The product ships weekly. Every capability below is live, dated, and documented — most were built to close a specific, published failure.
May – June 2026 — The grounding
- Every probability now anchors to a counted reference class: 190+ classes, each prior a count over enumerable historical cases (crisis databases, conflict archives, election records) with a documented derivation and audit trail — not model intuition.
- Provenance became enforceable: derivation scripts, audit records, and a pre-deploy gate that blocks any counted claim that can’t reproduce its own number.
- Epoch discipline: every scoring-relevant change lands at a stamped boundary, so the track record can never silently rewrite its own past.
Early July 2026 — The machine that argues with itself
- The Falsifiability Register: every analysis now carries its own counter-case — counted counter-evidence with era stamps, machine-declared blind spots, and typed contradictions where the same dataset reads both ways. On its first production run it caught the exact contradiction it was built to find.
- Grounded Lenses: five schools of analysis — realist, institutionalist, constructivist, bargaining, domestic-organizational — read every story independently, take positions on the same scenarios, and are corrected by counted history where their own canonical claims fail. The strongest dissent from our house view is published beside it.
- Every reader-facing claim is visibly graded: counted (green, with the historical basis and era) or argued (amber). The difference is the product.
Mid-July 2026 — Accountability machinery
- Questions gained durable identity: consecutive analyses of the same situation now thread, so revisions are visible and scoreable instead of silent.
- A daily sweep checks every published analysis against what its own falsifiers said would prove it wrong — 361 records walked on its first run, 69 candidates surfaced, 31 human-adjudicated within a day.
- VERA now declines to analyze articles the world has overtaken: a generation gate checks each article’s core premise against verified newer evidence, and refuses — with a cited reason — rather than publish confidently on a dead premise.
- Institutional sources went from spec to watched: Lloyd’s JWC circulars signature-verified inside the source documents; UN Security Council, IAEA, CENTCOM and FCDO feeds polled on a clock, every finding citable.
- Resolution criteria for the first scoring cohort were adversarially tested, found wanting, repaired, and re-blessed — before a single measured score was produced under them.
August 2026 — The measurement
- The first cohort of predictions reaches its horizon in early August and resolves under pre-registered, repaired criteria. Measured Brier scores from then on. The track record page is where they land.
© 2026 VERA Narrative Intelligence Ltd · VERA analyses are for informational purposes only and do not constitute financial, investment, or legal advice.