Skip to content

docs: close the loop on 009, 010 and 011 - #18

Merged
sam-aydlette merged 2 commits into
mainfrom
docs/close-the-loop
Aug 28, 2026
Merged

sam-aydlette merged 2 commits into
mainfrom
docs/close-the-loop

Conversation

@sam-aydlette

@sam-aydlette sam-aydlette commented Aug 28, 2026 •

Copy link
Copy Markdown
Owner

Housekeeping plus one real finding, from verifying the merged result end to end.

The three shipped specs are marked DONE

009 → #16, 010 → #17, 011 → #15. Same drift #14 just corrected; not recreating it a week later.

A deployment gap the tool caught itself

Adding a feed to config/feeds/ does not add it to the database. A beat resolves feeds from config, so a new source resolves, passes its tests, and CI goes green — but the fetcher reads the rss_feeds table, which is only populated by FeedManager.load_feeds_to_database().

Five trade outlets were configured by #16. CI was green. The corpus had zero articles from any of them.

What surfaced it was beat coverage, from #17:

[UNMATCHED] FedRAMP director reinstated
    no article from this beat's feeds matched
      and no feed anywhere in the corpus carried it either.

After running the sync (added=5) and fetching:

[MATCHED] FedRAMP director reinstated
  135 article(s) from this beat's feeds in window
    seen by FedScoop (2026-08-26)
      FedRAMP Director reinstated at GSA after weeks of leave  [name redacted]

3 matched + 0 unmatched + 1 inconclusive of 4 probe(s) declared
PASS: 3 of 4 probe(s) answered, none missed.
  1 probe(s) could not be answered and are counted above, not dropped.   [exit 0]

#17 was built to catch exactly this class and caught it on its first real use — against an incomplete deployment of #16, which merged an hour earlier. Recorded in the README's known limitations so the next person adding a feed doesn't lose an afternoon to a green CI and an empty corpus.

Two limitations corrected because they are no longer true

  • "Only one timescale" — fixed by calibration: let the operator stake and resolve their own predictions #15. Replaced with the narrower truth: the four pre-existing standing questions have no cadence and need re-declaring.
  • "The calibration record is empty" — still empty, but the machinery now exists. Rewritten to keep that distinction explicit: the fix is a mechanism, not a result. An operator can now stake a dated claim; nobody has. Until something is staked the tool's central claim remains unmet, and the README should not imply otherwise because a feature shipped.

Marks the three shipped specs DONE with the PRs that landed them, so the
backlog does not drift back into claiming nothing shipped -- the drift PR #14
just corrected.

Records a real deployment gap found while verifying the merged result:
adding a feed to config/feeds/ does not add it to the database. A beat
resolves feeds from config, so a new source resolves, passes its tests, and
fetches nothing until FeedManager.load_feeds_to_database() is run. Five trade
outlets were configured, CI was green, and the corpus stayed empty of them.
The tool's own coverage command is what surfaced it, reporting 'no feed
anywhere in the corpus carried it either'.

Corrects two limitations that are no longer true as written: the one-timescale
limitation is fixed, and the calibration machinery now exists though nothing
has been staked in it yet. The record stays honest about that distinction --
the fix is a mechanism, not a result.
Task 006 permits a person's name to appear as an attribute of a specific
document event and never as an accumulated record, on the reasoning that a
document attribute expires with the document while a durable row becomes a
file on someone. SOURCES.md and a task file are durable artifacts in a public
repository, not document events, and both quoted a news headline naming a
named federal official in connection with a personnel matter -- leave and
reinstatement.

That is the precise image the boundary was written to avoid: a public repo
belonging to someone who works in this domain, containing the personnel
situation of an identifiable official in it.

The point both passages were making -- that this feed carries full body text,
and that this is the event the beat was blind to -- does not require the name
and is preserved without it. The probe config never contained it: the shipped
probe is terms ['FedRAMP'] with any_of [['director','administrator'],
['reinstat*','return*','restored']], which finds the story without naming
anyone. The boundary held where it was designed in; it failed in prose nobody
had applied it to.

Note git history retains the original text. This removes it going forward,
which is what can be done without a force-push.
@sam-aydlette
sam-aydlette merged commit 4aed2b1 into main Aug 28, 2026
5 checks passed
@sam-aydlette
sam-aydlette deleted the docs/close-the-loop branch August 28, 2026 20:10
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant