feat(provenance): verworfene Daten mit Grund + Ersatz im Verlauf zeigen

Der Herkunfts-Verlauf nennt jetzt auch, wenn ein Wert aus einer Quelle VERWORFEN
wurde — mit Grund und was stattdessen verwendet wurde (⚠-Markierung), z. B.:
- „⚠ Vater-Kandidat ‚Danielle' (*04.03.2020) verworfen — falsches Geschlecht für
  die Vaterrolle; ‚Hagrid Rubeus' verwendet." (Molly)
- „⚠ Geschlecht weiblich aus ‚Wurfchronik.docx' verworfen — abweichend; männlich
  aus ‚Alberto Kids.xlsx' verwendet (Mehrheit)." (Solice)
- „⚠ Vater ‚Jasper Jacob' (*2018) verworfen — unplausibel (7 Jahre älter); kein
  Ersatz." / DOB-Remap: „⚠ Geburtsdatum 22.08.2018 verworfen — per manueller
  Entscheidung korrigiert; 14.07.2019 verwendet." (Kazuya)

Import: generischer _format_discard()-Helfer; Verwerfungen aus Mehrheitsentscheid,
pick_parent_ref-Ablehnungen (explain_pick_rejections), Rollen-Normalisierung,
Eltern-Alter-Sanity-Check und DOB-Remaps werden über _discarded/_prov_args an die
Tiere gehängt; Gerbil-Provenance wird in einem finalen Pass gebaut (nachdem alle
Verwerfungen feststehen). Frontend hebt ⚠-Zeilen ab (+ Legende).

Tests grün (python merge/extract, vitest 129, playwright 8, dotnet 212).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

gerbil-manager-web/e2e/mock-data.ts

@@ -135,7 +135,11 @@ export function seedDb(): MockDb {
           'Geburtsdatum (12.03.2025) aus „Stammbaum von Krümel.xlsx“.',
           'Genotyp aus „Stammbaum von Krümel.xlsx“.',
           'Auch in „Wurfchronik-Detail.docx“ gefunden → Datensätze zusammengeführt.',
+          // VERWORFEN: abweichender Wert aus einer Quelle wurde verworfen (Grund + Ersatz).
+          '⚠ Geburtsdatum 14.06.2015 aus „Wurfchronik-Detail.docx“ verworfen — abweichend; 12.03.2025 aus „Stammbaum von Krümel.xlsx“ verwendet (Mehrheit).',
           'Eltern über Position im Stammbaum erkannt (Quelle: „Stammbaum von Krümel.xlsx“).',
+          // VERWORFEN: unplausibler Elternteil verworfen, kein Ersatz verwendet.
+          '⚠ Vater „Jayjay“ (*19.06.2013) verworfen — unplausibel (9 Jahre älter als das Kind); kein Ersatz.',
           'Angaben aus der Wurfchronik übernommen.',
         ],
       }),

gerbil-manager-web/e2e/provenance.spec.ts

@@ -21,6 +21,17 @@ test('Tierakte: "Datenherkunft"-Button öffnet den Nachverfolgungs-Dialog', asyn
     'Auch in „Wurfchronik-Detail.docx“ gefunden → Datensätze zusammengeführt.',
   )
 
+  // VERWORFEN: eine Quelle wurde verworfen — mit Grund und verwendetem Ersatz.
+  await expect(dialog).toContainText(p.discardLegend)
+  await expect(dialog).toContainText(
+    '⚠ Geburtsdatum 14.06.2015 aus „Wurfchronik-Detail.docx“ verworfen — abweichend; 12.03.2025 aus „Stammbaum von Krümel.xlsx“ verwendet (Mehrheit).',
+  )
+  await expect(dialog).toContainText(
+    '⚠ Vater „Jayjay“ (*19.06.2013) verworfen — unplausibel (9 Jahre älter als das Kind); kein Ersatz.',
+  )
+  // Die Verwerfen-Zeile ist visuell als solche markiert (eigene CSS-Klasse).
+  await expect(dialog.locator('.provenance__history-step--discard').first()).toBeVisible()
+
   // Quelldateien werden angezeigt.
   await expect(dialog).toContainText(p.sourceFilesTitle)
   await expect(dialog).toContainText('Stammbaum von Krümel.xlsx')

gerbil-manager-web/src/components/ProvenanceDialog.tsx

@@ -12,6 +12,11 @@ import { de } from '../strings/de'
 import type { EntityProvenance } from '../api/types'
 import './provenanceDialog.css'
 
