit the nominated sky graph as closely as W. Its rank within the 512-word material library 
conditions on that graph. S5 separately reports unrestricted words, words feasible under the original 
integer compiler, and the original word held fixed. The fixed-word comparison ranks Cygnus first 
across its bright-star role panel without an archaeological chronology filter. These observations 
concern different candidate universes; none supplies a global historical chance probability 
(Supplementary Notes 9, 44–46 and 52; S5 Appendix).
3.14. Geometric recovery of temporal roles
Cygnus is the only four-arm star-shaped minimum spanning tree in each 88-quintet catalogue panel. 
The geometric rule selects Deneb–Sadr–Albireo and leaves Aljanah–Fawaris as the remaining pair, 
recovering the existing temporal roles. At J2000 in the common Hipparcos catalogue, Albireo makes 
a 168.702° bearing angle from Deneb at Sadr, compared with 107.449° for Aljanah and 84.895° for 
Fawaris. No selected role changes between J1991.25 and J2000 in the 88 Hipparcos groups. 
Recomputing the clock with the recovered roles gives 2965.425 BCE and 0.390126 sidereal minutes 
RMS, reproducing the common-catalogue result in section 3.9. The original 5,280-class comparison 
and its 1.42 runner-up ratio remain the fixed-unit temporal benchmark.
At that epoch, all six kinematic scenarios retain Sadr as a certified equal-weight spherical median of 
both quintet and selected trio, the four-arm tree and the same role assignment. The proportional 
change of the trio’s angular shape relative to J2000 is 0.0394–0.0472%. Across 1,501 annual 
epochs per scenario from 3000 to 1500 BCE, all 9,006 sampled model cases preserve both median 
certificates, the tree and the selected continuation; the minimum tree margin is 0.3990°. The 
additional 88-group checks at 3000 BCE, the existing optimum and 1500 BCE also leave Cygnus as 
the sole four-arm case. S6 gives the certificates, scenario definitions, complete comparison panels 
and secondary Galactic-axis diagnostic.
3.15. Dimensionless selection under a local horizon condition
With one geometrically selected assignment per group and no horizon or chronology restriction, 
Cygnus ranks second among 88 groups. Its minimum is E=0.649026% near 2985.482 BCE. Tucana 
ranks first at E=0.134236% near 3237.025 BCE, followed by Norma at 2.801116% near 7314.476 
BCE. The Cygnus right-ascension intervals are (6.954903°,18.170633°,4.006198°); their common 
numerical scale is discarded. Every one of the 88 unrestricted scores reproduces its S5 free-scale 
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856

Stonehenge | v1.39 | 33
value, and both search-grid resolutions recover the same minima. Table 6 distinguishes this panel 
from the unrestricted 5,280-role comparison in section 3.9.
Table 6. Leading dimensionless fits after the same geometric role rule. The last column 
reports the best score after reoptimization over epochs satisfying the necessary horizon 
condition. Scores are percentages; smaller is better. The search uses no archaeological date 
filter.
Group
Global E (%)
Global epoch (BCE)
Horizon E (%)
Tucana
0.134236
3237.025
None
Cygnus
0.649026
2985.482
0.649026
Norma
2.801116
7314.476
2.801116
Aquila
4.360427
2496.611
4.360427
Draco
4.623273
1299.828
4.623273
Seventy-one groups have at least one admissible epoch under the necessary horizon condition. 
Cygnus ranks first after reoptimization within all such epoch segments; Norma is second. Tucana 
has no admissible segment. Four of its five stars remain below the geometric horizon throughout the 
numerical search, with refined maximum upper culminations of −7.336°, −15.700°, −13.178° and 
−3.835°. At its best temporal fit, all five stars have negative upper culminations. Cygnus, in contrast, 
permits all five stars simultaneously above 64.719° near its optimum, at an appropriate sidereal time. 
The local-observation ranking therefore follows a common physical requirement rather than a 
chronology selection.
The dimensionless Cygnus optimum lies within the 3000–2920 BCE initial-enclosure interval, 
although it is just older than the separate 2980–2720 cal BCE 95% ditch interval. Within the later 
2620–2480 BCE sarsen phase, its smallest E is 10.422720%, 16.059 times the global minimum. 
This reproduces the S5 free-scale phase contrast. The 2985.482 BCE optimum and the earlier fixed-
unit 2965.425 BCE optimum are outputs of different objectives, not independent dates or uncertainty 
bounds. Across the full 5,280-role panel, free-scale alternatives remain as reported in section 3.9; 
the geometric selector specifies which one role assignment per group is retained here. The model 
still requires archaeological justification of the count map and any proposed transmission across 
phases.
3.16. Fixed angular recovery without constellation boundaries
The nominal Hipparcos J2000 census contains 19,939 distinct quintets across 25,161 generating 
radius intervals. Under the fixed target (7°,18°,4°), Cygnus ranks first without requiring four-arm 
topology or horizon admissibility (Table 7). Its minimum R=0.097531516° near 2965.425 BCE is 
equivalent to 0.390126064 quarter-degree units. The runner-up, HIP 17440–16368–15968–13884–
16245 in role order, has R=0.214695196° near 2430.126 BCE, giving a 2.2013 error ratio. This 
group spans the modern regions Reticulum, Hydrus and Horologium, which were not inputs to 
enumeration. The result retains the fixed target and geometric roles while removing regional 
boundaries and a four-arm entry filter.
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888

