Seked

Thesis

Numbers first, then the model

There are two ways to build this. The first is to model the pyramids beautifully and paint the claims on afterwards as annotations. The second is to put the numbers first: every dimension lives in a database with its source and its uncertainty, the geometry is generated from those numbers, and each claim is a formula over the same numbers. The second way is what makes the model high fidelity rather than a poster. It is also far more interesting to use. Switch the survey from Petrie 1883 to Dash 2015, nudge the cubit by a millimetre, drag the sky back to 10,500 BCE, and watch each claim's residual move in real time.

Five rules follow from that, and every later section is an application of them.

1. I

   One source of truth. Every number has provenance: who measured it, when, how, and with what error. Nothing is typed twice.

2. II

   Claims are data. A claim is a file: inputs, formula, target, tolerance, sources, and an overlay spec. Adding a claim never touches app code.

3. III

   Two monuments. "As built" (reconstructed, with casing and pyramidion, carrying error bars) and "as today" (stripped, 8 m shorter). Claims are about the first; the evidence comes from the second.

4. IV

   Show the fit. Each overlay displays measured versus claimed, the residual, and a count of the free choices the claim needed: which base line, which unit, which epoch. The app never says true or false. It shows the number.

5. V

   One frame. Origin at the Great Pyramid's base centre, axes east-north-up, georeferenced to WGS84. The Heliopolis line, the shaft alignments and the equinox sun all live in the same coordinates.

Deliverables

Six things that exist at the end

• D1

  Measurement database

  Versioned JSON with a schema, one record per dimension, provenance on every value, and named survey presets.

• D2

  Claims registry and dossier

  One YAML file per claim; a generated document (Markdown and PDF) with the maths, the fit and the sources for each.

• D3

  Parametric geometry library

  TypeScript that turns D1 into meshes, plus a landmarks file naming every point and axis a claim can reference.

• D4

  Blender scene

  Plateau terrain, all structures, materials, LODs, exported glTF, and the hero renders and animations.

• D5

  Web app

  Viewer with layer toggles, section cuts, a walk-in interior, a real sky with a time scrubber, and one overlay per claim.

• D6

  Deployment

  Static build on Cloudflare Pages, with a README and the dossier alongside.

Scope

The site, in four tiers

The necropolis is enormous. Tiers keep the mastaba fields from eating the schedule while still giving the plateau its real shape.

Plate I

The Great Pyramid in section

Drawn to scale from the survey values below, north on the left. Solid lines are the monument as built. Dashed lines mark today's truncated top and the two ScanPyramids discoveries. Blue lines are the four shafts, the subject of claim C2. This is the thing Tier 1 produces, in outline.

D1

Measurement database

The most valuable and most tedious deliverable. A few hundred records, each hand-entered from a primary source with its uncertainty. The sources rank as follows, and a record always names which one it came from.

1. Petrie 1883, The Pyramids and Temples of Gizeh. The base survey for the interior and the casing angle; public domain; measured in British inches.
2. Cole 1925, Survey of Egypt Paper 39. The classic base-and-orientation survey of G1, per side.
3. Glen Dash Foundation with AERA, 2012–2018. Modern total-station survey of the G1 base, corners and orientation.
4. Lehner & Hawass 2017, Giza and the Pyramids. The canonical rounded values for everything on the plateau; the default preset.
5. Maragioglio & Rinaldi 1965, vol. IV. Measured drawings of every interior element.
6. Gantenbrink 1993, the Upuaut robot survey. The only precise geometry of the four shafts, including the bends.
7. ScanPyramids: Morishima et al. 2017 (Big Void), Procureur et al. 2023 (north-face corridor).
8. Legon 1979 and Dorner's survey for the relative positions of the three pyramids; Nell & Ruggles 2014 for orientations across the site.

Starting sheet

These are the values I would seed the database with. Every one gets re-verified against its primary source in Phase 0; this table is a starting sheet, not the database.

Record shape

One record per measured quantity. Derived quantities (perimeter, apothem, volume) are never stored; the geometry library computes them so they cannot drift from their inputs.

{
  "id":        "g1.base.side.north",
  "structure": "g1",
  "quantity":  "length",
  "value":     230.253,
  "unit":      "m",
  "sigma":     0.02,
  "source":    "cole-1925",
  "method":    "casing-edge",
  "note":      "north side, casing baseline"
}

Presets bundle a coherent set of records: petrie-1883, cole-1925, dash-2015, and canonical (Lehner & Hawass rounded values, the default). Switching preset regenerates the geometry and re-evaluates every claim. Units are metres internally, with royal cubits, British inches and Smyth's pyramid inch (1.001 in) as display units, because several claims only work in one of them.

