Five worked uses of the open JUNO datasets, from download to scoring a period that is not in the record yet.
The JUNO datasets are open. These five worked uses go from downloading a file to scoring a period that is not in the record yet. Every file linked here is in the site repository; counts are computed from the files themselves (as of 20 November 2026).
JUNO_Unified_Dataset_v3_2026-10-19.csv: 41,914 rows, 42 societies, 1752 to 2026, one row per society, year and node. Columns: Society, Year, Node, Coherence, Capacity, Stress, Abstraction, Node_Value_v2, Source_File, Source_Note. What changed in v3 is in the changelog; the wider archive is on Datasets.
import pandas as pd
df = pd.read_csv("JUNO_Unified_Dataset_v3_2026-10-19.csv")
au = df[df.Society == "Australia"]
au.pivot_table(index="Year", columns="Node", values="Stress").tail()
In a spreadsheet, filter the Society column instead. Each row is one node in one year, so a full year is eight rows: Helm, Shield, Lore, Stewards, Craft, Hands, Archive, Flow.
Society, Year, Node, Coherence, Capacity, Stress, Abstraction, Node Value, Bond Strength).Advanced Analysis groups uploaded rows by Year and Node only. Upload one society at a time: a multi-society file such as the v3 corpus would be averaged across societies. Filter it first (use 1), or pick a society from the page’s built-in list.
In the browser: CAMS Compare sets two societies side by side on one kernel with a year slider. In code, from the v3 corpus:
pair = df[df.Society.isin(["Australia", "Argentina"]) & df.Year.between(1950, 2026)]
pair.assign(SK=pair.Stress / pair.Capacity) \
.groupby(["Society", "Node"]).SK.mean().unstack(0).round(2)
The ratio of Stress to Capacity per node shows where each system carries load it cannot absorb. Read it as stress levels and coupling, not as a score of merit: a score is a state, not a grade. Before reading anything into the network quantities, see the Temporal Inelasticity audit, which limits what bond-derived measures can carry.
A new score is one reading by one scorer. The site’s own panels use ensembles of independent passes; treat a single score as a hypothesis to test against them.
The Blind Analysis Challenge gives anonymised society files with years replaced by offsets, a protocol prompt for any capable model, and a sealed key. Run it before you look at the named data, so the result is yours and not a recognition of history you already know.