+/** Leading marker the Python merge prefixes onto discard ("verworfen") history
+ *  lines (data thrown away with reason + replacement). Kept in sync with
+ *  DISCARD_MARK in tools/import/merge_and_resolve.py. */
+const DISCARD_PREFIX = '⚠ '
+
 interface ProvenanceDialogProps {
   open: boolean
   onClose: () => void
@@ -92,12 +97,27 @@ export default function ProvenanceDialog({ open, onClose, provenance, entityLabe
             {p.history.length > 0 && (
               <section className="provenance__section">
                 <div className="provenance__section-title">{t.historyTitle}</div>
+                {p.history.some((s) => s.startsWith(DISCARD_PREFIX)) && (
+                  <p className="provenance__discard-legend">{t.discardLegend}</p>
+                )}
                 <ol className="provenance__history">
-                  {p.history.map((step, i) => (
+                  {p.history.map((step, i) => {
-                    <li key={`${i}-${step}`} className="provenance__history-step">
+                    // Discard lines (Python prefixes them with „⚠ “) are styled
-                      {step}
+                    // distinctly so a thrown-away value stands out from accepted facts.
-                    </li>
+                    const isDiscard = step.startsWith(DISCARD_PREFIX)
-                  ))}
+                    return (
+                      <li
+                        key={`${i}-${step}`}
+                        className={
+                          isDiscard
+                            ? 'provenance__history-step provenance__history-step--discard'
+                            : 'provenance__history-step'
+                        }
+                      >
+                        {step}
+                      </li>
+                    )
+                  })}
                 </ol>
               </section>
             )}

gerbil-manager-web/src/components/provenanceDialog.css

@@ -163,6 +163,19 @@
   word-break: break-word;
 }
 
+/* Verworfene Werte (Datenherkunft): hervorgehoben, damit klar wird, dass hier
+   Daten aus einer Quelle verworfen wurden — mit Grund und Ersatz. */
+.provenance__history-step--discard {
+  color: var(--color-danger, #9a3412);
+  font-style: italic;
+}
+
+.provenance__discard-legend {
+  margin: 0 0 0.5rem;
+  font-size: 0.78rem;
+  color: var(--color-text-muted, #888);
+}
+
 .provenance__actions {
   display: flex;
   justify-content: flex-end;

gerbil-manager-web/src/strings/de.ts

@@ -896,6 +896,9 @@ export const de = {
     },
     /** Überschrift des chronologischen, dateibezogenen Verlaufs (Primärinhalt). */
     historyTitle: 'Verlauf',
+    /** Legende für die mit „⚠“ markierten Verlaufszeilen, in denen Daten aus
+     *  einer Quelle verworfen wurden (mit Grund und verwendetem Ersatz). */
+    discardLegend: '⚠ markiert verworfene Quelldaten (mit Grund und verwendetem Ersatz).',
     /** Überschriften / Feldbeschriftungen. */
     sourceFilesTitle: 'Quelldateien',
     sourceFilesCount: (n: number) => (n === 1 ? 'aus 1 Quelle' : `aus ${n} Quellen`),

tools/import/extract.py

@@ -774,6 +774,9 @@ def dedup(animals):
             # FEAT-8c: machine-readable quarantine marker so the API loader can skip
             # conflicting records without parsing the German review report.
             "conflict": False,
+            # DOB-Remap (manuelle Entscheidung): original/corrected birthdate so the
+            # downstream Datenherkunft can record the discarded original value.
+            "dobRemap": next((a["dobRemap"] for a in grp if a.get("dobRemap")), None),
         }
         merged.append(out)
         # conflict: same animal, GENUINELY disagreeing genotype (presence-vs-absence is NOT a
@@ -1037,6 +1040,9 @@ def apply_dob_remaps(raw_animals, path):
         dob = norm_dob(a.get("dob", ""))
         new = remaps_full.get((nc, zc, dob)) or remaps_name.get((nc, dob))
         if new and a.get("dob") != new:
+            # Keep the discarded original DOB so the provenance/Datenherkunft can
+            # show „Geburtsdatum X verworfen, per Entscheidung auf Y geändert.“
+            a["dobRemap"] = {"original": a.get("dob"), "corrected": new}
             a["dob"] = new
             n += 1
     return n

