---
title: "Fact-Check Policy"
date: 2026-08-27
author: "Editor"
---

# Fact-Check Policy

Technology reporting fails most often not because writers invent things, but because they repeat a spec, a benchmark, or a version number from somewhere that never checked it. TechRT treats every figure, model number, and vendor claim as untrusted input until it has passed the verification described below.

## Scope

The verification stack applies to every factual claim in a TechRT article. “Factual claim” means anything a reader could later quote or act on. In practice that covers:

- **Reported figures:** benchmark scores, battery-life minutes, storage and memory capacities, prices, download counts, release-cycle timings
- **Identifiers:** app package names, model and part numbers, OS builds and version numbers, CVE numbers where a fix is discussed
- **Dates:** release dates, update rollouts, end-of-support and end-of-life dates, price-check dates
- **Attributions:** who said what, in which release note, developer document, support page, or filing
- **Derived figures:** calculations from primary data, with the underlying inputs and the arithmetic shown

Editorial analysis, forward-looking assessment, and opinion are flagged separately. When we draw a conclusion, we cite the evidence behind it.

## Source Hierarchy

We rank sources by how close they sit to the original information.

| Tier | Examples | Use | |—|—|—| | Primary, technical | Vendor documentation and release notes, developer documentation, official support pages, standards bodies (IEEE, IETF, USB-IF, Bluetooth SIG, Wi-Fi Alliance) | Default for technical claims | | Primary, governmental | FCC filings and rulings, court filings, regulatory orders and consumer-protection notices | Default for recalls, rulings, and statutory claims | | Primary, commercial | Manufacturer spec sheets, official store listings, first-party benchmark data with a stated methodology | Used with the methodology disclosed in our article | | Secondary, accepted | Coverage by an outlet that itself cites a tier above and links to it | Treated as a pointer; we then cite the primary directly | | Secondary, rejected | Aggregator spec sites, affiliate roundups that tested nothing, marketing decks, press releases without underlying data, forum hearsay presented as fact, AI-generated summaries | Not used |

If another publication breaks a story, we name them and link to the original story, then trace and cite the primary source ourselves.

## The Verification Stack

The stack runs in three phases. Each phase has hard gates: an article cannot move past a phase until every claim in it has passed.

### Phase 1: Capture (deterministic, automated)

- **Fetch and hash.** Every source URL is fetched, the publication date is recorded, and a SHA-256 of the page is stored. Re-fetching later detects silent edits.
- **Excerpt extraction.** The verbatim sentence or paragraph that supports each claim is captured before the article is drafted.
- **Claim tagging.** Every factual sentence in the draft links to the excerpt that backs it. Untagged factual sentences fail this gate.

### Phase 2: Cross-check (mixed, automated and reviewed)

- **Verbatim match.** Quoted strings in the article must match the captured excerpt character for character.
- **Math check.** Every derived figure is recomputed from the underlying inputs; any mismatch blocks publication.
- **First-pass classifier.** Each claim is checked against its excerpt and returned as supported, uncertain, or contradicted, with a confidence score.
- **Second-pass re-check.** Numerical claims, superlatives, and any first-pass result below threshold are re-run against a stricter check.
- **Source-tier upgrade.** Any claim leaning on a secondary source is re-checked against a primary equivalent before publishing.

### Phase 3: Sign-off (human, named)

- **Volatility refresh.** For zero-day fixes, live outages, and products still in rollout, any source older than 48 hours is re-fetched before publishing.
- **Editorial review.** A named TechRT reviewer with relevant credentials reads the piece end-to-end. Reviewer name and credentials are visible on the published article.
- **Audit log archived.** The full verification log (every claim, source, decision, score) is archived against the article ID for later reference.

## Threshold and Failure Handling

Each claim carries an internal confidence grade. The publish threshold is fixed; claims below it cannot ship. If the verification stack flags a claim mid-process, the writer has three options: re-source from a higher-tier reference, rewrite the claim narrower so the existing source supports it, or remove the claim. There is no fourth option.

## Independence

TechRT earns advertising revenue and accepts sponsored posts. Sponsored content is labeled and produced separately from editorial; it is never published as editorial. Advertisers have no say in what we cover, test, or rate. Reviewers disclose relevant employment or advisory positions on their author profiles, and an article is reassigned where a reviewer has a conflict.

## Refresh Cadence

Tech data ages quickly. Evergreen guides and statistics articles are revisited on a fixed schedule; how-to guides are re-tested when an OS or app update changes a step. Each article carries its last-reviewed date.

## Corrections

Send factual-error reports through our [contact page](/contact/). We respond to every correction request, log the change against the article, and refresh the audit log.

Last reviewed: 27 August 2026.