fix(pii): remove real member names tied to payments

- Scrub three parity fixtures that still held raw bank messages
- Use fictional names in the junior regression test, its changelog
  entry and its plan file
- State the rule in CLAUDE.md: a real name as a bare test input is
  fine; a real name next to payment details, or any junior or parent
  name, is not
- Tick T4 in the cleanup task list

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

CHANGELOG.md

@@ -1,5 +1,12 @@
 # Changelog
 
+## 2026-10-09 00:21 CEST — fix(pii): remove real member names tied to payments
+
+- Three parity fixtures held raw bank messages with real names; the names are now `Member_<hash>` pseudonyms, as the scrubber produces.
+- The junior regression test, its changelog entry and its plan file used a real junior's and parent's name with paid months; replaced with fictional names.
+- `CLAUDE.md` now states the rule: a real name as a bare test input is fine, a real name next to payment details or any junior/parent name is not.
+- Git history still contains the old text.
+
 ## 2026-10-08 21:27 CEST — feat(fees): junior rate for 2026-08 is 0
 
 - Junior monthly rate for 2026-08: 500 → 0 CZK, same as adults.
@@ -294,7 +301,7 @@
 ## 2026-05-06 16:38 CEST — fix: include juniors in payment-inference roster
 
 - `scripts/infer_payments.py`: union adults + junior rosters so junior-only members are visible to the matcher.
-- Root cause: `get_members_with_fees()` reads only the adults sheet; junior-only kids like Jáchym Kubík were absent from `member_names`, causing the exact-match short-circuit to never fire and a different adult sharing the first name to win via fuzzy review.
+- Root cause: `get_members_with_fees()` reads only the adults sheet; junior-only kids like Jáchym Dvořák were absent from `member_names`, causing the exact-match short-circuit to never fire and a different adult sharing the first name to win via fuzzy review.
 - Two regression tests added to `tests/test_match_members.py`.
 
 ## 2026-05-06 16:05 CEST — feat(go/M2.10): port domain/reconcile.Reconcile

CLAUDE.md

@@ -29,6 +29,9 @@ These exist only in the Go app on purpose. Do not port them to the legacy Python
 ## Key Constraints
 
 - **PII separation**: Member data (names, emails, payment info) must never be committed to git. Enforce config/data separation from day one.
+  - Tests may use a real member's name as a bare matching input (diacritics, nicknames, word order).
+  - Never commit a real name together with payment details or raw bank messages, and never commit names of juniors or their parents. Use fictional names there.
+  - Parity fixtures under `go/tests/fixtures/` must be scrubbed (`Member_<hash>`); check new or regenerated fixtures before committing.
 - **Configuration**: External service IDs, credentials, and tunable parameters are centralized in `go/internal/config/config.go` (mirrored in `legacy/scripts/config.py`). Domain-specific constants (fees, merged months) stay in their respective modules.
 
 ## Development Setup

docs/plans/2026-05-06-1626-infer-payments-junior-roster.md

@@ -2,12 +2,12 @@
 
 ## Context
 
-A bank payment from sender `JIŘÍ KUBÍK` with the message
+A bank payment from sender `JIŘÍ DVOŘÁK` with the message
-`Jáchym Kubík: 01/2026+03/2026+04/2026` is being inferred as
+`Jáchym Dvořák: 01/2026+03/2026+04/2026` is being inferred as
-`[?] Jáchym Hrušák (G)` instead of the obvious `Jáchym Kubík`, even though
+`[?] Jáchym Novotný (G)` instead of the obvious `Jáchym Dvořák`, even though
 the message contains his exact full name.
 
-**Root cause** (confirmed with the user): `Jáchym Kubík` is in the **junior**
+**Root cause** (confirmed with the user): `Jáchym Dvořák` is in the **junior**
 attendance sheet only — he does not appear on the main/adults sheet. But
 [scripts/infer_payments.py:101-102](scripts/infer_payments.py#L101-L102)
 builds `member_names` by calling `get_members_with_fees()`
@@ -15,17 +15,17 @@ builds `member_names` by calling `get_members_with_fees()`
 `EXPORT_URL` (the adults sheet). Junior-only members are therefore invisible
 to the matcher.
 
