Give a point-level correction its own coordinate verdict

README.md

@@ -200,12 +200,12 @@ What each side may rely on:
   `mirror_of`.
 - **Grouping is a proposal; pondie's verdict overrides it.** When pondie
   disagrees, it writes a revision: a complete `CoordinateParse` with `revision_of` set and
-  one `AnalysisVerdict` per original analysis (accept, relabel, split, merge, omit, each
-  with a reason and evidence). Accepted and relabelled analyses keep their keys. A split
-  or merge copies the claims and votes stored against the old keys to their replacements,
-  each copy noting where it came from; the revised version is kept unchanged as history
-  and left out of projections (`AnalysisVerdict` gives the rules). Region,
-  seed and target sets given only as definitions are relabelled `anchor` and become
+  one `AnalysisVerdict` per original analysis (accept, relabel, edit, split, merge, omit, each
+  with a reason and evidence). Accepted and relabelled analyses keep their keys. An edit
+  corrects a single analysis's points, so its key changes. An edit, split or merge copies the
+  claims and votes stored against the old keys to their replacements, each copy noting where it came from; the revised version is
+  kept unchanged as history and left out of projections (`AnalysisVerdict` gives the rules).
+  Region, seed and target sets given only as definitions are relabelled `anchor` and become
   CoordinateSets in the record, not Analyses with statistics.
 - **Uploads are asymmetric.** neurostore accepts a parse alone, a parse with a record, a
   record alone against a parse already stored (every verdict accept), or a revision with a

neuroimaging-paper-parse/coordinates.yaml

@@ -503,6 +503,8 @@ classes:
 
       - accept, relabel: nothing moves. The key is unchanged, so everything stored against
         it applies to the revision as it stands.
+      - edit: everything is copied to the one analysis in `replaced_by`, whose points are
+        the corrected ones.
       - split: everything is copied to each analysis in `replaced_by`.
       - merge: everything stored against each merged analysis is copied to the one
         analysis they merge into. Two copies of the same field with the same value become
@@ -530,7 +532,8 @@ classes:
         multivalued: true
         description: >-
           The keys in this revision that take the analysis's place: itself for accept and
-          relabel, the replacements for split and merge, none for omit.
+          relabel, the one corrected analysis for edit, the replacements for split and merge,
+          none for omit.
       role:
         any_of:
           - range: CoordinateRole

neuroimaging-paper-parse/parse_enums.yaml

@@ -193,6 +193,11 @@ enums:
           A reviewed correction of the ingestion workflow's proposed role, which wins: the same points
           in a different role, such as a region of interest, a seed, a stimulation target
           where the proposal said result. Its key is unchanged.
+      edit:
+        description: >-
+          The same analysis, with its points corrected: a value misread, a sign dropped, a point
+          added to this analysis, removed or moved. Its key changes because its content did.
+          Reordering points is not an edit.
       split:
         description: The points belong to more than one analysis. The replacements have new keys.
       merge:

src/study_schema/jsonschema/coordinate-parse.schema.json

@@ -25,7 +25,7 @@
     },
     "AnalysisVerdict": {
       "additionalProperties": false,
-      "description": "pondie's verdict on one analysis of the parse a revision revises, reached while extracting the whole record. The verdict overrides the original's proposal, grouping and role alike.\nWhat a verdict does to what is stored against the analysis -- claims, their evidence, and users' votes on them:\n- accept, relabel: nothing moves. The key is unchanged, so everything stored against\n  it applies to the revision as it stands.\n- split: everything is copied to each analysis in `replaced_by`. - merge: everything stored against each merged analysis is copied to the one\n  analysis they merge into. Two copies of the same field with the same value become\n  one claim; with different values they stay competing claims, resolved like any\n  other.\n- omit: nothing is copied. - add: nothing to copy; the analysis is new.\nEvery copy records the key and parse it was copied from, so a reviewer can tell a copied claim from one made against the revision. The revised parse and everything stored against it are kept, unchanged, as history: superseded rather than deleted, and left out of projections and studysets.",
+      "description": "pondie's verdict on one analysis of the parse a revision revises, reached while extracting the whole record. The verdict overrides the original's proposal, grouping and role alike.\nWhat a verdict does to what is stored against the analysis -- claims, their evidence, and users' votes on them:\n- accept, relabel: nothing moves. The key is unchanged, so everything stored against\n  it applies to the revision as it stands.\n- edit: everything is copied to the one analysis in `replaced_by`, whose points are\n  the corrected ones.\n- split: everything is copied to each analysis in `replaced_by`. - merge: everything stored against each merged analysis is copied to the one\n  analysis they merge into. Two copies of the same field with the same value become\n  one claim; with different values they stay competing claims, resolved like any\n  other.\n- omit: nothing is copied. - add: nothing to copy; the analysis is new.\nEvery copy records the key and parse it was copied from, so a reviewer can tell a copied claim from one made against the revision. The revised parse and everything stored against it are kept, unchanged, as history: superseded rather than deleted, and left out of projections and studysets.",
       "properties": {
         "evidence": {
           "description": "The sentences of the paper the verdict rests on.",
@@ -45,7 +45,7 @@
           "type": "string"
         },
         "replaced_by": {
-          "description": "The keys in this revision that take the analysis's place: itself for accept and relabel, the replacements for split and merge, none for omit.",
+          "description": "The keys in this revision that take the analysis's place: itself for accept and relabel, the one corrected analysis for edit, the replacements for split and merge, none for omit.",
           "items": {
             "type": "string"
           },
@@ -187,6 +187,7 @@
       "enum": [
         "accept",
         "relabel",
+        "edit",
         "split",
         "merge",
         "omit",