D2

Claims registry

Each claim is a card with the same six parts: what it claims, what it compares, the fit against the current preset, how many free choices it needed, its sources for and against, and the overlay it draws. Fits below are computed from the starting sheet with a 0.5236 m cubit. The n column counts free choices: a decision the proponent made that changes the answer, such as picking the socket base over the casing base, an invented unit, or an epoch. Zero means the number falls out of the monument with no help.

A · Proportion and geometry of the Great Pyramid

B · Earth and cosmos scale

C · Sky

D · Site plan and geodesy

E · Mass and construction, for context

Not claims, but numbers the overlays lean on: volume 2.59 million m³ from b²h/3; about 2.3 million blocks and 6 million tonnes; a 20-year build implies one block placed every three to four minutes of daylight; 203 courses survive, and Petrie measured every one of them, which is what drives the stepped profile in the "today" model.

Claim file

id: A1
title: π in the profile
group: proportion
inputs: [g1.base.perimeter, g1.height.original]
formula: perimeter / height
target: 2π
tolerance: 0.5%
free_choices: []
overlay:
  type: ghost-profile
  slope: atan(4/π)
sources:
  for: [smyth-1864, tompkins-1971, hancock-1995]
  context: [petrie-1883, rossi-2004]

packages/sky

Sky engine

Half the claims are about the sky, and the interesting epochs are 4,500 and 12,500 years ago. Standard astronomy code is not built for that, so this package needs care.

• Catalogue. HYG 4.2, which merges Hipparcos, the Yale Bright Star Catalogue and Gliese with proper motions and parallax in one CSV (CC BY-SA 4.0). Named subsets for Orion's belt, Sirius, Thuban, Kochab, Mizar and Regulus.
• Precession. The Vondrák, Capitaine and Wallace 2011 long-term model, valid for ±200,000 years. The IAU 2006 polynomials most libraries ship are wrong by degrees at 10,500 BCE, which is exactly where the Orion and Leo claims live. The model is a few coefficient tables; implement it in-house if no package does.
• Proper motion. Rigorous space motion. Sirius moves about 4.5° in 12,500 years; the belt stars barely move, which is why the belt's shape survives and Sirius' shaft alignment is epoch-sensitive.
• Frames. ICRS to mean equator of date (Vondrák), to horizontal altitude and azimuth at 29.9792° N, 31.1342° E through local sidereal time, to scene east-north-up. Refraction of about 34′ at the horizon and a local horizon profile for rising and setting claims.
• Sun. VSOP87-based positions from astronomy-engine; equinox and solstice events; ΔT matters for clock times of sunrise, not for its direction.
• Calendar. Julian Day numbers throughout, proleptic Julian calendar for BCE dates, astronomical year numbering (10,500 BCE is year −10499).
• Validation. Match Stellarium, which also uses Vondrák, for Alnitak's meridian-transit altitude at −2500 and −10500 before any sky overlay ships.

Architecture

One monorepo, one data flow

Everything downstream is generated from the measurement database. Blender and the web app consume the same glTF and the same landmark names, so a chamber centroid means the same point in a cinematic render and in a claim overlay.

seked/
  data/
    measurements/      one JSON file per structure
    presets/           petrie-1883 · cole-1925 · dash-2015 · canonical
    stars/             BSC5 + Hipparcos subset, as JSON
    terrain/           DEM references and the local-frame transform
  packages/
    units/             cubit · pyramid inch · seked ↔ degrees
    geometry/          pyramid() · passage() · chamber() · shaft() → buffers + landmarks
    sky/               vondrak.ts · frames.ts · sun.ts · catalogue.ts
    claims/            loader · evaluators · overlay descriptors
  blender/
    generate.py        bpy: build meshes from data/
    terrain.py         BlenderGIS import and re-origin
    export.py          glTF per structure
    scenes/            .blend files: plateau, sphinx, materials, cinematic
  apps/web/            Vite + React + R3F
  docs/
    dossier/           generated from the registry
    sources.md · decisions.md

