Kemet / geometry & reconstruction Independent research · Experiment 01
Break the vessel. Hide the answer.
Ask its fragments what belonged together.
A broken edge carries a relationship to another piece. One convincing fit is a beginning. Can those relationships recover an entire object—and agree when they meet again?
I’m exploring that question with a vessel inspired by pottery from Kemet, ancient Egypt. The object is synthetic. Its reconstruction is computed from the scattered fragments, using a geometric solver and GIGI.
K–01 / Recorded blind reconstruction
The assembly order, joins and loop checks come from the recorded solver run. Travel between positions is an animated presentation of those decisions, not a physical breaking or restoration simulation. In the reveal, teal traces the withheld original.
01 / What the pieces know
Take three neighboring points on a broken edge. Their three distances do not change when the fragment turns or moves. The solver looks for the same little triangle on another fragment, computes a rigid alignment, then checks for at least four matching boundary vertices.
The object’s color, cultural reference and original shape do not enter that calculation. This first experiment deliberately uses exact, unworn edges.
From a local fit to a shared frame
02 / The measured result
The evaluator opens the original only after the solver finishes. It aligns the assembled object as a whole, then measures the remaining vertex error. It never adjusts individual fragments to make the answer fit.
| Fixture | Recovered | Verified loops | Vertex RMS | Unassembled baseline RMS |
|---|
A control shifts one redundant proposed join by 12 mm. The loop check rejects it. Removing all candidate relations leaves only the anchor piece placed. These checks test specific mechanisms; they do not establish that every incorrect assembly will be found.
Loading the measurement-noise probe…
The vessel has small surface irregularities, which help distinguish its fragments. Erosion, missing fragments, scan noise and genuinely ambiguous assemblies remain further tests. This is a surface model, without material thickness or fracture physics.
03 / Why GIGI is here
GIGI stores the fragment geometry. The solver reads those records, computes candidate rigid transformations, writes the candidates back, and retrieves them for assembly. Its join decisions and closure checks are then stored and read back as the animation’s event record.
The registration and assembly mathematics live in the accompanying Python solver. This experiment uses GIGI’s record storage and retrieval; it does not treat a database curvature statistic as proof that the vessel is correct.
The separation is by program interface and process, not a security sandbox. All source is available for inspection.
04 / Kemet, as a point of departure
The reference is everyday ceramic form: a rounded body, a short neck and an open mouth. Yale’s account of the Umm Mawagir ceramics describes globular or oval jars with short necks; the Met’s collection offers further examples of Middle Kingdom pottery.
This mesh is an original computational fixture inspired by that context. It makes no claim to reproduce a particular object, ancient manufacturing method or archaeological reconstruction.
05 / Open the working record
Independent computational experiment · 26 October 2026. The browser replays saved results; it does not connect to a live GIGI service.