tools/import/merge_and_resolve.py

@@ -713,8 +713,63 @@ def pick_parent_ref(parent_refs, role, child_dob, avoid_name=None, gender_of=Non
     return role_refs[order[0]]
 
 
+def explain_pick_rejections(parent_refs, role, child_dob, chosen, avoid_name=None,
+                            gender_of=None):
+    """Explain why other refs for `role` lost to `chosen` in pick_parent_ref.
+
+    Returns a list of discard dicts (for _format_discard) — one per distinct
+    rejected candidate name that was beaten for a clear reason (wrong sex, age-
+    impossible, or duplicate of the other parent). Mirrors pick_parent_ref's
+    ranking so the history can explain the same decision it made.
+    """
+    role_refs = [p for p in parent_refs if p.get("roleGuess") == role]
+    if not role_refs or chosen is None:
+        return []
+    avoid = normalize_name(avoid_name) if avoid_name else None
+    expected = "male" if role == "father" else "female"
+    role_de = "Vaterrolle" if role == "father" else "Mutterrolle"
+    cand_de = "Vater-Kandidat" if role == "father" else "Mutter-Kandidat"
+    chosen_name = chosen.get("name")
+    repl_disp = chosen_name
+    if chosen.get("dob"):
+        repl_disp = f"{chosen_name} (*{_de_date(parse_date(chosen.get('dob'))) or chosen.get('dob')})"
+
+    out = []
+    seen = set()
+    for p in role_refs:
+        name = p.get("name")
+        if not name or normalize_name(name) == normalize_name(chosen_name or ""):
+            continue
+        key = normalize_name(name)
+        if key in seen:
+            continue
+        g = gender_of(name) if gender_of else None
+        dob = p.get("dob")
+        reason = None
+        if avoid is not None and key == avoid:
+            reason = "bereits als anderer Elternteil gewählt"
+        elif g in ("male", "female") and g != expected:
+            reason = f"falsches Geschlecht für die {role_de}"
+        elif dob and not parent_age_plausible(dob, child_dob):
+            reason = "unplausibles Alter für diesen Wurf"
+        if reason is None:
+            continue
+        seen.add(key)
+        disp = name
+        if dob:
+            disp = f"{name} (*{_de_date(parse_date(dob)) or dob})"
+        out.append({
+            "label": cand_de,
+            "value": f"„{disp}“",
+            "reason": reason,
+            "replacement": f"„{repl_disp}“",
+        })
+    return out
+
+
 def _build_gerbil_history(records, best_g, field_source, parent_method=None,
-                          any_decision=False, any_conflict=False, conflict_notes=None):
+                          any_decision=False, any_conflict=False, conflict_notes=None,
+                          field_discards=None, parent_discards=None):
     """Build an ordered, file-attributed German history for a resolved gerbil.
 
     Reads like a chronological log:
@@ -773,6 +828,13 @@ def _build_gerbil_history(records, best_g, field_source, parent_method=None,
         else:
             history.append("In weiterem Datensatz gefunden → Datensätze zusammengeführt.")
 
+    # Discarded field values from majority-vote conflict resolution: the LOSING
+    # values, their file, and what won instead.
+    for d in (field_discards or []):
+        line = _format_discard(d)
+        if line not in history:
+            history.append(line)
+
     # Parent derivation.
     if parent_method:
         method_label = {
@@ -791,6 +853,14 @@ def _build_gerbil_history(records, best_g, field_source, parent_method=None,
         else:
             history.append(f"Eltern über {method_label} erkannt.")
 