Stonehenge | v1.39 | 34
Under the uniform necessary horizon condition, 17,860 groups have at least one admissible epoch. 
Reoptimization over those epochs retains Cygnus first with a 2.6260 runner-up error ratio. Merely 
checking visibility at each unconstrained fixed optimum would retain 13,910 groups and would 
implement a different test. The four-arm subset contains 283 quintets when any member may be the 
hub, or 256 when the hub must also be a generating cap centre. Both subsets retain fixed-scale first 
place with a 12.9151 error ratio; their runner-up, HIP 85670–87833–94779–86414–87585, reaches 
its minimum near 1833.145 CE. Two-grid refinement preserves all reported ranks and margins (S8, 
section 9).
Table 7. Complete nominal Hipparcos J2000 census and nested subsets. All searches span 
astronomical years −8000 to +2000. The final column is runner-up/Cygnus error when Cygnus 
ranks first; R is used for fixed-scale rows and E for free-scale rows. Only the explicitly 
labelled row imposes the horizon condition. These candidate counts and error ratios are not 
calibrated probabilities.
Objective and candidate family
Groups
Cygnus 
rank
Error ratio
Fixed target; all fields
19,939
1
2.201
Fixed target; horizon reoptimized
17,860
1
2.626
Fixed target; four arms, any hub
283
1
12.915
Fixed target; four arms, cap-centred hub
256
1
12.915
Free scale; all fields
19,939
68
—
Free scale; four arms, any hub
283
1
1.0155
Free scale; four arms, cap-centred hub
256
1
1.0155
Removing absolute angular scale from the same 19,939 assignments gives Cygnus E=0.649026% 
near 2985.482 BCE and rank 68. Both four-arm subsets retain first place, with nominal runner-up 
HIP 109268 at E=0.659091% near 1582.264 CE: an error ratio of only 1.0155. The fixed and free-
scale Cygnus optima differ by 20.06 years and both fall in 3000–2920 BCE. The nearby epochs 
demonstrate compatibility under these two objectives, while the rank change shows that 
identification in the complete field family depends on the fixed scale. They are neither independent 
dates nor uncertainty bounds.
The earlier four-arm census recovers Deneb–Sadr–Albireo with Aljanah–Fawaris as the residual 
pair. This membership persists for cap radii approximately 16.239°–20.284° around Sadr. Its four 
nominal catalogue/geometry panels and six targeted kinematic variants below concern E and the 
stricter cap-centred four-arm family (Table 8), rather than the new complete fixed-scale census.
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912

Stonehenge | v1.39 | 35
Table 8. Four nominal cap-centred four-arm ratio censuses without constellation boundaries. 
The date labels specify the geometry used to generate groups; every temporal search covers 
astronomical years −8000 to +2000. The horizon column counts groups with at least one 
admissible epoch. The final column compares the runner-up and Cygnus errors after uniform 
horizon reoptimization.
Catalogue / 
geometry
Four-arm 
fields
Horizon 
fields
Cygnus rank 
global / local
Local error ratio
Hipparcos / J2000
256
232
1 / 1
1.971
Hipparcos / 3000 BCE
267
243
1 / 1
3.890
HYG / J2000
255
231
1 / 1
1.679
HYG / 3000 BCE
261
237
1 / 1
3.888
Cygnus ranks first in all four nominal four-arm ratio panels, before and after the horizon condition 
(Table 8), and at each nested magnitude ceiling V≤4, 5 and 6. HYG returns E=0.649007% near 
2986.051 BCE. In the Hipparcos J2000 panel, 232 fields have some admissible epoch. Reoptimizing 
each within its admissible intervals leaves Cygnus first, with runner-up error 1.971 times larger. The 
corresponding local error ratios across the other three nominal panels range from 1.679 to 3.890. 
The horizon rule constrains physical observability; it does not impose an archaeological date 
window.
In the cap-centred four-arm ratio family, all six targeted S6 kinematic variants recover the same 
Cygnus membership and roles and preserve first place under the uniform horizon condition, with the 
smallest runner-up/Cygnus error ratio 1.579. The unrestricted ranking falls to second in the HGCA, 
Albireo systemic-motion and Deneb-distance variants. Their Cygnus optima span 2984.95–2986.97 
BCE, a model-sensitivity range rather than dating precision. Thus the locally admissible selection 
survives every tested kinematic variant, while the global nominal first place does not. The four 
nominal panels and six variants share catalogue inputs and are not independent confirmations (S8, 
sections 1–3).
3.17. Invariance and additional geometric controls
The timing intervals can be written using only scalar and triple products of stellar directions with the 
physical terrestrial pole. This removes the arbitrary right-ascension zero and coordinate basis while 
retaining the pole and its precession. The relative dispersion E is a monotonic transform of a least-
squares fit of a common inverse scale: F₂=E/√(1+E²). In the earlier 71-group horizon-admissible 
regional panel, its error margin preserves Cygnus first for the specified relative-residual Lp family 
and positive common quadratic weights with maximum/minimum ratio at most 18. Those bounds 
belong to that regional panel and are not transferred to the smaller margins of the new field census. 
Permitting all six target orderings instead places Lyra first and Cygnus second in the regional panel, 
showing the importance of the stated number-to-interval association (S8, sections 4–5).
Using only the published Station side means and their calculated Euclidean diagonal, the continuous 
construction recovers the four interior summaries with 2.014% relative RMS, without imposing 
5:12:13 or an Aubrey diameter. This ablation retains the feature associations and uses two exterior 
measures to determine shape and size. A separate spherical audit finds Sadr–Albireo nearly equal 
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946