Stack

• Language. TypeScript everywhere; Python only inside Blender. The geometry library is pure TypeScript so Blender, the web app and the tests all get identical vertices.
• Tooling. pnpm workspaces, Vite, Vitest, zod schemas on every data file.
• 3D. three, @react-three/fiber, @react-three/drei, @react-three/postprocessing. Optional "today" layer via NASA-AMMOS 3d-tiles-renderer streaming Google Photorealistic 3D Tiles.
• State. zustand for the layer, preset, time and claim state; URL-encoded so a view is shareable.
• Astronomy. astronomy-engine for the Sun and coordinate utilities; in-house Vondrák 2011 precession written from the paper, since no maintained npm package implements it.
• Geo. proj4 for WGS84 to local east-north-up; QGIS to prepare the DEM and orthophoto in a projected CRS before Blender sees them.
• Assets. Blender 4.2 LTS with BlenderGIS; gltf-transform for Draco and KTX2 after export.
• Frame conventions. Origin at the G1 base centre; +X east, +Y north, +Z up in data and Blender; the glTF exporter converts to Y-up for the web. Metres everywhere.
• Deploy. Static build to Cloudflare Pages.

D4

Blender pipeline

For when you are home. The rule is that Blender models what only Blender can, and generates the rest.

• Generate, don't model. A bpy script reads the measurement files and builds exact-vertex meshes for G1, G2 and G3. Concavity is a shape key from 0 (flat) to 1 (0.9 m). The casing is a separate shell, so "today" is hide the casing and cap the core at 138.75 m.
• Real courses. Geometry Nodes instance the 203 surviving courses with Petrie's measured course heights, so the stripped profile shows the actual thick courses at 35, 44 and 67 rather than a uniform staircase. This is the single cheapest fidelity win in the project.
• Interior as separate solids. Chambers, passages, gallery and shafts are their own objects. Section views in the viewer come from clipping planes; a boolean-cut master exists for renders only.
• Terrain. BlenderGIS imports the DEM and orthophoto in a projected CRS; re-origin to the G1 base centre; check the base elevations against the survey before anything sits on it.
• Sphinx. A sculpt over Lehner's 1:50 plans, with the CC-BY scans open in a reference window. Temples and mastabas are extruded footprints with real heights.
• Materials. Tura casing (smooth, near white), nummulitic core (tan, course lines), Aswan granite (King's Chamber, G3's lower courses, G2's bottom course), basalt paving on the temple floors. Keep PBR to base colour, roughness and normal, and bake.
• Export. One uncompressed GLB per structure from Blender, then gltf-transform for compression, KTX2 textures and deduplication, since Blender's exporter does not write KTX2. The Z-up to Y-up conversion is the exporter's job; the origin convention is preserved.
• Renders. Equinox dawn from the east to show eight faces; the cutaway; and a sky-rollback animation whose star positions are baked from the sky package to CSV, so Blender and the web never disagree about where Alnitak was.

Schedule

Six phases

Durations assume evenings and weekends and are estimates. Each phase ends with a "done when" that can be checked, not felt.

1. 0Weekend 1

   Foundation

   Repo scaffold, the record schema, G1 exterior and interior values from Petrie, claims A1, A2 and B1 in the registry, and unit tests that recompute the perimeter-to-height ratio.

   Done when pnpm test is green and the dossier renders three claim cards from data alone.

2. 1Weeks 1–2

   Geometry

   G1 exterior with seked, concavity, casing and core, and truncation. Interior solids. The landmarks file. glTF export and a Blender import that proves the round trip.

   Done when Petrie's derived numbers (apothem, volume, the 41°59′ arris angle) come out of the model, not the database.

3. 2Weeks 2–4

   Blender

   Terrain, G2 and G3, Tier 3 masses, the Sphinx, materials, LODs, exports, and the first hero render.

   Done when the whole plateau loads in the viewer under a triangle budget of about 1.5 million at the closest LOD.

4. 3Weeks 3–5

   Viewer

   Scene loading, camera modes (orbit, fly, walk the passages), section plane, layer panel, sky dome with sun and a time scrubber from 12,000 BCE to today.

   Done when Alnitak's transit altitude matches Stellarium at −2500 and −10500 to within an arc-minute.

5. 4Weeks 5–8

   Overlays

   Group A, then B, then C, then D. Each claim gets its overlay, its maths panel, a sensitivity control and its sources. The preset switch and the cubit slider live here.

   Done when every card in the registry renders in 3D and the dossier regenerates from the same files.

6. 5Weeks 8–10

   Polish and ship

   Tour mode as a narrated sequence of views, dossier export, the Blender cinematic, deployment, README.