+    # Discarded parent candidates / links (pick_parent_ref rejections, role
+    # normalization drops, parent-age sanity check). Threaded in after the merge
+    # via the gerbil's _discarded list so they read in chronological order.
+    for d in (parent_discards or []):
+        line = _format_discard(d)
+        if line not in history:
+            history.append(line)
+
     # Manual decisions / conflicts.
     if any_decision:
         history.append("Zuordnung per manueller Entscheidung getroffen.")
@@ -821,6 +891,59 @@ def _de_gender(g):
     return {"male": "männlich", "female": "weiblich"}.get(g, g)
 
 
+# Leading marker that visually flags a discard ("data was thrown away") line in
+# the history timeline. The frontend keys discard styling off this marker.
+DISCARD_MARK = "⚠ "
+
+
+def _format_discard(d):
+    """Render one discard record into a German history line.
+
+    A discard record is a dict describing a value/candidate the pipeline threw
+    away. Recognised keys:
+      • text        — a fully pre-formatted line (used verbatim, marker added)
+      • label       — German field label (e.g. „Geburtsdatum“)
+      • value       — the discarded value (already display-formatted)
+      • file        — source file the discarded value came from (attributed)
+      • reason      — why it was dropped (e.g. „abweichend“, „unplausibel …“)
+      • replacement — what was used instead (already display-formatted)
+      • repl_file   — source file the replacement came from
+    Produces lines like:
+      „⚠ Geburtsdatum 14.06.2015 aus ‚A.xlsx‘ verworfen — abweichend;
+        14.06.2017 aus ‚B.xlsx‘ verwendet (Mehrheit).“
+    Generic across entity types so litters/contacts can reuse it.
+    """
+    if d.get("text"):
+        return DISCARD_MARK + d["text"]
+    parts = []
+    label = d.get("label")
+    value = d.get("value")
+    if label and value is not None:
+        parts.append(f"{label} {value}")
+    elif label:
+        parts.append(str(label))
+    elif value is not None:
+        parts.append(str(value))
+    head = " ".join(parts) if parts else "Wert"
+    if d.get("file"):
+        head += f" aus {_quote_file(d['file'])}"
+    line = f"{head} verworfen"
+    if d.get("reason"):
+        line += f" — {d['reason']}"
+    repl = d.get("replacement")
+    if repl is not None and repl != "":
+        instead = str(repl)
+        if d.get("repl_file"):
+            instead += f" aus {_quote_file(d['repl_file'])}"
+        suffix = d.get("replacement_note")
+        line += f"; {instead} verwendet"
+        if suffix:
+            line += f" ({suffix})"
+    elif d.get("no_replacement"):
+        line += "; kein Ersatz"
+    return DISCARD_MARK + line + "."
+
+
 def main():
     print("Loading color variety seeds...")
     variety_map = {}
@@ -864,6 +987,14 @@ def main():
     else:
         print(f"Warning: Seeds path not found at {SEEDS_PATH}")
 
+    # Reverse map (variety GUID → human name) for discard/replacement history
+    # lines that need to show a colour value rather than a raw GUID. First wins
+    # so we keep the canonical lower-case catalog name.
+    variety_id_to_name = {}
+    for name_lower, vid in variety_map.items():
+        if vid not in variety_id_to_name:
+            variety_id_to_name[vid] = name_lower
+
     md_files = sorted([f for f in os.listdir(DIR_PATH) if f.lower().endswith('.md')])
     print(f"Found {len(md_files)} markdown files in {DIR_PATH}.")
 
@@ -1187,7 +1318,20 @@ def main():
         mother_ref = pick_parent_ref(parent_refs, "mother", child_dob_raw,
                                      avoid_name=father_ref.get("name") if father_ref else None,
                                      gender_of=gender_of_name)
-        
+
+        # Record rejected parent-ref candidates (wrong sex / age-impossible /
+        # duplicate of the other role) so this animal's gerbil history can
+        # explain which Stammbaum positions were discarded and what won instead.
+        pick_discards = []
+        pick_discards += explain_pick_rejections(parent_refs, "father", child_dob_raw,
+                                                 father_ref, gender_of=gender_of_name)
+        pick_discards += explain_pick_rejections(
+            parent_refs, "mother", child_dob_raw, mother_ref,
+            avoid_name=father_ref.get("name") if father_ref else None,
+            gender_of=gender_of_name)
+        if pick_discards:
+            a["_pick_discards"] = pick_discards
+
         a["_mapped_litter_scoped_id"] = None
         if father_ref and mother_ref:
             f_name = get_normalized_gerbil_name(father_ref.get("name"))
@@ -1623,6 +1767,23 @@ def main():
             "_old_id": rg.get("Id") or rg.get("id")
         })
 