Stonehenge | v1.39 | 36
to Fawaris–Aljanah, differing by 0.0342% at the nominal ratio epoch. Direct axis and centre-radius 
comparisons do not support a proportional copy of the Station rectangle, and exact spherical 
equality gives a different epoch. Inverting the continuous temporal construction produces a Cygnus 
rectangle aspect 6.49% above the published value at its nearer root. These exploratory diagnostics 
are reported with their alternative comparisons; none supplies an independent archaeological date 
or a new precise metric recovery (S8, sections 5–7).
4. Discussion
The reconstruction makes continuity of design at Stonehenge quantitatively testable. One exterior 
calibration recovers four inner dimensions at 2.02% RMS. A specified angular target places Cygnus 
first near 2965 BCE among 19,939 geometrically assigned fields without constellation boundaries, a 
four-arm entry filter or an archaeological date constraint. Freeing scale retains a nearby Cygnus 
optimum near 2985 BCE but reduces its complete-census rank to 68. These results separate epoch 
compatibility from scale-dependent identification. They define two archaeological questions: whether 
relationships between components persisted across building phases, and whether a later 
arrangement intentionally retained an earlier reference epoch. Historical attribution requires 
evidence connecting the numerical construction with practice and transmission.
Copying, halving and subtracting lengths, together with constructing inradii or exradii, provide 
possible setting-out operations for the model. This gives the dimensional correspondence a 
constructive interpretation: the model specifies operations by which a length in one component can 
be carried into another. The shared-width relation gives the most direct example. It identifies the 
outer rectangle's short side with the inner Sarsen diameter, then uses the same frame to supply 
three further lengths. The trilithon-opening comparison and declared path rules extend the 
reconstruction from dimensions to candidate axes.
The exterior constrains both size and direction in the tested reconstruction. The same outer frame 
supplies the inner dimensional ladder and, through specified paths of lengths 34 and 62, axes that 
approximate the entrance and inner-opening normals without a fitted rotation. The entrance-based 
specification makes the role of the opening explicit: its normal selects a contour family, and the 
perpendicular selects a crossed family, before their lengths are inspected. This selection reuses the 
reported plan alignments. Orthogonality alone does not force a high assigned alignment. The 
contribution is the explicit connection from lengths to directions across spatially separated 
components. It gives a shared-design or inheritance hypothesis a concrete relationship to explain 
and test against original settings (Figure 9; Supplementary Notes 40 and 54).
Measured lengths and the specified inventory closures recover the same primitive frame within the 
adopted definitions. Published metre measurements select 5:12:13 within the stated comparison 
class; the chosen closures recover its doubled form 10:24:26. The 7|21|7|21 model is the closest 
tested discrete expression on the 56-position circuit. Two approximately contemporaneous inner 
outputs, the sarsen envelope and trilithon span, have 2.05% RMS under the fixed outer scale. The 
later Y/Z pair has 1.99% RMS, while the four-output combined RMS is 2.02%. These subsets are 
dependent on the same scale and feature choices, and neither provides an independent early-phase 
validation. An inheritance interpretation requires independently observed spatial relations persisting 
across phases, preferably shared non-ideal departures; approximate circle diameters alone do not 
establish that transfer.
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988

Stonehenge | v1.39 | 37
The fixed-scale minimum near 2965 BCE and dimensionless minimum near 2985 BCE supply 
alternative temporal references for the inheritance hypothesis. Both fall within the initial-enclosure 
range and the same Cygnus model favours this early branch over the later sarsen phase [2]. The 
fixed angular optimum near 2965 BCE also falls within the initial-enclosure range and, unlike the 
dimensionless optimum, within the separate 95% ditch interval [18]. Three of the four counted 
material roles are later than, or insecurely attributable to, the initial enclosure (Table 3). The 
complete inventory cannot directly date its own construction to either optimum. Its possible historical 
significance is that a later material arrangement retained an earlier reference epoch, a hypothesis 
requiring dated evidence of continuity in the numerical convention.
The temporal contribution lies in the sequence of inference: a specified count map supplies a 
numerical word, the astronomical calculation returns an epoch without fitting archaeological dates, 
and published chronology provides the external comparison. The same word, operator and 
nominated minimum are documented in the public Calanais archive of 26 August 2026; the 
Stonehenge count map was recognized retrospectively. This establishes the antecedent of the 
calculation, not a blinded discovery. Tests against alternative words, stellar assignments, scales and 
construction phases show what this model-based epoch inference recovers and where it depends 
on the adopted conventions. The 0.8% and 2.6% window fractions describe coverage of the search 
span, not calibrated probabilities of the discovery process. Section 3.15 removes the absolute 
timing-scale convention and reuses the already computed free-scale score. Its additional result is 
the combination of geometric role selection and a uniform local horizon test, rather than a second 
independent chronological confirmation. The complete S8 census extends the fixed target to a 
different field family; its matching scale-free control ranks Cygnus 68th. Neither the enlarged count 
nor the larger fixed-scale error ratio calibrates the retrospective discovery process.
The link between material geometry and the temporal calculation survives replacing the exact 
integer word with continuous metric targets. Cygnus remains first in all five declared variants at the 
original timing scale, and every global optimum falls within the initial-enclosure interval. When spatial 
and temporal scales are freed independently, Cygnus remains nondominated alongside other 
solutions. These controls identify which conclusions survive changes to the model; shared inputs 
and retrospective choices prevent treating them as independent historical confirmations.
The geometric selector gives the temporal pairs a compact structural rationale: two adjacent 
intervals belong to a hub-centred trio, and the third belongs to the remaining pair. The complete field 
census recovers Cygnus and its roles without modern regional boundaries or a four-arm requirement 
when the target is fixed in degrees. Ratios remove absolute angular scale, and a vector formulation 
removes the coordinate zero and basis. These operations leave the physical terrestrial pole, field 
and role choices, and the ordered count map. The scale-free complete-census rank of 68 contrasts 
with first place in the four-arm subsets. Regional, complete-field, four-arm and free-role comparisons 
therefore answer distinct questions; none establishes prehistoric use of its family or map.
The compiler is a modern analytical representation of a hypothesized ancient numerical convention; 
prehistoric use of that convention requires independent evidence. The spatially filtered census and 
fixed-word panel recover Cygnus under distinct conditions (Table 4). Lacerta and Scutum free-word 
alternatives fail the original integrality condition; Vela passes it but fails the three closures. Applying 
these mathematical constraints does not establish their ancient use. A broader archaeological 
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030

