The line between 'wait for a measurement' and 'write it now' is clear: if the conclusion depends on a run, wait; if it depends only on structure or rules, write it.
Five types you can write today
| Type | Why no waiting | Example |
|---|---|---|
| Indexes and tables of contents | Depends on existing structure, not on results | This page, the start index, the claims list |
| Directory layout notes | Directory names and purposes are facts | What lives in each $DSH_HOME directory |
| Flags and defaults | Copy the --help text and state the version | The startup flag table |
| Channel and version mapping | Look it up in the registry, note the capture date | The npm channel table |
| Boundaries and prohibitions | You set the rules; nothing to run | What never to paste into a session |
Looks writable, actually needs a run
- 'Which mode saves tokens' — must run, several times, keeping records.
- 'What changed in config across an upgrade' — actually upgrade and diff.
- 'What causes this error' — reproduce it, or it is only a compiled claim.
- Any performance number — no raw record, no number.
Why this list is worth maintaining
It decides when the site looks empty. With structural pages alone, a reader can install, look up flags, and know the boundaries — already more useful than most peers. Numeric content can be slow; one wrong number discounts the whole site.
Our line: anything marked Compiled carries a per-line source; anything marked To be tested stays blank rather than filled with an estimate. Not caution for its own sake — a wrong number costs more than an empty table.