+    # Build the discard list for one stammbaum animal: rejected parent-ref
+    # candidates (pick_parent_ref) plus a manual DOB-remap decision, if any.
+    def _stammbaum_discards(a):
+        discards = list(a.get("_pick_discards", []))
+        remap = a.get("dobRemap")
+        if remap and remap.get("original") and remap.get("corrected"):
+            orig = _de_date(parse_date(remap["original"])) or remap["original"]
+            corr = _de_date(parse_date(remap["corrected"])) or remap["corrected"]
+            if orig != corr:
+                discards.append({
+                    "label": PROV_FIELD_LABELS["DateOfBirth"],
+                    "value": orig,
+                    "reason": "per manueller Entscheidung korrigiert",
+                    "replacement": corr,
+                })
+        return discards
+
     # Map and append stammbaum animals to all_processed_gerbils
     for a in stammbaum_only_animals:
         a_id = a["id"]
@@ -1731,7 +1892,10 @@ def main():
             "_eff_dob": dob_val or "2010-01-01",
             "_birth_date": dob_val,
             "_filename": a.get("sourceFiles", ["Stammbaum"])[0],
-            "_old_id": a_id
+            "_old_id": a_id,
+            # Rejected Stammbaum parent-ref candidates for this animal (filled by
+            # pick_parent_ref above) — surfaced in this gerbil's discard history.
+            "_discarded": _stammbaum_discards(a),
         })
 
     # Map and append docx animals to all_processed_gerbils
@@ -1885,7 +2049,8 @@ def main():
                     
         return True
 
-    def build_provenance(records, best_g, extra_notes=None, field_source=None):
+    def build_provenance(records, best_g, extra_notes=None, field_source=None,
+                         field_discards=None):
         """Aggregate data-provenance across every raw record merged into one
         resolved gerbil. Returns a JSON string (stored on the Gerbil entity as a
         nullable text column) so the Rennmausakte can show where the entry came
@@ -1937,6 +2102,8 @@ def main():
             any_decision=any_decision,
             any_conflict=any_conflict,
             conflict_notes=extra_notes or [],
+            field_discards=field_discards,
+            parent_discards=best_g.get("_discarded"),
         )
 
         return build_entity_provenance(
@@ -1971,7 +2138,9 @@ def main():
         if is_placeholder:
             # Placeholders: do NOT merge, keep all separate
             for g in group:
-                g["Provenance"] = build_provenance([g], g)
+                # Provenance is built in a final pass (after parent resolution),
+                # so parent-link discards land in the right gerbil's history.
+                g["_prov_args"] = ([g], g, None, None, None)
                 resolved_gerbils.append(g)
                 gerbil_id_map[g["Id"]] = g["Id"]
             continue
@@ -1992,7 +2161,7 @@ def main():
         for sub in sub_groups:
             if len(sub) == 1:
                 g = sub[0]
-                g["Provenance"] = build_provenance([g], g)
+                g["_prov_args"] = ([g], g, None, None, None)
                 resolved_gerbils.append(g)
                 gerbil_id_map[g["Id"]] = g["Id"]
                 continue
@@ -2019,6 +2188,17 @@ def main():
             if best_g["Notes"]:
                 merged_notes.append(best_g["Notes"])
 
+            # Carry over discard records (rejected parent-refs / dob remaps) from
+            # every merged record so none are lost when the primary changes.
+            merged_discards = list(best_g.get("_discarded") or [])
+            for g in sub:
+                if g is best_g:
+                    continue
+                for d in (g.get("_discarded") or []):
+                    if d not in merged_discards:
+                        merged_discards.append(d)
+            best_g["_discarded"] = merged_discards
+
             # Merge photos
             merged_photos = list(best_g.get("_photos", []))
 