Stonehenge | v1.39 | 38
comparison requires complete alternative material maps, stellar conventions and phase constraints 
evaluated on equal terms.
The solar and lunar literature addresses material evidence that a stellar numerical reconstruction 
must also confront. Pollard and Ruggles related changing depositional patterns to both solstitial 
ordering and lunar symbolism, without treating those interests as mutually exclusive [36]. Ruggles 
and Chadburn distinguish the widely accepted solstitial sightlines from unresolved intentionality in 
the Station rectangle's lunar alignment [37]. The calendar debate between Darvill and Magli and 
Belmonte illustrates disagreement over the inference from architectural counts to a particular 
timekeeping system [5,38–39]. In this study, a better stellar RMS is not a likelihood comparison 
against those archaeological interpretations: they predict horizon directions, seasonal practice or 
deposition, whereas the timing word predicts three numerical widths. Solar, lunar and stellar 
proposals can coexist logically, but their historical support must be evaluated through common 
observations and phase constraints. No superiority of the stellar interpretation is established here. 
The earlier eclipse-prediction debate also belongs to this history [40–41].
Astronomical reconstructions in heritage research can draw on different kinds of evidence. The 
Bekaa temple study combines measured orientations, reconstructed horizons and cultic context [42], 
whereas the proposed Shepseskaf eclipse association compares a calculated event with 
architectural choices and historical chronology [43]. Our count-derived comparison instead requires 
an additional numerical mapping whose ancient use is unverified. Reproducible numerical selection 
therefore constrains this model without supplying the contextual evidence needed to identify a 
prehistoric practice. The dimensionless branch reduces the required assumptions by dispensing with 
the absolute timing unit. It retains the numerical proportions and chosen interval topology.
The artefactual record gives the site a documented funerary context. Willis et al. describe cremated 
remains in and around the Aubrey circuit, a polished gneiss macehead associated with a burial, and 
bone or antler skewers interpreted as pyre goods [44]. Historical recovery and the later reburial of 
mixed material limit spatial interpretation. These finds attest to activities at the monument; they do 
not independently identify the count compiler, stellar roles or angular convention. No portable 
astronomical instrument is required by the model, so its absence would not by itself falsify it. Positive 
contextual evidence for repeated setting-out operations or a shared numerical practice would be 
more informative than assigning an astronomical meaning to funerary objects.
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060

Stonehenge | v1.39 | 39
4.1. Scope and limitations
Figure 12 shows the original fixed-unit benchmark. Sections 3.15–3.17 and S7–S8 specify the 
regional ratio extensions, complete fixed angular census and four-arm subsets. Global fitting 
precedes chronology comparison. Interval-constrained tests are separate, and shared inputs do not 
provide independent confirmations.
Figure 12. Dependencies of the original fixed-unit analysis. (A) Exterior calibration predicts 
interior dimensions. (B) The count map yields the word and conditional uniqueness. (C) 
Fitting returns rank and epoch without archaeological date constraints. (D) Chronology and 
construction phases enter afterward. Shared inputs do not provide independent 
confirmations; Table 4 distinguishes alternative controls.
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071

