a. Topic landscape
Each point is a leaf topic; x = panel-conditional origin year, y = cross-journal reach, size = paper count, colour = the origin journal's source/bridge/terminal role. Hover for top-words + origin journal + recombination rate. Same panel-conditional + 1995-censoring caveats as the cascade tab.
b. Cascade flows
Cascade caveats: origins are panel-conditional ('first in this journal panel', not first in the world) and the corpus is 1995 left-censored (early topics read as 'present at panel start', not 'born 1995'). The cascade is validation-REAL: the gate-critical S1/Null-1 seeding-spread permutation PASSes at both grains and on the resolvable subset, the drop-one-journal origin jackknife flip-rate = 0, and the leaf-vs-mid granularity check holds. The held-out-years check is supporting-only.
c. Journal roles
Roles are jackknife-stable (role_rank_correlation well above the 0.7 bar) and robust across grain. Caveat: 'terminal' = within-panel citation/flow sink, not a global dead-end; high-impact clinical journals consolidate work (net inflow) while the source role belongs to basic-science + specialty venues. The source/bridge/terminal label is a coarse argmax; the continuous component scores are the nuanced output.
d. Novelty origins
Sector novelty is essentially a null (company sits mid-pack; novelty is not a tech/industry phenomenon here). Geography is the most differentiated axis (descriptive, likely topic-mix driven). Recombination is a corpus-internal lower bound. Funding and citation-intent layers are deferred (coverage gaps; $0 session). Institution type/country blanks are excluded, never imputed.
e. Trajectories
Trajectories are a descriptive per-topic state-space model (a local-linear-trend fitted on logit(share) / log(volume)) projected +5y → 2030 with 80% uncertainty bands; all 149/149 topics fit OK. The fit window is 1995–2025 with 2026 excluded (a partial harvest year). 'Share' is panel-conditional — the share of the 78-journal annual assigned output, not a global field share. This is a state-space extrapolation of past trend, explicitly NOT a predictive/causal forecast and NOT the F3 emergence GNN (which was signed NULL at Gate G4). Read the bands, not the point line.