@@ -2094,8 +2274,21 @@ def main():
                      if not any(kw in g["Notes"].lower() for kw in ["parent listed", "mutter von", "vater von", "dam of", "sire of"]):
                          merged_notes.append(g["Notes"])
 
+            # Display formatter per field for discard/replacement lines.
+            def _disp(field, val):
+                if val is None or val == "" or val == "unknown":
+                    return None
+                if field in ("DateOfBirth", "DateOfDeath"):
+                    return _de_date(val)
+                if field == "Gender":
+                    return _de_gender(val)
+                if field == "ColorVarietyId":
+                    return variety_id_to_name.get(val, "Farbschlag")
+                return str(val)
+
             # Reconcile fields based on number of source files supporting them
             conflict_notes = []
+            field_discards = []
             for field in ["DateOfBirth", "DateOfDeath", "Gender", "Genotype", "ColorVarietyId"]:
                 votes = {}
                 for g in sub:
@@ -2118,6 +2311,25 @@ def main():
                     winner = next((g for g in sub if g.get(field) == best_val), None)
                     if winner is not None:
                         field_source[field] = winner
+                    # Record each LOSING value: which file it came from, that it
+                    # was discarded as differing, and that the majority value (and
+                    # its file) was used instead.
+                    if len(votes) > 1:
+                        repl_disp = _disp(field, best_val)
+                        repl_file = _primary_file_of(winner) if winner is not None else None
+                        for lose_val in votes:
+                            if lose_val == best_val:
+                                continue
+                            loser = next((g for g in sub if g.get(field) == lose_val), None)
+                            field_discards.append({
+                                "label": PROV_FIELD_LABELS.get(field, field),
+                                "value": _disp(field, lose_val),
+                                "file": _primary_file_of(loser) if loser is not None else None,
+                                "reason": "abweichend",
+                                "replacement": repl_disp,
+                                "repl_file": repl_file,
+                                "replacement_note": "Mehrheit",
+                            })
                     # Keep helper fields in sync if we changed DateOfBirth
                     if field == "DateOfBirth":
                         best_g["_birth_date"] = best_val
@@ -2127,9 +2339,9 @@ def main():
                 best_g["Notes"] = " | ".join(merged_notes)
 
             best_g["_photos"] = merged_photos
-            best_g["Provenance"] = build_provenance(
+            # Defer provenance to a final pass so parent-link discards (added
-                sub, best_g, extra_notes=conflict_notes, field_source=field_source
+            # later by the parent-age / role-normalization passes) are included.
-            )
+            best_g["_prov_args"] = (sub, best_g, conflict_notes, field_source, field_discards)
 
             # Print merge trace
             print(f"Deduplicated same-animal name '{best_g['Name']}': merged {len(sub)} entries across files: {', '.join(sources)}")
@@ -2294,6 +2506,25 @@ def main():
         del l["_mother_name"]
         del l["_filename"]
 