   Done when a stranger can open the URL, click one claim, and understand what it says and how well it fits.

Risks

The hard parts, named

• Surveys disagree. The base differs by up to about 10 cm across Petrie, Cole and Dash, and the original height is projected from a surviving casing angle. Values are ranges, not points, and the overlays show the band.
• Which base? Casing edge versus socket corners changes the perimeter by roughly 1.4 m. Several classic claims silently pick whichever fits. Both are first-class in the database.
• Precession validity. IAU 2006 is wrong by degrees at 10,500 BCE. Use Vondrák or the Orion overlay is fiction in both directions.
• Interior geometry. The shafts are only known from Gantenbrink, and the Queen's north shaft bends. Model from polylines. Treat the ScanPyramids void as a position with error bars.
• Concavity. Measured on the core; whether the casing followed it is disputed. Ship both as a toggle and say so.
• Licensing. Google 3D Tiles require attribution and forbid extraction; Sketchfab scans are CC-BY but hobbyist grade; Digital Giza offers viewers, not downloads; Gantenbrink's drawings survive only on a mirror with no stated terms. Numbers from papers are facts and free to use.
• Scope. The tiers exist so Menkaure's interior and the mastaba fields cannot eat the schedule.
• Performance. Forty structures, terrain, 9,000 stars and the shafts. Budget about 1.5 million triangles at the closest LOD, lazy-load per structure, and keep the stars on a single point cloud.

Sourcing

Sourcing check

What is actually downloadable, under what terms, checked today. Items marked unverified could not be confirmed from a primary source and need a look from home.

Assets and terrain

Survey data

ScanPyramids

• Big Void. Morishima et al., Nature 552, 386–390 (2017), open access; arXiv PDF. Cross-section similar to the Grand Gallery, at least 30 m long, above the gallery. The paper leaves the inclination open, so the model carries it as a position with error bars and an inclination toggle.
• North-face corridor. Procureur et al., Nature Communications 14, 1144 (2023), open access, CC BY. About 9 m long, about 2 × 2 m in section, directly behind the chevrons above the original entrance. A 2025 resistivity follow-up in Scientific Reports confirms it.

Astronomy in JavaScript

• astronomy-engine (MIT) is right for the Sun, Moon and coordinate utilities. Its precession is the IAU 2006 polynomial, so it is not used for star positions at −10500.
• Vondrák 2011 has no maintained npm package. Implement it from the paper, which is two coefficient tables; a reference JavaScript snippet exists at astrogreg.com, and a package named falcon-ephemeris claims it (unverified). Stellarium's precession.c is GPL-2, so port it only if the whole package can be GPL.
• Stellarium is confirmed to use Vondrák 2011 in its source, so it stands as the validation reference.
• Star data. HYG 4.2 (CC BY-SA 4.0) merges Hipparcos, the Yale catalogue and Gliese with proper motions in one CSV, and is the easiest single file. BSC5 JSON mirrors exist on GitHub; the catalogue itself is public domain, the repositories carry their own licences.

Blender tooling

• BlenderGIS (GPL-3) is the standard for GeoTIFF and HGT import with a CRS. A third-party site claims Blender 4.x support from v2.2.15; check the official releases before installing.
• glTF. Blender's exporter does Draco but not KTX2. Export an uncompressed GLB, then run gltf-transform for compression and KTX2 textures; three.js loads the result with GLTFLoader plus the Draco and KTX2 loaders. Draco is pointless for the pyramid solids, which are a few hundred polygons; KTX2 matters for the terrain textures.

What this changes in the plan

• The shafts are re-drawn from Gantenbrink's published figures on the mirror, not imported from a DXF, unless the DXF turns up.
• The Sphinx is a sculpt over Lehner's plans with the CC-BY scan as reference. No survey-grade mesh is available.
• Vondrák precession is written in-house from the paper. astronomy-engine is used for the Sun only.
• HYG 4.2 replaces the planned join of the Yale catalogue and Hipparcos.
• The Google 3D Tiles layer stays optional and live-only, which the terms require anyway.

Discussion

Decisions to talk through

1. 1

   Product shape. Web app plus Blender assets, a Blender-only film, or a game engine build.

   Recommendation: web app plus Blender. It is the only shape where the claims are interactive, and it is shareable by URL.

2. 2

   First claim set. Which claims are in version one.

   Recommendation: A1–A4, B1–B3 and C1–C4, twelve claims that exercise every overlay type (ghost profile, chamber wireframe, scaled Earth, map inset, sky projection).

