Table columns

Package: TABLES_PACKAGE. Registered by createEngine(); for a slimmer engine, register it explicitly (see choosing packages).

A markdown table is otherwise the one block the engine reads and skips, so the numbers in it cannot be totalled from where they sit. A column can be named and read as data instead:

Expression	Meaning
sum of column "cost" in table above	the total of the named column
average of column "cost" above	the mean of the named column
min of column "cost" above	the smallest cell
max of column "cost" above	the largest cell
count of column "cost" above	how many numeric cells it has
median of column "cost" above	the middle cell
standard deviation of column "cost" above	how spread out it is
variance of column "cost" above	the squared spread
spread of column "cost" above	largest minus smallest
mode of column "cost" above	the most frequent cell

total of column and mean of column are accepted as synonyms of sum and average, minimum of column and maximum of column of min and max, and stdev of column of standard deviation of column. The address is optional: with only the nearest table to read from, sum of column "cost", sum of column "cost" above, and sum of column "cost" in table above all mean the same thing.

The table’s rows must start with a pipe, and the header needs a |---| separator under it. That separator is what makes the rows a table: without it, a line starting with a pipe is an expression, since | is also bitwise or (5 | 3 is 7). With it, every row is markup, answering nothing and reporting no error, and the table ends a block for total above the way a heading does. A cell may still hold an inline solve, which is worked out as it would be in a line of prose.

Every summary form the table above names, read from the same column:

| item | cost |
| ---- | ---- |
| rent | 1200 |
| food |  300 |
| taxi |   12 |

sum of column "cost" in table above     // 1,512
average of column "cost" above          // 504
min of column "cost" above              // 12
max of column "cost" above              // 1,200
count of column "cost" above            // 3
median of column "cost" above           // 300

The spread and shape aggregates read a column too, so a table of readings can be summarised where it sits:

| reading | score |
| ------- | ----- |
| a       |     2 |
| b       |     4 |
| c       |     4 |
| d       |     4 |
| e       |     5 |
| f       |     5 |
| g       |     7 |
| h       |     9 |

standard deviation of column "score" above  // 2
variance of column "score" above            // 4
spread of column "score" above              // 7
mode of column "score" above                // 4

The sample forms, which divide by one less than the count for readings that are a draw from a larger set, read a column the same way, as do the other spellings:

| reading | score |
| ------- | ----- |
| a       |     2 |
| b       |     4 |
| c       |     4 |
| d       |     4 |
| e       |     5 |
| f       |     5 |
| g       |     7 |
| h       |     9 |

sample standard deviation of column "score" above  // 2.14
sample variance of column "score" above            // 4.57
stdev of column "score" above                      // 2
minimum of column "score" above                    // 2
maximum of column "score" above                    // 9

The column name is matched case-insensitively.

A column of money totals in its currency, the way the same figures typed as lines do under total above. A cell reads as money when a currency symbol comes before the amount ($200, £12,570) or a currency code after it (1,200 GBP), and a plain number in the same column joins it as an amount in that currency:

| item | cost  |
| ---- | ----- |
| rent | 500   |
| food | $200  |
| car  | 1,200 |

total of column "cost" above   // $1,900.00
count of column "cost" above   // 3
max of column "cost" above     // $1,200.00

Two currencies in one column are refused by name, as total above refuses them, and so is the variance of a column of money, which would be in square dollars; its standard deviation is in dollars.

A cell with no figure in it, a label, a blank, or a figure with a unit such as 5 km, is skipped rather than counted, so a stray row does not break an otherwise-numeric column. Thousands have to be grouped in threes to read as one number: 4,812 is 4812, and 12,57 is text, since a misplaced comma is more likely a typo or a decimal comma than a thousands separator. A percentage cell is counted by count, and refused by the summaries that add or compare, because a percentage is a proportion rather than one of the figures. A column with no number or money cells at all, or a name that is not one of the headers, is a clear error rather than a silent zero.

Unit cells are not read by a column summary yet, and a table whose rows do not start with a pipe is not recognised. Both are left for a later slice.

To read one cell rather than a whole column, name its row with a table lookup: column "cost" for "food". A lookup reads cells the same way, and answers a percentage cell as a percentage. A table of thresholds and rates can be applied to an amount as banded rates.

Like line references, a column read only works inside a document, since it reads a table elsewhere in the note.