+    # Children-of-litter lookup so a dropped parent link can be explained in the
+    # offspring's Datenherkunft (the discard is most meaningful on the child).
+    children_by_litter = {}
+    for g in resolved_gerbils:
+        lid = g.get("LitterId")
+        if lid:
+            children_by_litter.setdefault(lid, []).append(g)
+
+    def _add_child_discard(litter, discard):
+        """Attach a parent-link discard to every child gerbil of the litter."""
+        for child in children_by_litter.get(litter["Id"], []):
+            dl = child.setdefault("_discarded", [])
+            if discard not in dl:
+                dl.append(discard)
+
+    def _pname(pid):
+        p = gerbil_by_id_final.get(pid)
+        return p.get("Name") if p else None
+
     # Role normalization: assign each resolved parent to the role matching its
     # gender, eliminate self-pairings (same animal in both roles), and never let
     # impossible duplicates survive (two males / two females). This corrects
@@ -2301,9 +2532,23 @@ def main():
     # (one parent of "unknown" gender, or a self-paired litter).
     role_fixes = 0
     for l in resolved_litters:
+        before = (l.get("FatherId"), l.get("MotherId"))
         father, mother = assign_parent_roles(l.get("FatherId"), l.get("MotherId"), _final_gender)
-        if (l.get("FatherId"), l.get("MotherId")) != (father, mother):
+        if before != (father, mother):
             role_fixes += 1
+            after = {father, mother}
+            # A parent id present before but gone after was dropped by role
+            # normalization (self-pairing or two-of-the-same-sex). Explain it.
+            for pid in before:
+                if pid and pid not in after:
+                    pname = _pname(pid)
+                    if before[0] == before[1]:
+                        reason = "Selbstverpaarung — Tier kann nicht beide Elternteile sein"
+                    else:
+                        reason = "ein Wurf hat nur einen Vater und eine Mutter"
+                    _add_child_discard(l, {
+                        "text": f"Elternteil „{pname}“ verworfen — {reason}",
+                    })
         l["FatherId"] = father
         l["MotherId"] = mother
     if role_fixes:
@@ -2323,6 +2568,23 @@ def main():
             p = gerbil_by_id_final.get(pid)
             if p and not parent_age_plausible(p.get("DateOfBirth"), ldate):
                 age_drops.append((l.get("Name"), role, p.get("Name"), p.get("DateOfBirth"), ldate))
+                # Explain the drop in each child's history: which parent, its DOB,
+                # why (implausible age), and that no replacement was used.
+                role_de = "Vater" if role == "FatherId" else "Mutter"
+                pdob_disp = _de_date(p.get("DateOfBirth")) or "unbekannt"
+                age_reason = "unplausibles Alter für diesen Wurf"
+                pd = date_to_days(parse_date(p.get("DateOfBirth"))) if p.get("DateOfBirth") else None
+                ld = date_to_days(parse_date(ldate)) if ldate else None
+                if pd is not None and ld is not None:
+                    years = abs(ld - pd) / 365.25
+                    if ld - pd <= 0:
+                        age_reason = "unplausibel (nicht vor dem Kind geboren)"
+                    else:
+                        age_reason = f"unplausibel ({years:.0f} Jahre älter als das Kind)"
+                _add_child_discard(l, {
+                    "text": (f"{role_de} „{p.get('Name')}“ (*{pdob_disp}) verworfen "
+                             f"— {age_reason}; kein Ersatz"),
+                })
                 l[role] = None
     if age_drops:
         print(f"Parent-age sanity check: dropped {len(age_drops)} implausible parent link(s):")
@@ -2402,6 +2664,27 @@ def main():
     resolved_litters = deduped2
     litter_by_scoped_id = {l["Id"]: l for l in resolved_litters}
 
+    # Final gerbil-provenance pass: now that parent links are fully resolved and
+    # all discards (majority-vote conflicts during dedup; rejected parent-refs;
+    # role-normalization drops; parent-age drops) are attached to each gerbil's
+    # _discarded list, render the provenance JSON with the discard history lines.
+    n_discards = 0
+    for g in resolved_gerbils:
+        args = g.pop("_prov_args", None)
+        if g.get("_discarded"):
+            n_discards += len(g["_discarded"])
+        if args is not None:
+            records, best_g, conflict_notes, field_source, field_discards = args
+            g["Provenance"] = build_provenance(
+                records, best_g, extra_notes=conflict_notes,
+                field_source=field_source, field_discards=field_discards,
+            )
+        else:
+            g["Provenance"] = build_provenance([g], g)
+        g.pop("_discarded", None)
+    if n_discards:
+        print(f"Datenherkunft: recorded {n_discards} discard line(s) across gerbils.")
+
     # Datenherkunft for litters: which source files contributed, whether this is
     # a Wurfchronik litter vs a Stammbaum-reconstructed ("virtual") litter, how
     # many raw records merged into it, plus human-readable notes. Accumulators

tools/import/test_merge_resolve.py

@@ -277,6 +277,74 @@ check("_de_date: ISO → DD.MM.YYYY", m._de_date("2022-03-27") == "27.03.2022")
 check("_de_date: passes through non-ISO", m._de_date("unbekannt") == "unbekannt")
 
