How We Rate a Reading
Updated
A reading here is the result of one step in a chain of arithmetic. This page explains how each reading gets its rating, where its inputs come from, and what happens when a source cannot be read.
What do clear, hazy and overcast mean?
Seeing is a word borrowed from astronomy, where unsteady air blurs fine detail. On this site it rates how far the conclusion of a step can be trusted. It does not say whether the arithmetic is right, which has its own column.
- Seeing: clear. The inputs have a named first-hand source, the model's assumptions are listed, and their effect on the result has been estimated as a range. A value this site chose, or an assumption still open, that would move the result noticeably rules out clear, however exact the arithmetic is under that condition.
- Seeing: hazy. The estimate has reached two-road agreement, or the inputs are reliable but the assumptions move the result noticeably. That covers conditional arithmetic on a single road, a written figure known only second-hand or possibly from a shared origin, a check that can only run on a finite list, and a parameter this site picked.
- Seeing: overcast. The step lies after the dotted line, or two roads differ by more than 3 times.
Error ranges and conditions are still written out in the text (“if the Moon kept its mass”). A rating never stands in for them. Where a method has an error of its own, such as a fitted range or a finite sample, the page reports that too.
How do you read a Check my math table?
Each row is one step, with its input value, unit and source number. Three columns then answer three separate questions:
| Column | Values | Question it answers |
|---|---|---|
Source | primary, secondary, site assumption | Where did this input come from? |
Recomputed | yes, no | Was this formula re-run from the listed inputs? |
Seeing | clear, hazy, overcast | How far can the conclusion be trusted? |
A site assumption is a value this site chose, such as a round number used to set up a sum. It is labeled so you can swap in your own. The same data is available as a CSV beside the article, and some pages add a small table that changes one parameter.
What does recomputing prove?
Formulas are re-run in code, with units checked and the sums in the text compared against the output. A slip of a factor of 10 or more stops a page from going out. Where a sum mixes metric and customary units, the CSV keeps a unit column.
That tests the arithmetic only. A correct sum on a wrong input is still wrong, so the source chain answers for the inputs, and the Seeing rating answers for the assumptions. Passing a re-run never raises a step to clear.
Why two roads?
An estimate on this site tries to reach the same quantity along two paths whose main inputs come from different sources. Each road is calculated separately and the gap is printed.
Within 3 times counts as agreement. More than 3 times means overcast. The factor of 3 is this site's convention, not a statistical test. Agreement shows only that two roads are consistent, so the page also states what they share, because a common bias would survive it.
Which sources come first?
When sources disagree, this order applies:
- International and national institutions' data
- Peer-reviewed papers
- Descriptions from institutions that hold the original
- Reference books
- Everything else, with open wikis used only to find a way in and never as the final source
Sources at the same level that disagree are shown side by side, and the page does not pick a winner it cannot justify. For a quotation, the scan of the original publication settles it.
Every input number, quotation, “first” or “earliest” claim, founding and closing date, population figure and attribution of a scholarly position has a source. Anything likely to go out of date carries its year or an “as of”. The site uses one source version for each constant across all its pages.
What if a source cannot be read?
Where an original or a holding institution's record could not be consulted, the page says so and names who the figure or quotation came through. It cites what could be read and does not write as if it had seen the rest. A forecast known only through such a report can still be graded, but its original wording is not quoted, and it does not lead a section.
The Sources list on each page reflects this. If a source is listed with a note on what was read, such as an abstract or a single value, the claims in the text stay within that.
If a source's terms rule out a use, it is not used that way. Some sources are consulted to check a figure, and their tables are not reproduced. Where a source's terms rule out even that, a different source is found or the sentence goes.
How are quotations and data handled?
Quotations carry the publication, date and page or passage. Older texts are quoted only from editions whose translator and version are clearly out of copyright. Otherwise the page quotes the original language and adds this site's own translation, as with Verne's French. Work still in copyright gets a short quotation with its location, and modern studies are cited for their findings.
Check my math CSV files hold single values and this site's own results, with a source number on every row. Third-party data tables are not reprinted whole.
How are forecasts, communities and paradoxes judged?
Forecasts are graded item by item as hit, costume error, miss or untestable, and an item with no source is not graded. Untestable also covers an item that the sources we read cannot settle. A famous quotation with no traceable original appears only when the trace is the subject. Finding nothing earns the label Unverified. Calling a quotation misattributed or fabricated takes positive evidence.
Communities need two sources for dates and people. A detail with only one source is marked that way or left out.
Paradoxes state their formal assumptions and credit each position to whoever holds it. A table that works under those assumptions shows only what follows from them, and a finite run does not prove anything about the infinite.
What rules apply to images?
Figures are drawn from the same CSV as the numbers. A public-domain image needs a stated basis that holds in the United States, Germany and France, checked one image at a time, including the particular scan used. Scans keep their original paper color, and any crop is named in the credit line.
Illustrations are always labeled, never stand in for a real object or event, and carry no measurements. Details are on Reuse and Licenses.
What this site will not do
It does not predict real markets, elections, technology trends or doomsday dates. It does not offer medical, financial or legal guidance, or steps for experiments at home. A hypothetical's conclusion may be speculation. Its premise may not, so a premise that cannot be verified means no page.
Mistakes are handled on the Corrections page.