Proof inventory

Validation artifacts and institutional evidence register

Northflow follows a strict no-claim-without-artifact discipline. Every core capability is mapped to a concrete, reviewable artifact with status, update date, and access pathway.

Status legend: Operational, In development, In evaluation, Planned. Dates indicate latest release or target publication window.

Artifacts

Verifiable evidence inventory

Each entry below links to documentation, downloadable packages, or operational context used in institutional review cycles.

Gaia DR3 validation report

Operational

AVAILABLE — FEBRUARY 2026

Operational validation report documenting HGE execution over Gaia Data Release 3 with deterministically replayable outcomes, uncertainty handling, and reproducibility controls.

  • 1.8 billion object catalog coverage
  • Deterministic run-to-run reproducibility
  • Audit-ready provenance trail included

Evidence bundle specification

Operational

VERSION 1.0 — AVAILABLE

Technical specification for signed evidence bundles, schema constraints, cryptographic signatures, and verification procedures used across institutional review workflows.

  • Schema-level validation rules
  • W3C PROV-aligned provenance fields
  • Signature and integrity verification steps

Deterministic replay demonstration

Operational

DEMO PACKAGE — AVAILABLE

Replay demonstration package showing identical output generation under controlled inputs and policy constraints, used for institutional technical briefings.

  • Reproducible execution trace
  • Policy-gated verification checkpoints
  • Evidence chain integrity confirmation

Red-team tamper test results

Operational

SUMMARY — JANUARY 2026

Summary of red-team tamper resistance testing covering six adversarial attack vectors against evidence generation, storage, and verification pathways.

  • 6 attack vectors tested
  • Tamper detection and rejection paths
  • Incident evidence verification workflow

CERES famine forecasting — arXiv preprint

Published

PUBLISHED — MARCH 2026 — arXiv:2603.09425

Peer-reviewed preprint describing the CERES probabilistic famine early warning system. 43-country coverage, weekly 90-day IPC Phase 3+/4+/5 forecasts, AUC 0.84. The first famine EWS with a public write-once prospective verification ledger. Published on OCHA Humanitarian Data Exchange.

  • 43 countries — ~95% of active IPC Phase 3+ caseload
  • AUC = 0.84, n=87 historical IPC transitions
  • Back-validation: Somalia 2011, South Sudan 2017, Ethiopia Tigray 2022, Yemen 2021
  • Public grading ledger via /v1/grades endpoint
  • Dataset on OCHA HDX — CC BY 4.0, updated weekly

Sentinel adaptation progress logs

In development

MONTHLY UPDATES — ACTIVE

Implementation logs tracking HGE adaptation to ESA Sentinel data pipelines for wildfire risk, deforestation verification, and infrastructure stress indicators.

  • Adapter integration milestones
  • Validation checkpoint updates
  • Institutional engagement timeline

OQTOPUS quantum validation report

Operational

DELIVERED — FEBRUARY 2026 — UNIVERSITY OF OSAKA

26-experiment technical evaluation of HGE on the OQTOPUS quantum processing unit (University of Osaka, Dr. Naoyuki Masumoto). Depth invariance hypothesis confirmed at 90% confidence across 3 phases. Validates HGE instrument-agnostic architecture across astronomical and quantum physics domains.

  • 26 autonomous experiments across 3 phases
  • Depth invariance confirmed at 90% confidence
  • Ed25519-signed evidence bundles delivered
  • Full deterministic replay verified
  • Report delivered to University of Osaka

Institutional benchmark annex

Planned

PLANNED — Q2 2026

Cross-domain benchmark annex for institutional reviewers comparing reproducibility, verification latency, and governance-mode behavior across deployment contexts.

  • Comparative benchmark protocol
  • Governance-mode performance matrix
  • Publication readiness checkpoints

Report overview

Gaia DR3 validation report — public summary

This section summarizes what the validation demonstrates and what artifacts are available for review. It intentionally omits implementation details that constitute proprietary methodology.

Scope and dataset context

The Gaia Data Release 3 (DR3) catalog contains approximately 1.8 billion observational objects. Northflow’s validation demonstrates that the Hypothesis Generation Engine (HGE) can operate over this scale while producing institutionally reviewable evidence outputs.

At-a-glance

  • Dataset: Gaia DR3 catalog
  • Scale: ~1.8B objects
  • Output: signed evidence bundles
  • Property: deterministic replay

What this validation demonstrates

  • Deterministic replay

    A given run can be replayed to reproduce the same outputs under the same inputs and policy constraints — a requirement for audit-grade evidence handling.

  • Provenance-aware outputs

    Outputs are packaged as evidence bundles with traceable provenance context, enabling reviewers to understand where conclusions originate and how they are verified.

  • Institutional review readiness

    Artifacts are structured for evaluation in regulated environments where traceability, repeatability, and integrity checks are mandatory — not optional.

  • Scale without hand-curation

    The validation is designed to show that reviewable evidence workflows can operate at catalog scale, without requiring ad-hoc manual reconstruction of results.

Artifacts available for institutional review

Public materials provide a high-level validation narrative and a public-safe summary for review. Additional packages may be shared selectively for qualified institutional stakeholders.

  • Public validation summary (this page)
  • Evidence bundle schema + integrity verification steps
  • Replay demonstration context for briefings
  • Security / tamper testing summary (red-team results)

What we do not publish

For safety, integrity, and IP reasons, some implementation details are not published openly. We focus on publishing evidence, not recipes.

  • Internal heuristics, ranking logic, and decision policies
  • Operational parameters tied to deployment governance constraints
  • Security-sensitive mechanisms that would weaken tamper resistance if disclosed

Explorer

Evidence bundle explorer

Sample signed evidence bundle (anonymized). Expand each section to inspect how claims, provenance, integrity, and audit controls are represented.

Bundle ID

HGE-EVB-2026-02-017

Signature status

Verified

Replay status

Deterministic match

Hypothesis representation

Machine-readable hypothesis schema (JSON-LD envelope)

Contains formal claim statement, parameter bounds, uncertainty priors, and falsifiability criteria. Each block is versioned and linked to originating evidence context.

Provenance chain

W3C PROV-aligned event trace from ingestion to conclusion

Provenance events record ingestion, transformations, policy gates, and verification checkpoints used to reach each claim.

Cryptographic signature

Signed hash envelope with bundle integrity fingerprint

Cryptographic envelope includes signature metadata, key identifiers, and integrity hashes for tamper detection and institutional verification.

Audit invariants

Policy, replay, and tamper controls enforced per contract

Audit invariants define required controls for replay, policy enforcement, and tamper resistance across each execution context.

Sentinel adaptation progress logs

Monthly update stream for Earth Observation adaptation. Entries are maintained for institutional traceability and planning review.

February 2026

Sentinel ingestion adapter baseline completed. Initial wildfire-risk hypothesis templates validated in sandbox environment.

January 2026

Schema mapping finalized for Sentinel metadata fields. Provenance event contract aligned with evidence bundle specification v1.0.

December 2025

Domain transition plan approved: astronomy validation baseline → Earth Observation adaptation workflow with phased governance controls.