-With Kubík absent from `member_names`, the matcher in
+With Dvořák absent from `member_names`, the matcher in
 [scripts/match_payments.py:65](scripts/match_payments.py#L65) processes the
-combined text `jiri kubik jachym kubik: 01/2026+03/2026+04/2026` against an
+combined text `jiri dvorak jachym dvorak: 01/2026+03/2026+04/2026` against an
 adults-only roster:
 
 - The exact-full-name short-circuit (`match_payments.py:75-84`) finds nothing —
   no adult's full name is in the text.
-- Hrušák `(G)` is the only adult with first name `Jáchym`. He fails the
+- Novotný `(G)` is the only adult with first name `Jáchym`. He fails the
   auto-rules (his surname isn't in the text) but hits the partial-first-name
-  review rule (`match_payments.py:123-125`) → returned as `("Jáchym Hrušák (G)",
+  review rule (`match_payments.py:123-125`) → returned as `("Jáchym Novotný (G)",
-  "review")`, rendered as `[?] Jáchym Hrušák (G)`.
+  "review")`, rendered as `[?] Jáchym Novotný (G)`.
 
 The user's original framing — "exact match in message should win over
 everything" — is already implemented for any candidate that **is** in the
@@ -33,7 +33,7 @@ roster (the May-04 short-circuit). The bug is upstream: the right candidate
 was never even considered.
 
 **Goal:** make `infer_payments` consider junior members as candidates, so
-junior-only names like `Jáchym Kubík` get matched correctly.
+junior-only names like `Jáchym Dvořák` get matched correctly.
 
 ## Approach
 
@@ -98,23 +98,23 @@ changes are needed.
   the adults version, so `m[0]` works for both.
 - [scripts/match_payments.py:65-137](scripts/match_payments.py#L65-L137) —
   `match_members` already handles the precedence the user wants (exact full-name
-  short-circuit), so once Kubík is in `member_names`, the case will be auto-matched
+  short-circuit), so once Dvořák is in `member_names`, the case will be auto-matched
   with no `[?]`.
 
 ## Verification
 
 1. **Manual sanity** — re-run inference on the offending row:
-   - Clear `Person`/`Purpose` for the Kubík row in the payments sheet.
+   - Clear `Person`/`Purpose` for the Dvořák row in the payments sheet.
    - `make infer`.
-   - Expect `Person = Jáchym Kubík`, `Purpose = 2026-01, 2026-03, 2026-04`,
+   - Expect `Person = Jáchym Dvořák`, `Purpose = 2026-01, 2026-03, 2026-04`,
      no `[?]`.
 
 2. **Unit test** — extend
    [tests/test_match_members.py](tests/test_match_members.py) (or add a small
    `tests/test_infer_payments.py`) to assert that, given a roster that
-   includes `Jáchym Hrušák (G)` and `Jáchym Kubík`, the message
+   includes `Jáchym Novotný (G)` and `Jáchym Dvořák`, the message
-   `Jáchym Kubík: 01/2026+03/2026+04/2026` resolves to
+   `Jáchym Dvořák: 01/2026+03/2026+04/2026` resolves to
-   `[("Jáchym Kubík", "auto")]` only. This is really a regression test for
+   `[("Jáchym Dvořák", "auto")]` only. This is really a regression test for
    the May-04 short-circuit — the new behavior under test is just that
    `infer_payments` now feeds in juniors.
 
@@ -122,7 +122,7 @@ changes are needed.
 
 4. **Dashboard smoke** — `make web`, open `/payments`, confirm the row now
    shows the correct member; open `/juniors`, confirm the payment is
-   credited to Kubík for the three months listed.
+   credited to Dvořák for the three months listed.
 
 5. **Changelog** — once the user confirms the fix, append an entry to
    [CHANGELOG.md](CHANGELOG.md) per [CLAUDE.md](CLAUDE.md):

docs/plans/2026-10-08-1026-pre-feature-cleanup-tasks.md

@@ -58,7 +58,8 @@ before new feature work starts. Companion to the Go rewrite tracker,
     current year.
   - Same applies to the `sync-2026` Makefile target.
   - Moot if Python is retired before then.
-- [ ] **T4** Remove member names from tracked files (PII rule in `CLAUDE.md`).
+- [x] **T4** Remove member names from tracked files (PII rule in `CLAUDE.md`).
+  Done 2026-10-09 in a narrower form, see Notes & decisions. Original task:
   - `CHANGELOG.md` line 29 names a member in a fix description.
   - `tests/test_match_payments.py`, `tests/test_reconcile_exceptions.py` and
     `go/internal/domain/reconcile/reconcile_test.go` use names that may be
@@ -172,3 +173,16 @@ Plan: [2026-10-08-1103-api-parity-vs-months-to-show.md](2026-10-08-1103-api-pari
 ## Notes & decisions
 