src/study_schema/jsonschema/parsed-paper.schema.json

@@ -25,7 +25,7 @@
     },
     "AnalysisVerdict": {
       "additionalProperties": false,
-      "description": "pondie's verdict on one analysis of the parse a revision revises, reached while extracting the whole record. The verdict overrides the original's proposal, grouping and role alike.\nWhat a verdict does to what is stored against the analysis -- claims, their evidence, and users' votes on them:\n- accept, relabel: nothing moves. The key is unchanged, so everything stored against\n  it applies to the revision as it stands.\n- split: everything is copied to each analysis in `replaced_by`. - merge: everything stored against each merged analysis is copied to the one\n  analysis they merge into. Two copies of the same field with the same value become\n  one claim; with different values they stay competing claims, resolved like any\n  other.\n- omit: nothing is copied. - add: nothing to copy; the analysis is new.\nEvery copy records the key and parse it was copied from, so a reviewer can tell a copied claim from one made against the revision. The revised parse and everything stored against it are kept, unchanged, as history: superseded rather than deleted, and left out of projections and studysets.",
+      "description": "pondie's verdict on one analysis of the parse a revision revises, reached while extracting the whole record. The verdict overrides the original's proposal, grouping and role alike.\nWhat a verdict does to what is stored against the analysis -- claims, their evidence, and users' votes on them:\n- accept, relabel: nothing moves. The key is unchanged, so everything stored against\n  it applies to the revision as it stands.\n- edit: everything is copied to the one analysis in `replaced_by`, whose points are\n  the corrected ones.\n- split: everything is copied to each analysis in `replaced_by`. - merge: everything stored against each merged analysis is copied to the one\n  analysis they merge into. Two copies of the same field with the same value become\n  one claim; with different values they stay competing claims, resolved like any\n  other.\n- omit: nothing is copied. - add: nothing to copy; the analysis is new.\nEvery copy records the key and parse it was copied from, so a reviewer can tell a copied claim from one made against the revision. The revised parse and everything stored against it are kept, unchanged, as history: superseded rather than deleted, and left out of projections and studysets.",
       "properties": {
         "evidence": {
           "description": "The sentences of the paper the verdict rests on.",
@@ -45,7 +45,7 @@
           "type": "string"
         },
         "replaced_by": {
-          "description": "The keys in this revision that take the analysis's place: itself for accept and relabel, the replacements for split and merge, none for omit.",
+          "description": "The keys in this revision that take the analysis's place: itself for accept and relabel, the one corrected analysis for edit, the replacements for split and merge, none for omit.",
           "items": {
             "type": "string"
           },
@@ -187,6 +187,7 @@
       "enum": [
         "accept",
         "relabel",
+        "edit",
         "split",
         "merge",
         "omit",

src/study_schema/layouts.py

@@ -194,6 +194,13 @@ def check_revision(original: CoordinateParse, revision: CoordinateParse) -> list
         if verdict.key in judged:
             problems.append(f"{verdict.key} has more than one verdict")
         judged.add(verdict.key)
+        if verdict.verdict == "edit" and len(verdict.replaced_by or []) != 1:
+            problems.append(f"edit {verdict.key}: replaced_by must name exactly one key")
+        if verdict.verdict == "edit" and verdict.replaced_by == [verdict.key]:
+            problems.append(
+                f"edit {verdict.key}: replaced_by names the analysis itself; "
+                "unchanged content is an accept"
+            )
         for key in verdict.replaced_by or []:
             if key not in after:
                 problems.append(