Stonehenge | v1.39 | 40
The compiler, geometric associations and later diagnostics were developed with relevant data 
available. Their shared inputs prevent multiplication as independent evidence. The uniqueness 
theorem is conditional on three closures; the temporal map alone has a nontrivial kernel. The 512-
word comparison fixes the Cygnus operator; the crossed panel retains its word library, regional star 
selection, temporal topology and search window. It does not release all maps, boundaries, subsets 
and operators together. Thus its candidate counts are not calibrated probabilities of prehistoric 
intention. Metric variants retain the construction, and joint-scale nondominance is shared. Table 4 
and Supplementary Notes 1, 14, 29 and 53–54 delimit these conditional results. S7 fixes one 
geometric assignment per existing regional quintet and considers the necessary local horizon 
constraint uniformly. These restrictions define a narrower question than the full free-role search and 
were combined after the underlying findings were known. S8 removes regional boundaries under 
explicit field and role rules. The complete nominal Hipparcos census puts Cygnus first with a fixed 
angular target and 68th with free scale. Its full fixed-scale ranking has not been repeated across the 
four catalogue/geometry panels or six kinematic variants. Those sensitivity results belong to the cap-
centred four-arm ratio family: its narrow nominal global first place is lost in three variants, while local 
first place survives all six. None of these conclusions extends automatically to other star counts, field 
shapes, target orders, losses or propagation models.
The metric ladder uses rounded summaries and an explicit Sarsen outer-envelope boundary. Its 
2.02% RMS describes dimensional agreement. The local control in section 3.1 fixes the scale and 
permits nearest-integer reassignment; it does not calibrate the complete retrospective selection 
process. Drawing proxies, restoration and absent settings constrain finer spatial claims. The 
approximately 0.09 m difference between ideal rectangles is not an archaeological error estimate; 
shared-source circle fits are not independent surveys. Sensitivity bounds, including ±0.25 m 
coordinate scenarios, are stipulated rather than measured uncertainties.
The path-family, orthogonality and entrance-based choices are retrospective. The entrance 
selector’s perpendicular criterion partly anticipates V’s inner-opening alignment. All directional 
scores reuse the same plan; exchanged near-orthogonal channels restate that alignment. The 
direction-only Gram does not identify the numerical labels supplied by the paths. Selection within 
four route classes neither establishes ancient anchors nor exhausts alternatives; the rational 
rectangle also passes the directional rule (Supplementary Note 54). Stellar coordinate types do not 
themselves define terrestrial directions: v is mean latitude, and triangle angles require a reference 
axis. Fixed-J2000 spatial coordinates and modern IAU boundaries in the earlier regional censuses 
are analytical conventions. The S8 field census uses neither the ecliptic tuple nor regional 
boundaries, but retains explicit field and role choices, with topology imposed only in the named 
subsets; Calanais remains author-reported comparative context. The flat-horizon night calculation 
leaves terrain, extinction, weather and any actual observing occasion unspecified (Supplementary 
Notes 17–19, 24, 36, 38–42 and 48).
Temporal sensitivity must be distinguished from archaeological uncertainty. Channel weights move 
the optimum across approximately 2985–2915 BCE, while scores within 10% of the minimum span 
about 65 years; neither is a confidence interval. The channelwise bounds preserve preference over 
Stage 2, not the exact optimum or its interval membership. Across six losses and three input 
branches, optima span about 2933.5–3000.6 BCE. The free-scale optimum near 2987 BCE uses a 
different objective and admits competitors. The fixed-J2000 spatial convention and quarter-degree-
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114

Stonehenge | v1.39 | 41
per-count target lack independent cultural calibration. Converting sidereal minutes to degrees 
changes no fit; releasing absolute scale changes the model. The dimensionless branch removes the 
absolute scale and does not use the ecliptic tuple, while retaining the count map, equator-of-date 
intervals and chosen loss. Its optimum near 2985 BCE is distinct from the fixed angular optimum 
near 2965 BCE, and its complete-census Cygnus rank is 68. Chronological overlap therefore does 
not establish construction date or transmission. The 0.8% and 2.6% window fractions describe 
search-span coverage; phase RMS ratios describe error degradation, not historical odds. The 68% 
constrained ranking leaves the original optimum outside that interval and retains the alternatives in 
sections 3.9–3.12. Finally, fixed-scale robustness to arbitrary nonnegative weights and robustness 
under tested free-scale losses do not jointly establish robustness to arbitrary weights and scales. 
The much narrower 2984.95–2986.97 BCE range from the six S8 state variants concerns a fixed 
ratio objective and targeted astrometry changes only; it must not replace the broader loss and 
modelling sensitivities or be read as an archaeological uncertainty interval.
4.2. Further tests of geometric inheritance and stellar attribution
A prospective test of geometric inheritance would use independently surveyed, georeferenced 
coordinates and freeze the registration before examining later features. Define the early Aubrey 
centre and circuit departures without reference to Stations or inner structures; use a declared subset 
of independently dated later anchors to predict withheld sarsen, Y/Z and socket positions, 
propagating survey and construction uncertainty. Phase-specific and feature-specific scale models 
must receive the same holdouts (Supplementary Notes 13, 30–32 and 41–42). Persistent residuals 
beyond defensible tolerances would contradict the specified transfer. Original socket and mounting-
axis records can test the opening directions, including the unscored reflection prediction for Stone 
59. Agreement would support geometric continuity; chronological memory additionally requires 
independently dated evidence linking the numerical convention across phases.
For astronomy, a candidate matching or improving the fixed-word benchmark under identical inputs 
would contradict Cygnus’s sole first-place status in that panel. Distinguishing the historical stellar 
attribution requires an unused observable for which Cygnus and complete competing models predict 
different outcomes, evaluated under the same archaeological constraints. A geometric prediction 
shared by all candidates cannot identify the constellation. Evidence for the material map and 
reference convention remains necessary; missing artefacts or incomplete excavation are 
uncertainty, not automatic falsifiers (S5 Appendix, section 7). For each dimensionless local-
observation branch, a counterexample must belong to the stated candidate family, use its geometric 
role rule and meet the same horizon condition at its fitted epoch. Releasing the role rule returns the 
broader alternatives already documented. The complete fixed-scale claim additionally requires the 
same nominal Hipparcos field enumeration and one geometric assignment per quintet; its 
denominator is not a count of all possible stellar subsets or analytical choices.
5. Conclusions
One exterior calibration recovers four interior dimensions with 2.02% RMS, while a continuous two-
side ablation gives 2.014%. The specified count map supplies the fixed angular target (7°,18°,4°). 
Cygnus ranks first among 19,939 nominal Hipparcos J2000 fields near 2965 BCE, with 0.0975° 
RMS and a 2.20 runner-up error ratio, without constellation boundaries, four-arm filtering or a 
horizon constraint. Horizon reoptimization retains first place among 17,860 groups. Freeing scale 
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156

