[doc] RETENTION examples: example 4 references a non-existent column; example 3 expected output contradicts its INSERT

[boluor]

Verifying the RETENTION aggregate-function page (sql-manual/sql-functions/aggregate-functions/retention.md) on a live cluster (master daily; same text in version-3.x / version-4.x) surfaced two doc bugs. The page creates its own tables, so these are content errors, not missing setup.

Bug 1 — example 4 queries a column that doesn't exist (errors instead of returning NULL)

SELECT RETENTION(date = '2022-10-12') AS r FROM retention_test2 where uid is NULL;

retention_test2 is defined as (uid int, flag boolean, flag2 boolean) — there is no date column, so this fails with:
errCode = 2, detailMessage = Unknown column 'date' in 'table list'.
The example's intent ("empty aggregation returns NULL") would be shown correctly with a column that exists, e.g. SELECT RETENTION(flag) AS r FROM retention_test2 WHERE uid IS NULL; (the WHERE uid IS NULL already filters to zero rows → NULL).

Bug 2 — example 3 printed output contradicts its own INSERT

INSERT into retention_test2 values (0, false, false), (1, true, NULL);
SELECT * from retention_test2;

The doc prints:

| uid  | flag | flag2 |
|    0 |    1 |  NULL |
|    1 |    0 |     0 |

but the inserted rows are (0, false, false) and (1, true, NULL), i.e. the actual result is (0, 0, 0) and (1, 1, NULL). The printed flag/flag2 values don't match the INSERT (they look transposed). The expected output should be corrected to match the data.

Both reproduce on a fresh single-BE cluster. Happy to send a PR correcting example 4's column and example 3's expected output if that direction is agreed.

[zclllyybb]

Breakwater-GitHub-Analysis-Slot: slot_c99895fc08c6

Thanks for the clear report. I verified the issue against the apache/doris-website RETENTION page. This is a documentation content bug, not a Doris runtime/setup issue.

Findings:

• retention_test2 is declared with only uid, flag, and flag2, so example 4 cannot validly query date. To demonstrate "empty aggregation returns NULL", the query should use an existing expression such as RETENTION(flag) while keeping WHERE uid IS NULL.
• Example 3 has inconsistent source data and printed output. Based on the following examples, the printed table appears to be the intended data: example 5 says flag is true for uid = 0, and example 6 relies on the uid = 0 row being skipped because flag2 is NULL. Those statements match (0, true, NULL), (1, false, false), not the current INSERT (0, false, false), (1, true, NULL).
• I found the same affected snippet in the current docs and in the version-3.x/version-4.x English plus zh-CN/ja source copies. I did not find an existing related issue or PR in apache/doris-website.

Suggested next step:
A doc-only PR should fix example 4 and then choose one consistent direction for example 3:

1. Preferred: update the INSERT to (0, true, NULL), (1, false, false) so the existing printed output and examples 5-7 stay coherent.
2. Alternative: keep the current INSERT, update the example 3 output to (0, 0, 0) and (1, 1, NULL), and also adjust the prose in examples 5 and 6 because they currently refer to the wrong uid.

No query profile or cluster logs are needed to proceed; the markdown content is internally inconsistent enough to confirm the issue.