uššira piṭāti — the Amarna letters as a network, done carefully

The Amarna letters (c. 1360–1330 BC) are the diplomatic correspondence of the Egyptian court under Amenhotep III and Akhenaten — 382 clay tablets in Akkadian, found in 1887 at Tell el-Amarna: letters from the “Great Powers” (Babylon, Mittani, Hatti…) and from the small vassal kings of the Levant, most of them begging pharaoh for troops that never came.

A well-known study (Cline & Cline 2015, popularized in Cline’s Love, War, and Diplomacy, 2025) read the archive as a social network and reported a “small world” with clustering 48.75× a random graph, brokered above all by Rib-Hadda of Byblos, the archive’s most prolific complainer. This project rebuilds the Amarna networks from the Oracc lemmatized edition with fully reproducible code — one command regenerates every number, table, figure, and the typeset paper, byte for byte — and asks which of those findings survive proper null models and corrections for the archive’s structure. It is, after all, an ego-network of Egypt, excavated in Egypt: centrality in it measures salience to Egypt, not historical importance.

What survives, what doesn't

The correspondence network has zero triangles.: Everyone writes to Egypt; almost no one writes to each other (two vassal–vassal letter-pairs in 302 letters). Whatever the “small world of the Amarna letters” is, it is not a property of who wrote to whom.
The small world is real — but ×1.6, not ×49.: Rebuilt at full scale, the mention network (persons named together in a letter) matches Cline & Cline’s object (244 persons to their 246; clustering 0.409 to their 0.391). Against a null model that knows letters come in groups — a bipartite configuration null — the observed clustering exceeds chance by a factor of about 1.6. The astronomical multiplier came from comparing against the weakest available baseline.
Geography structures whom letters talk about, not who talks to Egypt.: Distance decays mention ties (r = −0.147, p < 0.001) and conflict ties (r = −0.298, p = 0.001): neighbors name and accuse neighbors. Letters to the court show no robust distance effect — the obligation to write did not thin with the road to Akhetaten.
The two-tier system fits in two coefficients.: An ERGM finds strong reciprocity (+3.78, concentrated in the Great-Power “brotherhood of kings”) and strong tier anti-homophily (−2.21): brotherhood above, servitude below, both materializing as letters to Egypt and almost never within tier.

Figures

Two network diagrams: the correspondence layer as a star centered on Pharaoh, and the clustered mention layer — **The two layers.** (a) Correspondence is a star on the court — its clustering is exactly zero. (b) The mention layer — who is named together — is where all the triadic structure lives.

Dot and interval chart comparing observed clustering against three null models — **Which null?** The multiplier you report is a choice of baseline: ×16.5 against Erdős–Rényi, ×1.6 against the fair bipartite null. The correspondence layer sits below its own null band, at zero.

Map of the Levant with accusation and alliance edges drawn between polities — **The conflict network in space.** Accusations (hand-coded from hostile language, every edge carrying its source quotation) run between neighbors: the Phoenician coast, the crowded hill country between Gezer, Gath, Shechem, and Jerusalem. The one long-range structure is the Qadesh–Damascus axis of the Hittite crisis.

Map of the eastern Mediterranean showing located vassal polities, Egyptian administrative centers, and Akhetaten — **The corrected geography of the corpus.** Vassal polities sized by letters to pharaoh, the five Egyptian garrison/commissioner seats, and Akhetaten, where the archive was found. Every coordinate is verified against the Pleiades bulk dump.

The paper

The Small World of the Amarna Letters, Revisited: A Reproducible, Bias-Aware Network Analysis (working draft; target venue Journal of Historical Network Research). Thirteen pages, six figures, and every number in it regenerates from ./run_all.sh at the repository root.

Reproduce it

Code (MIT) and hand-curated registries (CC-BY) live at git.sr.ht/~calgacus/ussira-pitati. The corpus itself is the Oracc aemw/amarna edition (Lauinger & Yoder, CC-BY-SA), fetched at build time and never redistributed. Every hand-coded row — entity aliases, conflict edges, adjudications against Moran (1992) and Rainey (2015) — quotes its evidence.