src/study_schema/models/paper_parse.py

@@ -296,6 +296,10 @@ class AuditVerdict(str, Enum):
     """
     A reviewed correction of the ingestion workflow's proposed role, which wins: the same points in a different role, such as a region of interest, a seed, a stimulation target where the proposal said result. Its key is unchanged.
     """
+    edit = "edit"
+    """
+    The same analysis, with its points corrected: a value misread, a sign dropped, a point added to this analysis, removed or moved. Its key changes because its content did. Reordering points is not an edit.
+    """
     split = "split"
     """
     The points belong to more than one analysis. The replacements have new keys.
@@ -1608,6 +1612,8 @@ class AnalysisVerdict(ConfiguredBaseModel):
     What a verdict does to what is stored against the analysis -- claims, their evidence, and users' votes on them:
     - accept, relabel: nothing moves. The key is unchanged, so everything stored against
       it applies to the revision as it stands.
+    - edit: everything is copied to the one analysis in `replaced_by`, whose points are
+      the corrected ones.
     - split: everything is copied to each analysis in `replaced_by`. - merge: everything stored against each merged analysis is copied to the one
       analysis they merge into. Two copies of the same field with the same value become
       one claim; with different values they stay competing claims, resolved like any
@@ -1617,7 +1623,7 @@ class AnalysisVerdict(ConfiguredBaseModel):
     """
     key: str = Field(default=..., description="""The revised parse's analysis key; for `add`, the new analysis's key in this revision.""")
     verdict: AuditVerdict = Field(default=...)
-    replaced_by: Optional[list[str]] = Field(default=None, description="""The keys in this revision that take the analysis's place: itself for accept and relabel, the replacements for split and merge, none for omit.""")
+    replaced_by: Optional[list[str]] = Field(default=None, description="""The keys in this revision that take the analysis's place: itself for accept and relabel, the one corrected analysis for edit, the replacements for split and merge, none for omit.""")
     role: Optional[Union[CoordinateRole, str]] = Field(default=None, description="""For relabel, the role the points have instead.""")
     reason: str = Field(default=...)
     evidence: Optional[list[TextSpan]] = Field(default=None, description="""The sentences of the paper the verdict rests on.""")

tests/test_layouts.py

@@ -84,3 +84,40 @@ def test_a_different_text_is_caught(tmp_path):
     (tmp_path / STUDY / "processed" / "pubget" / "text.txt").write_text("edited")
     paper = PaperParse.read(tmp_path / STUDY / "parse")
     assert any("text_sha256" in p for p in check_paper(paper, tmp_path / STUDY))
+
+
+def test_an_edit_names_exactly_one_replacement():
+    original, revision = _pair()
+    [split] = [v for v in revision.verdicts if v.verdict == "split"]
+    split.verdict = "edit"
+    assert any("edit" in p and "exactly one" in p for p in check_revision(original, revision))
+
+
+@pytest.mark.parametrize("replaced_by", [[], None])
+def test_an_edit_with_no_replacement_is_caught(replaced_by):
+    original, revision = _pair()
+    [split] = [v for v in revision.verdicts if v.verdict == "split"]
+    split.verdict = "edit"
+    split.replaced_by = replaced_by
+    assert any("edit" in p and "exactly one" in p for p in check_revision(original, revision))
+
+
+def test_an_edit_that_names_its_one_replacement_is_accepted():
+    original, revision = _pair()
+    [split] = [v for v in revision.verdicts if v.verdict == "split"]
+    split.verdict = "edit"
+    kept, dropped = split.replaced_by
+    split.replaced_by = [kept]
+    revision.analyses = [a for a in revision.analyses if a.key != dropped]
+    assert check_revision(original, revision) == []
+
+
+def test_an_edit_that_names_its_own_key_is_refused():
+    original, revision = _pair()
+    [split] = [v for v in revision.verdicts if v.verdict == "split"]
+    split.verdict = "edit"
+    split.replaced_by = [split.key]
+    assert (
+        f"edit {split.key}: replaced_by names the analysis itself; unchanged content is an accept"
+        in check_revision(original, revision)
+    )

tests/test_models.py

@@ -179,3 +179,24 @@ def test_names_differing_only_in_whitespace_share_a_key():
     assert analysis_key("PO > Sil", "MNI", points) == analysis_key(" PO>\u00a0Sil", "MNI", points)
     assert analysis_key("PO > Sil", "MNI", points) != analysis_key("PO < Sil", "MNI", points)
     assert analysis_key("PO > Sil", "MNI", points) != analysis_key("PC > Sil", "MNI", points)
+
+
+def _with_edit_verdict(example_paths):
+    data = load(example_paths["revision"])
+    [split] = [v for v in data["verdicts"] if v["verdict"] == "split"]
+    data["verdicts"] = [
+        {
+            "key": split["key"],
+            "verdict": "edit",
+            "replaced_by": split["replaced_by"][:1],
+            "reason": "A minus sign was dropped from one point of the analysis.",
+        }
+    ]
+    return data
+
+
+def test_an_edit_verdict_validates(example_paths):
+    data = _with_edit_verdict(example_paths)
+    assert CoordinateParse.model_validate(data).verdicts[0].verdict == "edit"
+    real = pytest.importorskip("jsonschema", reason="optional: JSON Schema validation")
+    real.validate(data, jsonschema.load("coordinate-parse"))