Zero external issues and zero Discussions: post and pin the who's-using-this thread

[shivamlalakiya]

What's wrong

Every issue ever filed against this repository is maintainer-authored, and there are zero
Discussions even though Discussions is enabled.

That is one specific line on the JOSS reviewer checklist:

Collaborative effort: Does the project show evidence of community engagement such as
contributions from multiple developers, issues/discussions from external users, or community
influence on the development described in the paper?

Half of that item is already strong: 17 merged external pull requests from 5 external
contributors. The other half reads as absent. A reviewer sees an issue tracker where every entry
is the maintainer talking to themselves.

There is also no instrument by which a user could signal existence. The library makes no network
calls and has no telemetry, deliberately, and that is worth keeping. So a pinned Discussion is the
only available channel.

Where

File	Locate with
the drafted thread body, written and never posted	ls scripts/issue-drafts/_DISCUSSION_who_is_using_this.md
the existing generic Discussions link, near the bottom	grep -in 'discussion' README.md
the Maturity block the direct link should sit near	grep -n -i 'maturity' README.md

What to change

1. Post the drafted Discussion at scripts/issue-drafts/_DISCUSSION_who_is_using_this.md and
   pin it. Read it first: it was drafted before the paper landed and may want a sentence about
   the JOSS submission.
2. Link the thread directly from README.md. There is already a generic
   /discussions link, but it sits in the contributing section near the bottom of the file, where
   a prospective user does not reach it. Add a direct link to the pinned thread near the Maturity
   block.
3. Message the five external contributors (@slegarraga, @shubhrai23, @fuleinist, @BortnikMaxim,
   @AsavariCharati) and ask whether they hit anything worth filing. They touched the docs and the
   test suite; they will have noticed something. One issue from a contributor who is not you costs
   one message and changes the shape of the tracker.
4. Do not manufacture the signal. Do not file issues from a second account, and do not seed the
   Discussion with fake replies. A reviewer can read timestamps, and the whole point of this
   project's honesty passes is that the record survives inspection.

Tests to add or extend

None. Nothing here touches code.

Done when

# Exits 0 only once at least one Discussion exists.
R=PhilanthroPy-Project/PhilanthroPy
test "$(gh api graphql -f query="{repository(owner:\"${R%%/*}\",name:\"${R##*/}\"){discussions{totalCount}}}" \
  --jq '.data.repository.discussions.totalCount')" -gt 0

# And that the README links the thread itself, not just the generic tab.
grep -qE 'discussions/[0-9]+' README.md

Plus, and this is the part that actually matters and no shell check can assert: at least one issue
or Discussion reply authored by somebody who is not the maintainer.

Partly self-healing, but not in time

JOSS reviewers are instructed to file issues directly against the submitted repository under their
own accounts, so this gap closes during review. It does not help at triage, which is when the
checklist is first read.

[slegarraga]

External contributor here (5 merged PRs today: #106/#107/#118/#119/#120) — answering the question in point 3 before you even send the message.

Yes, I hit real things. Filed as #129: CRMCleaner.transform silently corrupts complex amounts into wrong finite floats (3+4j → -34.0) via the object-cast retry path interacting with the currency-string parser. Found it while writing coverage tests, reproduced on current main.

Two smaller observations from the same work, reported in the relevant PRs rather than as issues since they're documented there:

1. The except (OverflowError, OutOfBoundsTimedelta) guard and the tz_localize fallback in _encounter_recency.py are unreachable from public inputs on pandas 2.x and 3.x alike (evidence in test: close the EncounterRecencyTransformer coverage gap (79% -> 98% branch) #107). Defensive dead code, not a bug — but a reviewer poking at branch coverage will ask.
2. DX: pytest --cov=... --cov-report=term-missing fails with numpy's "cannot load module more than once per process" under this repo's --import-mode=importlib config on macOS, while coverage run -m pytest works. Cost me a detour; worth either a docs note or a CI-mirrored make target.

Nothing else blocked me — the runbooks (MAINTAINERS_DEPENDENCIES.md, the good-first-issue specs with their own verification commands) are genuinely the best I've worked from in any repo this size.

[slegarraga]

Correction to my own count above: six merged PRs today (#106, #107, #118, #119, #120, #123), not five — #123 (remote-path rejection) landed while I was writing this. Thirteen lifetime now.