Stonehenge | v1.39 | 42
gives a nearby Cygnus optimum near 2985 BCE but rank 68 in the complete census; four-arm 
subsets retain first place with a narrow nominal margin. The older ratio-panel kinematic tests and 
regional and free-role controls remain distinct, including their adverse outcomes.
The contribution is a reproducible dimensional reconstruction and a stellar comparison with explicit 
selection rules. The complete field census removes modern regional boundaries and the four-arm 
entry condition from the fixed-scale result. The matching free-scale control separates nearby epoch 
recovery from identification under an absolute angular target. The count map, quarter-degree 
convention, five-member circular fields and geometric role assignment remain substantive 
assumptions. Evidence for prehistoric use and transmission is required to attribute these 
relationships to the builders. The astronomical optima are model outputs rather than construction 
dates.
Supplementary Materials
S1–S8 accompany the manuscript. S1 supplies the original analysis record and a reference 
crosswalk; the S5 README provides navigation and execution history for the earlier controls. The 
S6 Appendix and code archive document geometric role selection, its information order and 
reproduction commands. S7 supplies the dimensionless combination, all 88 results, horizon 
reoptimization and unit-invariance checks. S8 integrates the complete fixed angular field census and 
its matched free-scale control, preserves the earlier four-arm ratio panels and targeted kinematic 
sensitivity, and supplies invariance and geometric-ratio audits with complete adverse controls and 
reproduction instructions.
S1 Appendix: source ledger, methods, derivations and 54 supplementary notes. Its revision guide 
and marginal notices map the retained reference numbering and qualifications to the main text.
S2 Code: primary geometric, algebraic and temporal analyses, pinned inputs, shared Hipparcos 
catalogue, expected outputs and integrity manifest.
S3 Code: constructive, directional, material and source-review extensions, including the Note 54 
route-family audit.
S4 Code: stellar identification, brightness, ablation and observability censuses, with full outputs and 
a hash-checked dependency on the S2 catalogue.
S5 Appendix and Code: crossed selection, compiler feasibility, fixed-word and free-scale searches, 
phase contrasts, provenance and additional robustness tests. Sections 9–12 cover losses, restricted 
role censuses, joint scales and metric words; section 13 maps the files. Figure sources and full 
numerical precision are retained.
S6 Appendix and Code: geometric recovery of temporal roles, the two 88-quintet catalogue panels, 
clock reproduction, six-scenario persistence checks and subsequent historical scope check. All 
inputs and numerical outputs needed for S6 are included; the separate stellar manuscript is 
identified as unpublished [47].
S7 Appendix and Code: dimensionless timing with the S6 role rule, unrestricted and horizon-
constrained searches for all 88 groups, refined upper-culmination checks, unit-rescaling verification 
and comparison with S5. Every numerical input needed by S7 is included.
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195

Stonehenge | v1.39 | 43
S8 Appendix and Code: complete nominal Hipparcos field enumeration, fixed angular and matched 
free-scale searches, fixed-scale horizon reoptimization and nested four-arm subsets; the earlier four 
nominal four-arm ratio panels and six targeted kinematic recensus variants; coordinate invariance, 
regional loss and weight bounds, target permutations, continuous spatial ablation, spherical axis 
ratios and inverse-construction controls. Three portable modules retain their original bytes. A fourth, 
04_FixedScale, supplies the complete 19,939-group census, both search grids, all scores and 
numerical checks.
Author Contributions
Conceptualization, L.G.R. and M.A.S.; methodology, L.G.R.; investigation, L.G.R.; formal analysis, 
L.G.R.; validation, M.A.S.; visualization, L.G.R.; writing—original draft preparation, L.G.R.; writing—
review and editing, L.G.R. and M.A.S.
Funding
This research received no external funding.
Institutional Review Board Statement
Not applicable. This study analyzes published archaeological measurements, archived drawing 
proxies and astronomical catalogues; it involves no new human-participant or animal research.
Informed Consent Statement
Not applicable.
Data Availability Statement
Numerical inputs, outputs, archived plan proxies, figure sources and protocols are supplied in S1–
S8. The prior grammar and temporal operator are documented at 
https://zenodo.org/records/22103786 and https://zenodo.org/records/22691368; S5 verifies their 
provenance. The shared Hipparcos catalogue is distributed in S2. The third-party Rees raster used 
in one S3 diagnostic must be obtained from its source [28]; S3 records its expected hash. No 
restricted primary survey coordinates are used. The S5 README identifies dependencies, 
preserved versions, execution scope and integrity manifests. S6 includes its catalogue extracts, 
pinned kinematic states, role-selection code, clock engine and complete outputs; no external 
download is required to rerun S6. S7 includes the shared catalogue, unchanged precession engine, 
S5 and S6 regression benchmarks, its analysis code and complete outputs. The new calculation 
needs no external download. S8 includes the full archived Hipparcos and HYG inputs, the HYG 
licence, pinned kinematic variants, all candidates and scores, and four audit modules. Its 
computations require no external download once the stated Python dependencies are installed.
The Rees click-plan source [28] and English Heritage chronology summary [35] are identified in the 
main text. The Bulford excavation announcement [45], cited in the retained S1 record, is contextual 
reporting rather than a quantitative input to the Stonehenge model.
Portable code, pinned inputs and complete numerical outputs support reuse [46]. The S5 README 
preserves the earlier execution history. For revision v1.35, the outer-scale calculation, local 
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232

Stonehenge | v1.39 | 44
perturbation control and conditional integer audit were rerun, and the S5 checks recomputed 
reported temporal scores and verified archived ranks and robustness summaries. No new data or 
new full S2–S4 execution is claimed. Revision v1.36 adds the executed S6 diagnostic and its 
reproduction and integrity checks; the earlier S1–S5 archives and figure files are preserved 
unchanged. Revision v1.37 executes S7 and integrates its dimensionless comparison. All 88 roles 
and unrestricted scores are checked against S6 and S5, and both horizon-constrained and 
unrestricted minima agree across the two search grids. The earlier S1–S6 and figure archives 
remain unchanged. Revision v1.38 integrates the subsequently executed invariance, axis and field 
audits as S8. Their original numerical files and clean-extraction reproduction records are retained; all 
26 field-census output files reproduced byte for byte before integration. Revision v1.39 
independently enumerates the complete 19,939-group Hipparcos J2000 family and executes fixed-
scale, free-scale and fixed-scale horizon searches on 1-year and 0.5-year grids. The earlier 
03_Fields module was also rerun from clean extraction, reproducing all 26 outputs byte for byte. S1–
S7 and figures remain unchanged. The packaging checks document claims and file integrity; no new 
full execution of S1–S7 is claimed.
Acknowledgments
The authors thank the Ágora Study Group for sustained discussion and independent study support. 
During preparation, OpenAI ChatGPT was used for the purposes disclosed in Materials and 
Methods. The authors are responsible for the methods, source checks, results and interpretation.
Conflicts of Interest
The authors declare no conflict of interest.
References
1. Cleal RMJ, Walker KE, Montague R. Stonehenge in its landscape: twentieth-century excavations. 
English Heritage Archaeological Report 10; 1995.
2. Darvill T, Marshall P, Parker Pearson M, Wainwright G. Stonehenge remodelled. Antiquity. 
2012;86:1021–1040. doi: 10.1017/S0003598X00048225.
3. Parker Pearson M, Cleal R, Marshall P, Needham S, Pollard J, Richards C, et al. The age of 
Stonehenge. Antiquity. 2007;81:617–639. doi: 10.1017/S0003598X00095624.
4. Abbott M, Anderson-Whymark H. Stonehenge laser scan: archaeological analysis report. English 
Heritage Research Report Series 32/2012; 2012.
5. Darvill T. Keeping time at Stonehenge. Antiquity. 2022;96:319–335. doi: 10.15184/aqy.2022.5.
6. Hawkins GS. Stonehenge decoded. New York: Doubleday; 1965.
7. Thom A. Megalithic sites in Britain. Oxford: Oxford University Press; 1967.
8. Ruggles CLN. Astronomy in prehistoric Britain and Ireland. New Haven: Yale University Press; 
1999.
9. Snoeck C, Pouncett J, Claeys P, Goderis S, Mattielli N, Parker Pearson M, et al. Strontium 
isotope analysis on cremated human remains from Stonehenge support links with west Wales. Sci 
Rep. 2018;8:10790. doi: 10.1038/s41598-018-28969-8.
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270

Stonehenge | v1.39 | 45
10. Parker Pearson M, Pollard J, Richards C, Welham K, Casswell C, French C, et al. Megalith 
quarries for Stonehenge’s bluestones. Antiquity. 2019;93:45–62. doi: 10.15184/aqy.2018.111.
11. Madgwick R, Lamb AL, Sloane H, Nederbragt AJ, Albarella U, Parker Pearson M, et al. Multi-
isotope analysis reveals that feasts in the Stonehenge environs and across Wessex drew people 
and animals from throughout Britain. Sci Adv. 2019;5:eaau6078. doi: 10.1126/sciadv.aau6078.
12. Nash DJ, Ciborowski TJR, Ullyott JS, Parker Pearson M, Darvill T, Greaney S, et al. Origins of 
the sarsen megaliths at Stonehenge. Sci Adv. 2020;6:eabc0133. doi: 10.1126/sciadv.abc0133.
13. Gaffney V, Baldwin E, Bates M, Bates CR, Gaffney C, Hamilton D, et al. A massive late Neolithic 
pit structure associated with Durrington Walls henge. Internet Archaeol. 2020;55. doi: 
10.11141/ia.55.4.
14. Parker Pearson M, Pollard J, Richards C, Welham K, Kinnaird T, Shaw D, et al. The original 
Stonehenge? A dismantled stone circle in the Preseli Hills of west Wales. Antiquity. 2021;95:85–
103. doi: 10.15184/aqy.2020.239.
15. Nash DJ, Ciborowski TJR, Darvill T, Parker Pearson M, Ullyott JS, Damaschke M, et al. 
Petrological and geochemical characterisation of the sarsen stones at Stonehenge. PLoS One. 
2021;16:e0254760. doi: 10.1371/journal.pone.0254760.
16. Ciborowski TJR, Nash DJ, Darvill T, Chan B, Parker Pearson M, Pullen R, et al. Local and exotic 
sources of sarsen debitage at Stonehenge revealed by geochemical provenancing. J Archaeol Sci 
Rep. 2024;53:104406. doi: 10.1016/j.jasrep.2024.104406.
17. Clarke AJI, Kirkland CL, Bevins RE, Pearce NJG, Glorie S, Ixer RA. A Scottish provenance for 
the Altar Stone of Stonehenge. Nature. 2024;632:570–575. doi: 10.1038/s41586-024-07652-1.
18. Greaney S, Hajdas I, Dee M, Marshall P. Beginning of the circle? Revised chronologies for 
Flagstones and Alington Avenue, Dorchester, Dorset. Antiquity. 2025;99:655–671. doi: 
10.15184/aqy.2025.28.
19. Gaffney V, Baldwin E, Allaby R, Bates M, Bates R, Finlay A, et al. The perils of pits: further 
research at Durrington Walls henge, 2021–2025. Internet Archaeol. 2025;69. doi: 10.11141/ia.69.19.
20. De Smedt P, Garwood P, Chapman H, Deforce K, De Grave J, Hanssens D, et al. Novel insights 
into prehistoric land use at Stonehenge by combining electromagnetic and invasive methods with a 
semi-automated interpretation scheme. J Archaeol Sci. 2022;143:105557. doi: 
10.1016/j.jas.2022.105557.
21. Liu X, Yang S, Zhou C, Guo Z, Zhang T, Hou M. Using Source Feature Matrix for interpreting the 
result in virtual reconstruction of cultural heritage sites. NPJ Herit Sci. 2025;13:83. doi: 
10.1038/s40494-025-01658-2.
22. Ribeiro LG, Simonssini MA. The hidden code of Cygnus [O Código Oculto de Cygnus]. 2026. 
Portuguese. ISBN: 978-65-266-6755-2.
23. Petrie WMF. Stonehenge: plans, description, and theories. London: Edward Stanford; 1880.
24. Atkinson RJC. Some new measurements on Stonehenge. Nature. 1978;275:50–52. doi: 
10.1038/275050a0.
25. MacKie E. A new look at the astronomy and geometry of Stonehenge. Cult Cosmos. 2012;16:89
–107.
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310

