A page where every AI claim is paired with a public probe you can rerun

August 10, 2026 by Asif Waliuddin

What this page is

This is a selected, non-exhaustive set of NXTG.AI artifacts. Each factual row below is paired with the public probe that supports the narrow claim beside it — nothing more — as of 2026-08-10. A resolving URL proves that an artifact exists and is reachable; it does not, on its own, establish quality, adoption, usage, revenue, reliability, peer review, originality, or endorsement, and nothing here should be read to imply those.

Every claim on this page is intended to be falsifiable: run the probe yourself and treat the result as time-stamped evidence. If a rerun fails, the probe no longer supports a current-state reading — report the failure and preserve the 2026-08-10 observation as time-stamped evidence; a later outage or removal does not retroactively falsify a dated reading.

We deliberately do not publish usage, revenue, or user-count figures here. Their omission does not imply undisclosed traction exists; it means those figures are out of scope for a page whose only claim is public existence and reachability.

The probe table (every row is independently checkable)

Each URL in this table was probed live on 2026-08-10 and returned a resolving artifact. Re-run any of them. Rows 8–10 are Wikidata records created by NXTG.AI's own agents during this work — see the disclosure below.

What each row supports — stated narrowly

1. A source-available developer tool listed in the official MCP Registry

Forge Orchestrator is listed on registry.modelcontextprotocol.io

— the official Model Context Protocol registry — as an active server (rows 1–2). The code is public and source-available under the Functional Source License 1.1 (FSL-1.1-ALv2, which converts to Apache-2.0 on 2028-03-18): anyone can read it, run it, fork it under the license terms, or audit the tool surface directly against the repository. It ships tagged v1.6.1 with CI and a published crate. A pull request to the community directory is open (row 3).

The narrow, checkable claim is exactly that: Forge is a public, source-available (FSL-1.1), registry-listed MCP orchestrator. We attach no install count, adoption figure, or production-readiness assertion — the registry entry and the open repository are the whole of the claim.

2. Public research deposits with permanent identifiers

Three papers carry Zenodo DOIs that resolve today (rows 4–6), with the research PDF index served live from nxtg.ai/research/pdf

(row 7). A Zenodo DOI is a persistent, third-party-minted identifier for a public research deposit — it is not peer review, and resolving does not certify the contents. The lead deposit, the CRUCIBLE Protocol, documents a proposed test-and-verification

method (discussed below). 3. A publicly inspectable knowledge graph

The work is registered as structured, interlinked items on Wikidata (rows 8–10): a Forge product item, an author item carrying a P496

statement that links to an ORCID identifier, and three paper items connected to the author item via the P50

(author) property — all serving HTTP 200.

Disclosure: these Wikidata records were created by NXTG.AI's own agents as part of this work. Wikidata pages are publicly inspectable structured assertions that anyone can review or edit — they are not an independent third-party endorsement of NXTG.AI, and this page does not present them as one. What they provide is machine-readable linkage a search or answer engine can traverse.

The editorial controls applied to this page

This page was prepared under three editorial controls, stated here as the policy applied to this artifact — not as a measured claim about how NXTG.AI works everywhere:

Drafting and final content review were assigned to separate principals. The party that drafts is not the party that signs off.Final review used a different model family under an adversarial brief— the reviewer's task was to falsify the claims, not to agree with them.** Each proof-table row names its public probe**— the re-runnable instrument that grounds it, as in the table above.

These are the controls applied here; this page does not measure how consistently they are applied elsewhere. Where they surfaced a problem, the fix is the version you are reading — but treat that as declared editorial process, not an independently verifiable public record, because the review was internal and its logs are not a probe you can run.

On error-catching in general: any "N of N caught" figure would be a statistical illusion — an uncaught error leaves no row to count — so we publish no catch rate, sensitivity, or no-misses claim. Observed catches show only that catches have occurred, never a rate.

The test-verification layer of this discipline — a method intended to catch hollow tests and silent failures — is written down and deposited as the CRUCIBLE Protocol (row 4). The DOI proves the deposit is public and citable; it does not, by itself, prove the method works.

How to read this page

Don't trust it — check it. Open the registry entry. Resolve the DOIs. Load the Wikidata items and read who created them. Clone the repository and read the tool definitions against the registry entry. Every row was written to be falsifiable: if the probe and the claim disagree at the time checked, the claim fails that check. That is the only kind of claim worth publishing.

Selected and non-exhaustive. Evidence here establishes public existence and reachability at the probe date — not usage, revenue, adoption, reliability, peer review, originality, or endorsement.