Files
weatherreporter/docs/operations.md

163 lines
5.0 KiB
Markdown

# Weatherreporter Operations
## Normal Workflow
The implemented generation workflows are:
```text
weatherreporter generate daily --date 2026-05-29
weatherreporter generate tomorrow
weatherreporter generate three-day
weatherreporter generate weekend
weatherreporter generate storm --start 2026-05-29T18:00 --end 2026-05-30T06:00
weatherreporter run morning
weatherreporter run evening
```
These commands fetch weather data, build a briefing for the resolved valid
period, build the prompt input data package, run `scriptorium render`, run
`scriptorium run`, and write inspectable artifacts under the configured
workspace. The evening run resolves only the Tomorrow Planning Brief. The
morning run generates Daily Today and the 3-Day Outlook, plus Weekend Outlook
except on Sunday. Storm Report generation is manual and uses the explicit
`--start` and `--end` bounds as its valid period.
Scheduled run commands print a JSON aggregate summary to stdout and compact
per-report status lines to stderr. If one report fails, remaining independent
reports are still attempted. The command returns nonzero after the run when any
report failed.
## Filesystem Layout
The default workspace root is `workspace`.
```text
workspace/
snapshots/
daily/
YYYY-MM-DD/
<run_id>.briefing.json
<run_id>.metadata.json
three-day/
YYYY-MM-DD/
<run_id>.briefing.json
<run_id>.metadata.json
weekend/
YYYY-MM-DD/
<run_id>.briefing.json
<run_id>.metadata.json
storm/
YYYY-MM-DD/
<run_id>.briefing.json
<run_id>.metadata.json
data-packages/
daily/
YYYY-MM-DD/
<run_id>.data_package.json
three-day/
YYYY-MM-DD/
<run_id>.data_package.json
weekend/
YYYY-MM-DD/
<run_id>.data_package.json
storm/
YYYY-MM-DD/
<run_id>.data_package.json
preflight/
daily/
YYYY-MM-DD/
<run_id>.render.json
three-day/
YYYY-MM-DD/
<run_id>.render.json
weekend/
YYYY-MM-DD/
<run_id>.render.json
storm/
YYYY-MM-DD/
<run_id>.render.json
reports/
daily/
<run_id>.md
three-day/
<run_id>.md
weekend/
<run_id>.md
storm/
<run_id>.md
```
The Markdown report is written to a RunID-managed report path. When `--out` is
provided to `generate daily`, `generate tomorrow`, `generate three-day`,
`generate weekend`, or `generate storm`, the managed report is also copied to
that path.
For `run morning` and `run evening`, `--out-dir PATH` writes extra Markdown
copies using each report definition's default filename, such as `daily.md`,
`three-day.md`, `weekend.md`, or `tomorrow.md`.
## Run Identifiers
Run IDs are based on generation time plus report ID, such as:
```text
20260529T100000.123456789Z_daily_today
```
Managed artifact filenames use the RunID so repeated runs for the same valid
date do not overwrite each other.
## Metadata
Each generated report writes metadata that links:
- RunID
- report ID and prompt ID
- generation time and valid period
- source location, source hashes, and source warnings
- briefing snapshot path
- prompt input data package path
- preflight output path
- rendered report path
Run summaries include each report ID, prompt ID, RunID, status, error text when
applicable, valid period, and artifact paths known to the application.
## Recent Changes
When a prior comparable Daily briefing snapshot exists for the same valid local
date, the app compares structured briefing data before writing the prompt input
data package. Daily Today and Daily Tomorrow can compare with each other when
they cover the same valid local date. Meaningful changes are included under
`recentChanges.items`.
3-Day Outlook generation compares against a prior compatible 3-Day briefing
snapshot for the same valid local date when one exists.
Weekend Outlook generation compares against a prior compatible Weekend briefing
snapshot for the same weekend window when one exists. Friday evening and
Saturday runs may narrow the valid start while keeping the same Monday endpoint.
Storm Report generation currently leaves Recent Changes empty. Its explicit
event window is still recorded in briefing and metadata artifacts.
When no prior comparable snapshot exists, or no configured threshold is crossed,
the Recent Changes list is empty.
## Recovery
If render preflight exits nonzero after producing a result, the captured stdout,
stderr, exit code, and command are still written to the preflight artifact, and
metadata is still written for inspection.
If `scriptorium run` exits nonzero after writing a report, the generated report
and metadata remain available for inspection. Exit code `2` is still returned as
an error because it indicates validation failed, even if report output exists.
For scheduled runs, inspect stdout first for the aggregate JSON summary, then
use the per-report artifact paths in that summary to inspect briefing,
data-package, preflight, metadata, and rendered report files.
The application does not currently implement resume, cleanup, archive, or
remote storage behavior.