Stonehenge | v1.39 | 46
26. Lawson AJ. The structural history of Stonehenge. Proc Br Acad. 1997;92:15–37.
27. Field D, Pearson T. Stonehenge landscape: an archaeological analysis of the field survey. 
English Heritage Research Department Report Series 109-2010; 2010.
28. Daw T. A georeferenced plan of Stonehenge. Rees 1989 sheet MP/STO0861 click-plan 
[Internet]. Archived numerical extraction and source hashes in S2 Code [cited 2026 Sep 19]. 
Available from: https://timdaw37.github.io/stonehenge-plan/
29. Laskar J. Secular terms of classical planetary theories using the results of general theory. Astron 
Astrophys. 1986;157:59–70.
30. Verbunt F, van Gent RH. The star catalogues of Ptolemaios and Ulugh Beg: machine-readable 
versions and comparison with the modern Hipparcos catalogue. Astron Astrophys. 2012;544:A31. 
doi: 10.1051/0004-6361/201219596.
31. van Leeuwen F. Validation of the new Hipparcos reduction. Astron Astrophys. 2007;474:653–
664. doi: 10.1051/0004-6361:20078357.
32. Vondrák J, Capitaine N, Wallace P. New precession expressions, valid for long time intervals. 
Astron Astrophys. 2011;534:A22. doi: 10.1051/0004-6361/201117274.
33. Vondrák J, Capitaine N, Wallace P. New precession expressions, valid for long time intervals 
(corrigendum). Astron Astrophys. 2012;541:C1. doi: 10.1051/0004-6361/201117274e.
34. Wenger M, Ochsenbein F, Egret D, Dubois P, Bonnarel F, Borde S, et al. The SIMBAD 
astronomical database. Astron Astrophys Suppl Ser. 2000;143:9–22. doi: 10.1051/aas:2000332.
35. English Heritage. History of Stonehenge [Internet]. [cited 2026 Sep 26]. Available from: 
https://www.english-heritage.org.uk/visit/places/stonehenge/history-and-stories/history/
36. Pollard J, Ruggles C. Shifting perceptions: spatial order, cosmology, and patterns of deposition 
at Stonehenge. Cambridge Archaeol J. 2001;11:69–90. doi: 10.1017/S095977430100004X.
37. Ruggles C, Chadburn A. Missing data. Cosmovisiones/Cosmovisões. 2024;5(1):99–109. doi: 
10.24215/26840162e007.
38. Magli G, Belmonte JA. Archaeoastronomy and the alleged ‘Stonehenge calendar’. Antiquity. 
2023;97:745–751. doi: 10.15184/aqy.2023.33.
39. Darvill T. Times they are a-changin’: a response to Magli and Belmonte. Antiquity. 2023;97:752–
754. doi: 10.15184/aqy.2023.61.
40. Hoyle F. Stonehenge—an eclipse predictor. Nature. 1966;211:454–456. doi: 10.1038/211454a0.
41. Atkinson RJC. Moonshine on Stonehenge. Antiquity. 1966;40:212–216. doi: 
10.1017/S0003598X0003252X.
42. Magli G. Archaeoastronomy of the Temples of the Bekaa Valley. Heritage. 2021;4:1526–1537. 
doi: 10.3390/heritage4030084.
43. Magli G. The 1 April 2471 b.C. Eclipse and the End of the Fourth Egyptian Dynasty. Heritage. 
2025;8:492. doi: 10.3390/heritage8110492.
44. Willis C, Marshall P, McKinley J, Pitts M, Pollard J, Richards C, et al. The dead of Stonehenge. 
Antiquity. 2016;90:337–356. doi: 10.15184/aqy.2016.26.
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348

Stonehenge | v1.39 | 47
45. Wessex Archaeology. Discovery led by Phil Harding reveals 5,000-year-old prototype for 
Stonehenge solar alignment [Internet]. 2026 Jun 18 [cited 2026 Sep 27]. Available from: 
https://www.wessexarch.co.uk/Prototype-stonehenge-solar-alignment
46. Richards J, Brimblecombe P. The transfer of heritage modelling from research to practice. Herit 
Sci. 2022;10:17. doi: 10.1186/s40494-022-00650-4.
47. Ribeiro LG, Simonssini MA. Spherical certificates decouple centre and connectivity update times 
in evolving stellar patterns. 2026. Unpublished manuscript.
1349
1350
1351
1352
1353
1354
1355