 (Add entries as you go. Format: `YYYY-MM-DD — short note`.)
+
+- 2026-10-09 — T4 scope decided by the owner: a real name as a bare test
+  input is acceptable; a real name next to payment details, raw bank
+  messages, or any junior or parent name is not. Replaced the junior
+  regression-test names (test, changelog, plan) with fictional ones and
+  scrubbed three parity fixtures that still held raw bank messages. The
+  other test names stay. Git history was not rewritten. Rule recorded in
+  `CLAUDE.md`.
+- 2026-10-09 — The `real_*` parse_month_references seeds pick rows from the
+  live payments cache by index (`_REAL_PMR_INDICES`). The sheet has changed
+  since capture, so `make capture-fixtures` now rewrites those fixtures with
+  different messages. Commit only the fixtures a change needs, or replace
+  the index-based seeds with fixed synthetic ones.

go/tests/fixtures/pure/parse_month_references/real_filip_prosinec_leden_unor.json

@@ -3,7 +3,7 @@
   "func": "scripts.czech_utils.parse_month_references",
   "captured_at": "2026-05-06",
   "input": {
-    "text": "Filip Halamka - prosinec, leden, unor",
+    "text": "Member_607b748d - prosinec, leden, unor",
     "default_year": 2026
   },
   "output": {

go/tests/fixtures/pure/parse_month_references/real_jachym_numeric_multi.json

@@ -3,7 +3,7 @@
   "func": "scripts.czech_utils.parse_month_references",
   "captured_at": "2026-05-06",
   "input": {
-    "text": "Jáchym Kubík: 01/2026+03/2026+04/2026",
+    "text": "Member_71343342: 01/2026+03/2026+04/2026",
     "default_year": 2026
   },
   "output": {

go/tests/fixtures/pure/parse_month_references/real_martin_prosinec_leden.json

@@ -3,7 +3,7 @@
   "func": "scripts.czech_utils.parse_month_references",
   "captured_at": "2026-05-06",
   "input": {
-    "text": "Martin Bolvansky Pausal Prosinec Leden",
+    "text": "Member_51a8ec2d Pausal Prosinec Leden",
     "default_year": 2026
   },
   "output": {

legacy/tests/test_match_members.py

@@ -51,21 +51,21 @@ class TestMatchMembersExact(unittest.TestCase):
     def test_shared_first_name_junior_in_roster_wins_exact(self):
         # Regression: two members share first name "Jáchym"; message has full name
         # of the junior-only member → exact match must win, no [?] on the adult.
-        roster = ["Jáchym Hrušák (G)", "Jáchym Kubík"]
+        roster = ["Jáchym Novotný (G)", "Jáchym Dvořák"]
         result = match_members(
-            "JIŘÍ KUBÍK Jáchym Kubík: 01/2026+03/2026+04/2026", roster
+            "JIŘÍ DVOŘÁK Jáchym Dvořák: 01/2026+03/2026+04/2026", roster
         )
-        self.assertEqual(result, [("Jáchym Kubík", "auto")])
+        self.assertEqual(result, [("Jáchym Dvořák", "auto")])
 
     def test_shared_first_name_without_junior_in_roster_falls_back(self):
-        # Without Kubík in the roster (old behaviour), Hrušák wins via first-name
+        # Without Dvořák in the roster (old behaviour), Novotný wins via first-name
         # partial match — confirms the roster-expansion fix is the real solution.
-        roster = ["Jáchym Hrušák (G)"]
+        roster = ["Jáchym Novotný (G)"]
         result = match_members(
-            "JIŘÍ KUBÍK Jáchym Kubík: 01/2026+03/2026+04/2026", roster
+            "JIŘÍ DVOŘÁK Jáchym Dvořák: 01/2026+03/2026+04/2026", roster
         )
         names = [r[0] for r in result]
-        self.assertIn("Jáchym Hrušák (G)", names)
+        self.assertIn("Jáchym Novotný (G)", names)
 
 
 if __name__ == "__main__":