Update samorzad_gov_pl to de-identified upstream v0.2 (scoped from #15)

data/samorzad_gov_pl/NOTICE.md

@@ -7,8 +7,8 @@ platform.
 The platform footer states that textual content published in the service,
 excluding audiovisual content, is available under the [Creative Commons
 Attribution-ShareAlike 4.0 International](https://creativecommons.org/licenses/by-sa/4.0/)
-license (`CC-BY-SA-4.0`). Every source page retained in the published v0.1
-upstream dataset referenced the shared platform footer script.
+license (`CC-BY-SA-4.0`). Every page in the published v0.2 source dataset belongs
+to the v0.1 source set for which the shared footer reference was verified.
 
 For every released DynaWord document:
 
@@ -16,8 +16,9 @@ For every released DynaWord document:
 - `samorzad_gov_pl.attribution.jsonl` preserves the discovered `source_url`,
   fetched `final_url`, recommended `attribution_url`, article title,
   institution, tenant identifier, license URL and modification notice;
-- article text was extracted from source HTML, and this contribution normalizes
-  metadata and excludes short, non-Polish or duplicate records.
+- article text was extracted from source HTML and de-identified upstream;
+- this contribution normalizes metadata and excludes short, non-Polish or
+  duplicate records.
 
 When redistributing or adapting these texts, retain the source attribution,
 indicate changes and distribute adaptations under compatible

data/samorzad_gov_pl/samorzad_gov_pl.attribution.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e50acff3137b4680dc63a402667227b24cdd362e9c07895f1485a530ea5116c3
-size 68676369
+oid sha256:2def7488c34e749204842bcf21d33149738b34de2515115e3f79401f7eb271c4
+size 72438302

data/samorzad_gov_pl/samorzad_gov_pl.md

@@ -9,7 +9,7 @@ institutions and other public bodies.
 ## Dataset description
 
 - **Published source dataset:** https://huggingface.co/datasets/dawidmajewski/samorzad-gov-pl-articles
-- **Source revision:** `d5eaefc32c3f17cfa5a01d14573f8e4ee7d43385` (`v0.1`)
+- **Source revision:** `a998efea84bb0a62d94effc0b54f88b422d9898f` (`v0.2`)
 - **Original platform:** https://samorzad.gov.pl/
 - **Domain:** local government, education, social services and culture
 - **Language:** Polish (`pl`)
@@ -21,7 +21,7 @@ institutions and other public bodies.
 
 | read | kept | characters | tokens (`cl100k_base` proxy) | publishers |
 |---:|---:|---:|---:|---:|
-| 81,422 | 72,771 | 109,513,754 | 41,154,506 | 247 |
+| 81,418 | 72,472 | 107,130,266 | 40,011,099 | 247 |
 
 The released Parquet uses the canonical DynaWord schema:
 
@@ -32,40 +32,56 @@ institution in `author`. The names were obtained from the `og:site_name`
 metadata on each institution's platform homepage and frozen in
 `src/samorzad_gov_pl_publishers.json`.
 
-## Provenance
+## Provenance and de-identification
 
 The source release was collected directly from public samorzad.gov.pl pages.
 It contains every article URL found and successfully extracted during that
-crawl, rather than a topical or length-selected sample. The published source
-release contains 81,422 extracted articles from 247 institutions.
-
-`src/clean_samorzad_gov_pl.py` converts the published source Parquets into the
+crawl, rather than a topical or length-selected sample. Its v0.2 release
+contains 81,418 articles from 247 institutions after four corrupted source
+records were excluded.
+
+Consumer-facing text was de-identified upstream before conversion to DynaWord.
+Names were detected with the pinned Polish PII model
+[`ArkadiuszPawlak/fastpdn-ner-polish-pii`](https://huggingface.co/ArkadiuszPawlak/fastpdn-ner-polish-pii)
+at revision `636c57fe77b0afad8f065e7c76243dc051437035`. Deterministic
+rules and checksum validation covered structured identifiers and contact data.
+Removed values were replaced by typed placeholders such as `<PERSON>`,
+`<PHONE>`, `<EMAIL>` and `<URL>`.
+
+The upstream release audit scanned all 81,418 records for structured residuals,
+reran the PERSON model on an evenly spaced 1,000-record sample, validated
+placeholder structure, verified the 10 Parquet files by size and SHA-256, and
+checked record provenance against the pinned source manifest. All required
+release gates passed.
+
+`src/clean_samorzad_gov_pl.py` verifies and converts the v0.2 Parquets into the
 DynaWord schema, applies the common minimum-length and Polish-language checks,
-removes exact duplicate text and counts tokens with `cl100k_base`. The retained
-article text is otherwise unchanged.
+removes exact duplicate text and counts tokens with `cl100k_base`. Retained text
+is otherwise unchanged after upstream de-identification.
 
 ## Licensing — traceable basis and attribution
 
 The platform's shared footer states that textual content published in the
 service, excluding audiovisual content, is available under CC BY-SA 4.0. A
-release-time audit verified the live shared footer script and confirmed that
-all 81,422 preserved source pages referenced that script. The evidence URL,
-byte count and SHA-256 digest are recorded in
+release-time audit verified the live shared footer script for all 81,422 pages
+in the v0.1 source set. The four records omitted from v0.2 were a subset of that
+audited set, so every v0.2 article remains covered by the same evidence. The
+evidence URL, byte count and SHA-256 digest are recorded in
 `samorzad_gov_pl.license-evidence.json`.
 
 The fixed DynaWord schema does not contain article titles or URLs. Therefore,
 `samorzad_gov_pl.attribution.jsonl` maps every released `id` to its title,
 discovered `source_url`, fetched `final_url`, recommended `attribution_url`,
 publishing institution, platform tenant, license URL and modification notice.
-For 693 retained records the discovered and final URLs differ because the
+For 688 retained records the discovered and final URLs differ because the
 platform redirected the request; `attribution_url` uses the fetched final URL.
 
 Recommended attribution:
 
 > Publishing institution listed in the `author` column; article title and
 > canonical samorzad.gov.pl URL from the attribution sidecar; CC BY-SA 4.0.
-> Article text was extracted from HTML and normalized for inclusion in Polish
-> DynaWord.
+> Article text was extracted from HTML, de-identified upstream and normalized
+> for inclusion in Polish DynaWord.
 
 Photographs, recordings, videos and attachment bytes are not redistributed.
 The source platform gives audiovisual materials separate terms, and linked
@@ -76,11 +92,11 @@ not redistribute those files.
 
 | result | documents |
 |---|---:|
-| source records read | 81,422 |
-| shorter than 200 characters | 6,624 |
-| below the Polish-language heuristic | 35 |
-| exact duplicate text | 1,992 |
-| retained | 72,771 |
+| source records read | 81,418 |
+| shorter than 200 characters | 6,932 |
+| below the Polish-language heuristic | 34 |
+| exact duplicate text | 1,980 |
+| retained | 72,472 |
 
 Exact duplicate removal is performed after short-text and language checks.
 Existing DynaWord sources take priority when the same text is already present
@@ -88,11 +104,13 @@ elsewhere in the corpus.
 
 ## Limitations and recommended use
 
+- Automated de-identification reduces but cannot guarantee removal of every
+  identifying value; downstream users should apply controls appropriate to
+  their use case.
 - The collection is not a complete historical archive of samorzad.gov.pl. It
   contains the pages found and successfully extracted during one crawl.
-- Only 21,892 of the 81,422 source records expose a publication date. After
-  filtering, 52,214 retained records have an empty `created` value; no date is
-  inferred for them.
+- After filtering, 51,964 retained records have an empty `created` value; no
+  date is inferred for them.
 - The source text can contain spelling errors, extraction remnants or captions.
 - Very short source records are excluded by DynaWord's 200-character minimum.
 - Institution names describe the publishing body, not necessarily the person
@@ -112,18 +130,19 @@ elsewhere in the corpus.
 ## Reproduction
 
 The converter refuses to build unless the downloaded `manifest.json` and all
-10 Parquet shards match the pinned release. It verifies the manifest SHA-256,
-artifact set, shard sizes and SHA-256 values, row counts, schema, unique record
-IDs and the release-wide record-text digest before conversion. Outputs are
-written to temporary files, cross-checked, synced and transactionally replaced;
-a failed replacement restores the previous artifact set.
+10 Parquet shards match the pinned v0.2 release. It verifies the manifest
+SHA-256, artifact set, shard sizes and SHA-256 values, row counts, schema,
+unique record IDs and the release-wide record-text digest before conversion.
+Outputs are written to temporary files, cross-checked, synced and
+transactionally replaced; a failed replacement restores the previous artifact
+set.
 
 Download the published source Parquet shards and run:
 
 ```bash
 hf download dawidmajewski/samorzad-gov-pl-articles \
   --repo-type dataset \
-  --revision d5eaefc32c3f17cfa5a01d14573f8e4ee7d43385 \
+  --revision a998efea84bb0a62d94effc0b54f88b422d9898f \
   --include 'data/*.parquet' --include manifest.json \
   --local-dir /tmp/samorzad-gov-pl-articles
 

data/samorzad_gov_pl/samorzad_gov_pl.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b4f788030bb987b6c12487ea3e2e587bca1d1fd86294c1488961045bfe8fd137
-size 51930059
+oid sha256:3a683b4220b9b70698d7c524637d9ea5c1d72630b8f92ea7a871642a719cf2c0
+size 49612214

data/samorzad_gov_pl/samorzad_gov_pl.stats.json

@@ -1,25 +1,29 @@
 {
-  "authors_with_value": 72771,
+  "authors_with_value": 72472,
   "central_tenant_count": 1,
-  "chars": 109513754,
-  "cross_source_hashes_loaded": 4246372,
-  "drop_dup": 1992,
-  "drop_lang": 35,
+  "chars": 107130266,
+  "cross_source_hashes_loaded": 4255011,
+  "drop_dup": 1980,
+  "drop_lang": 34,
   "drop_ocr": 0,
-  "drop_short": 6624,
-  "kept": 72771,
+  "drop_short": 6932,
+  "kept": 72472,
   "license": "CC-BY-SA-4.0",
   "licenses": {
-    "CC-BY-SA-4.0": 72771
+    "CC-BY-SA-4.0": 72472
   },
   "local_public_tenant_count": 246,
   "publisher_count": 247,
-  "read": 81422,
-  "redirected_urls": 693,
-  "tokens": 41154506,
+  "read": 81418,
+  "redirected_urls": 688,
+  "tokens": 40011099,
   "upstream_artifact_count": 10,
   "upstream_dataset": "dawidmajewski/samorzad-gov-pl-articles",
-  "upstream_manifest_sha256": "97b5a266b2fd4f043648b7d3c757721cd826c7ca1048301761084b0abdee5c48",
-  "upstream_record_text_sha256": "6c35e1dc0bd35863ddb2f4e43d5bd08264c1fc27affdd60684ab5ecbc5457f63",
-  "upstream_revision": "d5eaefc32c3f17cfa5a01d14573f8e4ee7d43385"
+  "upstream_deidentification_model": "ArkadiuszPawlak/fastpdn-ner-polish-pii",
+  "upstream_deidentification_model_revision": "636c57fe77b0afad8f065e7c76243dc051437035",
+  "upstream_deidentification_policy": "samorzad-pii-v3",
+  "upstream_manifest_sha256": "9d349bb271033fef62a051e311115460e030740f9593bb7f3158cb1437f87639",
+  "upstream_record_text_sha256": "9378f130ce26fc95015f661910e660f8bbe6442e75640089185f78052533377d",
+  "upstream_revision": "a998efea84bb0a62d94effc0b54f88b422d9898f",
+  "upstream_source_quality_excluded_records": 4
 }

src/clean_samorzad_gov_pl.py

@@ -29,14 +29,14 @@ SOURCE = "samorzad_gov_pl"
 LICENSE = "CC-BY-SA-4.0"
 LICENSE_URL = "https://creativecommons.org/licenses/by-sa/4.0/"
 UPSTREAM_DATASET = "dawidmajewski/samorzad-gov-pl-articles"
-UPSTREAM_REVISION = "d5eaefc32c3f17cfa5a01d14573f8e4ee7d43385"
-UPSTREAM_TEXT_SHA256 = "6c35e1dc0bd35863ddb2f4e43d5bd08264c1fc27affdd60684ab5ecbc5457f63"
-UPSTREAM_MANIFEST_SHA256 = "97b5a266b2fd4f043648b7d3c757721cd826c7ca1048301761084b0abdee5c48"
+UPSTREAM_REVISION = "a998efea84bb0a62d94effc0b54f88b422d9898f"
+UPSTREAM_TEXT_SHA256 = "9378f130ce26fc95015f661910e660f8bbe6442e75640089185f78052533377d"
+UPSTREAM_MANIFEST_SHA256 = "9d349bb271033fef62a051e311115460e030740f9593bb7f3158cb1437f87639"
 UPSTREAM_EXPECTED = {
     "manifest_sha256": UPSTREAM_MANIFEST_SHA256,
     "dataset_id": UPSTREAM_DATASET,
     "schema_version": "1.0.0",
-    "record_count": 81_422,
+    "record_count": 81_418,
     "record_text_sha256": UPSTREAM_TEXT_SHA256,
 }
 MIN_CHARS = 200
@@ -251,9 +251,10 @@ def normalize_record(
         "license_url": LICENSE_URL,
         "modified": True,
         "modification_notice": (
-            "Article text was extracted from the source HTML. This DynaWord "
-            "release normalizes metadata and may exclude short, non-Polish or "
-            "duplicate records; the retained text is otherwise unchanged."
+            "Article text was extracted from the source HTML and de-identified "
+            "upstream. This DynaWord release normalizes metadata and may exclude "
+            "short, non-Polish or duplicate records; retained text is otherwise "
+            "unchanged after upstream de-identification."
         ),
     }
     return row, attribution
@@ -281,7 +282,7 @@ def existing_text_hashes(repo_root: Path, output_path: Path) -> set[bytes]:
 
 
 def _fsync_path(path: Path) -> None:
-    with path.open("rb") as handle:
+    with path.open("rb+") as handle:
         os.fsync(handle.fileno())
 
 
@@ -394,6 +395,7 @@ def build(
     seen = set(prior_hashes)
 
     candidates: list[dict] = []
+    metadata = manifest["dataset_metadata"]
     stats = {
         "read": len(records),
         "kept": 0,
@@ -420,6 +422,14 @@ def build(
         "upstream_record_text_sha256": UPSTREAM_TEXT_SHA256,
         "upstream_manifest_sha256": hashlib.sha256(manifest_path.read_bytes()).hexdigest(),
         "upstream_artifact_count": len(manifest["artifacts"]),
+        "upstream_deidentification_model": metadata["deidentification_model"],
+        "upstream_deidentification_model_revision": metadata[
+            "deidentification_model_revision"
+        ],
+        "upstream_deidentification_policy": metadata["deidentification_policy"],
+        "upstream_source_quality_excluded_records": metadata[
+            "source_quality_excluded_records"
+        ],
     }
     for record in records:
         text = str(record.get("text") or "").strip()

src/sources.py

@@ -331,12 +331,13 @@ SOURCES = {
         "license": "CC-BY-SA-4.0",
         "license_spdx": "CC-BY-SA-4.0",
         "traceable": "The shared samorzad.gov.pl platform footer publishes textual "
-                     "content under CC BY-SA 4.0. Every retained v0.1 source page "
+                     "content under CC BY-SA 4.0. The v0.2 source release is a "
+                     "de-identified subset of the audited v0.1 pages, all of which "
                      "referenced the shared footer script; audiovisual media and "
                      "attachment bytes are not redistributed.",
         "upstream": "https://huggingface.co/datasets/dawidmajewski/samorzad-gov-pl-articles",
-        "provenance": "Converted reproducibly from the published v0.1 source dataset "
-                      "by src/clean_samorzad_gov_pl.py. The source was collected "
+        "provenance": "Converted reproducibly from the published, de-identified v0.2 "
+                      "source dataset by src/clean_samorzad_gov_pl.py. The source was collected "
                       "directly from public samorzad.gov.pl pages across 246 local "
                       "public institutions and one central platform tenant. "
                       "Institution display names are frozen "

src/test_samorzad_gov_pl_contract.py

@@ -73,6 +73,7 @@ def test_normalize_record_maps_publisher_to_author_and_preserves_provenance():
     assert attribution["title"] == source["title"]
     assert attribution["license_url"] == "https://creativecommons.org/licenses/by-sa/4.0/"
     assert attribution["modified"] is True
+    assert "de-identified upstream" in attribution["modification_notice"]
 
 
 def test_verify_upstream_release_rejects_modified_shard(tmp_path):
@@ -183,14 +184,24 @@ def test_release_artifact_matches_stats_and_has_institution_authors():
 
     assert table.schema == DYNAWORD_SCHEMA
     assert table.num_rows == stats["kept"] == len(attributions)
+    assert stats["read"] == 81_418
+    assert stats["kept"] == 72_472
     assert stats["authors_with_value"] == stats["kept"]
     assert stats["publisher_count"] == 247
     assert stats["local_public_tenant_count"] == 246
     assert stats["central_tenant_count"] == 1
-    assert stats["redirected_urls"] == 693
+    assert stats["redirected_urls"] == 688
     assert stats["upstream_artifact_count"] == 10
+    assert stats["upstream_deidentification_model"] == (
+        "ArkadiuszPawlak/fastpdn-ner-polish-pii"
+    )
+    assert stats["upstream_deidentification_model_revision"] == (
+        "636c57fe77b0afad8f065e7c76243dc051437035"
+    )
+    assert stats["upstream_deidentification_policy"] == "samorzad-pii-v3"
+    assert stats["upstream_source_quality_excluded_records"] == 4
     assert stats["upstream_manifest_sha256"] == (
-        "97b5a266b2fd4f043648b7d3c757721cd826c7ca1048301761084b0abdee5c48"
+        "9d349bb271033fef62a051e311115460e030740f9593bb7f3158cb1437f87639"
     )
     assert set(table["source"].to_pylist()) == {"samorzad_gov_pl"}
     assert set(table["license"].to_pylist()) == {"CC-BY-SA-4.0"}