blackholes: dup-flood
The same posts ingested in two orders (arm B shuffled with seed 1): two graphs built side by side, with the claims the agreement metric paired linked across them. Matching is stateful, so where the graphs differ is where order mattered.
Fingerprints: what built each arm
blackholes (a) · baseline snapshot bh_small
epoch unknown commit unknown profile none: development defaults, not production models from run caps not recorded scored 2026-09-18 judge cost none
blackholes (b) · same configuration, second run
epoch 2026-09-domain-skills commit 2b6d590 profile none: development defaults, not production matcher z-ai/glm-5.3-flash steward z-ai/glm-5.3-flash curator z-ai/glm-5.3-flash extractor z-ai/glm-5.3-flash models from run caps stewardMaxRuns=2 stewardMaxIterations=25 curatorMaxRuns=2 curatorSweepRate=1 llmDailyTokenLimit=0 llmHourlyTokenLimit=0 scored 2026-09-18 judge cost none
System #0 Source submitted: «Astrophysical implications of hypothetical stable TeV-scale black holes»
Agreement a ↔ b: 10 claims paired, 10 unpaired · claim-set F1 0.67 · status agreement 100% · mean credence gap 0.00 · edge edit distance 2 · hover a claim to see its pair; “no pair” marks the rest
Source submitted: «Astrophysical implications of hypothetical stable TeV-scale black holes»
Steps · none recorded
Changed the graph · attribution: harness
- Source submitted: “Astrophysical implications of hypothetical stable TeV-scale black holes”
Past events · 0
Legend
System Extractor ↗ Matcher ↗ Claim Steward ↗ Curator ↗
verified supported inconclusive doubtful contradicted unknown unassessed (hatched: pending, not a verdict) merged away touched by the current event
Edges: requires grey, supports green, contradicts red dashed, assumes amber dotted. Gists in lists are derived by the exporter; anything marked verbatim is the model’s own text. Attribution “run-window” means a change was credited to a run by time and claim, not by a tool call that names it.