Every technical figure on CompatSense comes from a published source. This page explains how sources are ranked, how disagreements between them are settled, how AI-assisted research is handled, and what happens when someone reports an error — including a public log of the corrections made so far.
Source hierarchy
Sources are ranked by how authoritative they are for the claim being made. The rank sets the starting confidence of any evidence that relies on it (see the confidence section of Methodology).
| Rank | Source type | What it is | Starting confidence |
|---|---|---|---|
| 1 | Standards body | A specification ratified by a multi-vendor organization (VESA, HDMI Forum, USB-IF). | High |
| 2 | Technology owner | Official documentation from whoever controls a technology’s certification, when that is not a multi-vendor body (for example Intel for Thunderbolt). | High |
| 3 | Manufacturer | The official specification or user guide for a specific product. | High |
| 4 | Official documentation | Manuals, knowledge-base articles and release notes that are not a specification. | Medium |
| 5 | Chipset documentation | Datasheets from the maker of a controller or chip. | Medium |
| 6 | Technical media | Specialist technical publications with a record of accuracy. | Medium |
| 7 | Independent testing | Reproducible third-party testing with a published method. | Low |
| 8 | Community | Forums and user reports. | Low |
Manufacturer documentation and standards
The two top tiers answer different questions. A standards body defines what a protocol can carry — DisplayPort 1.4’s 32.4 Gbps raw rate, for instance — but says nothing about what a particular product implements. For a claim about a specific dock or monitor, the manufacturer’s own documentation is the authority: its output ports, its maximum refresh rate, its Power Delivery figure. Where a manufacturer does not publish a value, the value is not filled in from the protocol’s ceiling: it stays unknown.
Secondary and community sources
Technical media are used when a manufacturer does not publish a figure, and evidence that rests on them is capped at medium confidence. Some entries in the current dataset are of this kind, and each is marked as such in the source table. Community reports are never presented as an official specification. They may prompt an investigation, but a claim that only a community source supports is not published as a fact: it is left unknown or stated explicitly as unconfirmed.
Retrieval dates and review
Each source is stored with the date it was retrieved and the date it was last checked, and both appear on the coverage page. Editorial articles show when they were published and when they were last reviewed; the “Last reviewed” date changes when an article’s claims and sources are re-checked, not for typo fixes. Re-verification is done by hand when a source or a product changes — there is no automated re-check yet — so the dates tell you how recent the check was, not that a source is guaranteed to be current.
When sources disagree
The higher-ranked source wins, but the disagreement is never silently discarded: the evidence record of the losing source is flagged with a known conflict, which forces its confidence to low regardless of how authoritative the source is. Where two sources of similar rank disagree and neither can be verified, the value is left unknown rather than choosing one. The number of evidence rows with a recorded conflict is shown on the coverage page.
AI-assisted research and editorial work
AI tools may be used to find primary sources, organize evidence, compare sources, spot gaps and draft explanations. They are not autonomous and they do not publish: the operator reviews every piece before it goes live, and no pages are generated in bulk. AI-assisted work does not invent facts, claim physical testing that did not happen, or supply expertise its author does not have. The compatibility results themselves are not written by an AI at all: they are produced by a deterministic engine from the evidence described above.
Unsupported facts are not presented as verified
A specification without a source is not published as fact. A figure that was inferred is labeled as inferred. A value nobody documents is reported as unknown. Where a page describes a limitation or a caveat, it says what is unconfirmed rather than implying the opposite. No hardware testing is claimed anywhere on this site unless a page states it explicitly.
Reporting an error and how corrections work
Write to contact@compatsense.com with the page address, the specific claim, and the source you believe says otherwise. There is no comment form or public ticket system. Each report is checked against the cited source and against the source you provide. If it is right:
- the text or the dataset record is corrected;
- for an article, the “Last reviewed” date is updated; for a dataset record, the source’s checked date is refreshed;
- a correction that changed a figure, a conclusion or a claim of fact is added to the log below, with what was wrong and what it says now — it is not deleted quietly.
If the report turns out to be mistaken, the page stays as it is. No response time is promised, because this is a one-person project.
Corrections log
Material corrections, newest first. These were found during an internal review of the site’s published claims.
| Date | Page | What was wrong | What it says now |
|---|---|---|---|
| 2026-09-30 | Compatibility Checker | The worked example gave 15.998 Gbps as the bandwidth needed for 3840×2160@60 in 8-bit RGB. That figure is the 10-bit requirement; the correct 8-bit figure is 12.799 Gbps (533.28 MHz × 24 bits per pixel). | The example is now generated from the engine on each page build, so the figure cannot drift from the calculation. |
| 2026-09-30 | Compatibility Checker | The same example presented an HDMI 2.0 result for the same monitor as if its timing had been resolved automatically. No verified HDMI timing for that monitor and mode is on file; the figure came from a standard timing supplied by hand in an internal test. | The example now shows what the Checker really returns for that request: UNKNOWN, because no verified HDMI timing is on file. |
| 2026-09-30 | Troubleshooting | The page said its guides were ordered by which link is “statistically most likely” to be the real cause. No statistical data supports that. | The page now says the guides are ordered by dependency: a check comes first when later checks only matter if it passes. |
| 2026-09-30 | DisplayPort Bandwidth Explained | The article said its worked examples were outputs of “the calculator on this page”, but the calculator is a separate page. | The article now states that the figures come from the Bandwidth Calculator, links to it, and notes they are standard-derived estimates. |
| 2026-09-30 | Enrique Rubio | The page referred to an “Editorial Policy referenced from Methodology” that Methodology did not contain. | The page now points to this page, which is the source and corrections policy. |
| 2026-09-30 | Contact and author page | They said a page’s “reviewed date” is updated whenever a claim is corrected, without describing what that date actually means. | Both now describe the process on this page. |
See also: Methodology for how the engine reasons, Dataset & Evidence Coverage for the devices and sources, and Enrique Rubio for who maintains the site.