 
+# ── Discard history: a discarded source value records reason + replacement ──
+# _format_discard: majority-vote conflict (losing value + file → winner + file).
+_d_mehr = m._format_discard({
+    "label": "Geburtsdatum", "value": "14.06.2015", "file": "A.xlsx",
+    "reason": "abweichend", "replacement": "14.06.2017", "repl_file": "B.xlsx",
+    "replacement_note": "Mehrheit",
+})
+check("discard: starts with warning marker", _d_mehr.startswith(m.DISCARD_MARK))
+check("discard(majority): names losing value + its file",
+      "Geburtsdatum 14.06.2015 aus „A.xlsx“ verworfen" in _d_mehr)
+check("discard(majority): states the reason", "— abweichend" in _d_mehr)
+check("discard(majority): names replacement + its file + note",
+      "14.06.2017 aus „B.xlsx“ verwendet (Mehrheit)." in _d_mehr)
+
+# _format_discard: a parent dropped with NO replacement.
+_d_noerepl = m._format_discard({
+    "text": "Vater „Jayjay“ (*19.06.2013) verworfen — unplausibel (9 Jahre älter "
+            "als das Kind); kein Ersatz",
+})
+check("discard(text): verbatim text gets the warning marker",
+      _d_noerepl == m.DISCARD_MARK + "Vater „Jayjay“ (*19.06.2013) verworfen — "
+      "unplausibel (9 Jahre älter als das Kind); kein Ersatz")
+
+# explain_pick_rejections: a wrong-sex father candidate is explained (Molly case).
+_picks = m.explain_pick_rejections(
+    molly_refs, "father", "13.09.2021", fr, gender_of=gof,
+)
+_pick_father = next((d for d in _picks if "Danielle" in (d.get("value") or "")), None)
+check("pick-reject: wrong-sex father candidate is recorded",
+      _pick_father is not None)
+check("pick-reject: reason = wrong sex for the father role",
+      _pick_father and "falsches Geschlecht für die Vaterrolle" in _pick_father["reason"])
+check("pick-reject: replacement names the chosen Hagrid",
+      _pick_father and "Hagrid" in (_pick_father.get("replacement") or ""))
+
+# explain_pick_rejections: an age-impossible candidate is explained.
+_age_refs = [pref("Old", "father", "2010-01-01"), pref("Dad", "father", "2021-01-01")]
+_chosen = m.pick_parent_ref(_age_refs, "father", "2022-03-27")
+_age_picks = m.explain_pick_rejections(_age_refs, "father", "2022-03-27", _chosen)
+check("pick-reject(age): age-impossible candidate recorded with reason",
+      any("unplausibles Alter" in d["reason"] for d in _age_picks))
+
+# _build_gerbil_history threads field_discards (after merge) and parent_discards
+# (after the parent line) into the timeline.
+g_disc = {
+    "Name": "Solice", "DateOfBirth": "2022-03-27", "Gender": "male",
+    "Genotype": None, "ColorVarietyId": None, "DateOfDeath": None,
+    "ImportSource": "Stammbaum.xlsx", "_filename": "Stammbaum.xlsx",
+    "parentRefs": [],
+    "_discarded": [{"text": "Vater „Jayjay“ verworfen — unplausibel; kein Ersatz"}],
+}
+h_disc = m._build_gerbil_history(
+    [g_disc], g_disc, {},
+    field_discards=[{
+        "label": "Geschlecht", "value": "weiblich", "file": "Wurfchronik.docx",
+        "reason": "abweichend", "replacement": "männlich", "repl_file": "Stammbaum.xlsx",
+        "replacement_note": "Mehrheit",
+    }],
+    parent_discards=g_disc["_discarded"],
+)
+check("history: field-discard line present (majority vote)",
+      any("Geschlecht weiblich aus „Wurfchronik.docx“ verworfen" in s for s in h_disc))
+check("history: parent-discard line present (dropped parent)",
+      any("Vater „Jayjay“ verworfen" in s for s in h_disc))
+check("history: discard lines carry the warning marker",
+      all(s.startswith(m.DISCARD_MARK) for s in h_disc if "verworfen" in s))
+
+
 if check.failed:
     print(f"\n{check.failed} test(s) FAILED")
     sys.exit(1)