3. 3

   The "today" layer. Google Photorealistic 3D Tiles (needs an API key and attribution), CC scans, or none.

   Recommendation: start with none; the "today" model comes from the truncated core and Petrie's courses. Add 3D Tiles in Phase 5 if the terms fit.

4. 4

   The Sphinx. Scan, sculpt, or stylised massing.

   Recommendation: scan if a CC-BY one checks out, otherwise a sculpt. Massing would undercut the word "high fidelity".

5. 5

   Tone of the copy. Test bench, advocacy, or debunk. The model is identical in all three; only the words change.

   Recommendation: test bench. Show measured, claimed, residual and free choices, and let the reader decide. It is the most defensible and the most interesting.

6. 6

   Interiors for Khafre and Menkaure. Full or passages only.

   Recommendation: passages and chambers only, from published plans. No claim depends on their interiors.

7. 7

   Repository. New repo, public or private, monorepo as above.

   Recommendation: new private monorepo named after the working name, made public when the first overlays work.

8. 8

   Name. "Seked" is a placeholder that happens to be on-topic.

   Recommendation: keep it until something better appears; it costs nothing to rename before Phase 1.

Setup

First evening at home

• Install Blender 4.2 LTS and BlenderGIS from its official releases page, checking the 4.x compatibility note.
• Install QGIS.
• Install Node 22 and pnpm; scaffold the monorepo from the tree above.
• Download Petrie 1883 (Birdsall's transcription and the archive.org scan) and the Cole 1925 PDF.
• Download HYG 4.2 and cut it to stars brighter than magnitude 6.5.
• Download the Copernicus GLO-30 DEM tile N29 E031, which covers the plateau.
• Install Stellarium, set the location to Giza, and screenshot Alnitak's transit at −2500 for the validation test.
• Enter the starting sheet into the database, one record at a time, with the page number of its source.

Sources

Reading list

1. Petrie, W. M. F. (1883). The Pyramids and Temples of Gizeh. The base survey; public domain.
2. Cole, J. H. (1925). Determination of the Exact Size and Orientation of the Great Pyramid of Giza. Survey of Egypt Paper 39.
3. Dash, G. (2015–2018). AERAgram and Journal of Ancient Egyptian Architecture papers on the G1 base and orientation.
4. Lehner, M. and Hawass, Z. (2017). Giza and the Pyramids. The modern reference.
5. Lehner, M. (1997). The Complete Pyramids.
6. Maragioglio, V. and Rinaldi, C. (1965). L'Architettura delle Piramidi Menfite, vol. IV. Interior drawings.
7. Gantenbrink, R. (1993 onward). The Upuaut shaft survey; cheops.org is offline, mirrored at isida-project.org.
8. Morishima, K. et al. (2017). Nature 552, 386–390. The Big Void.
9. Procureur, S. et al. (2023). Nature Communications 14. The north-face corridor.
10. Legon, J. A. R. (1979). "The Plan of the Giza Pyramids." Archaeological Reports of the Archaeology Society of Staten Island.
11. Nell, E. and Ruggles, C. (2014). "The orientations of the Giza pyramids and associated structures." Journal for the History of Astronomy 45.
12. Spence, K. (2000). "Ancient Egyptian chronology and the astronomical orientation of pyramids." Nature 408.
13. Rossi, C. (2004). Architecture and Mathematics in Ancient Egypt. The seked explanation.
14. Vondrák, J., Capitaine, N. and Wallace, P. (2011). "New precession expressions, valid for long time intervals." Astronomy & Astrophysics 534, A22.
15. Smyth, C. P. (1864). Our Inheritance in the Great Pyramid. Origin of the pyramid inch and the land-centre claim.
16. Tompkins, P. (1971). Secrets of the Great Pyramid, with Stecchini's appendix on geodesy.
17. Bauval, R. and Gilbert, A. (1994). The Orion Mystery.
18. Hancock, G. (1995). Fingerprints of the Gods; Hancock, G. and Bauval, R. (1996). The Message of the Sphinx; Hancock, G. (2015). Magicians of the Gods.
19. Krupp, E. C. (1997). "Pyramid Marketing Schemes." Sky & Telescope, February. The inversion objection to C4.
20. Collins, A. (2006). The Cygnus Mystery. The alternative in C7.
21. Schoch, R. (1992). "Redating the Great Sphinx of Giza." KMT 3(2). The water-erosion argument behind C5.