SCRIP-CORPUS-087: recover Petersburg extraction-strategy preflight

corpus/candidates/bely-petersburg-1916-ru.md

@@ -28,6 +28,16 @@ The contract freezes only future composition semantics: ascending PDF-page order
 
 The retained Commons file description identifies the facsimile as the first 1916 book publication and marks the work public domain. That source/legal evidence is retained in the structured trace. The candidate is not collapsed with a generic source-less transcription or the revised 1922 Berlin edition.
 
+## Extraction-strategy preflight
+
+**SCRIP-CORPUS-087** adds a source-free preflight before choosing direct PDF text extraction versus raster OCR. On 24 September 2026 the official Commons surface still exposed the frozen candidate as a 632-page PDF facsimile of the first 1916 book edition, with the current file history rooted at the 20 March 2016 upload, explicit public-domain notice, PDF format 1.6, and a current-file no-overwrite notice.
+
+Two provider PDF screenshots were inspected directly: the title-page surface at carrier page 1 and a prose surface at carrier page 21, where printed folio 19 is visible. Those previews visibly look like facsimile page imagery, but that fact **does not prove that the PDF lacks a hidden text layer**. Conversely, the provider metadata does not prove that a usable embedded text layer exists.
+
+Therefore the embedded text layer remains **unverified**. Scriptorium does not yet choose either direct PDF text extraction or raster OCR. The next bounded evidence is an exact-carrier, version-recorded text-layer probe that retains only aggregate counts, an output digest and page-segment coverage; any extracted text remains transient. Even a non-empty text layer will still require separate correctness and literary-page-selection evidence before body promotion.
+
+Canonical source-free evidence: [`source-edition-traces/bely-petersburg-1916-ru.extraction-preflight.json`](source-edition-traces/bely-petersburg-1916-ru.extraction-preflight.json), physical SHA-256 **`093f251cb8d03846a5158a1ac29d4b66e1796a7cc2bc65a77ade3d9e3cebcff1`**.
+
 ## What is still open
 
 Defining the promotion contract does **not** freeze the literary text. Scriptorium still needs an independently evidenced bound contract version with exact literary-page selection and reproducible rasterizer/OCR identities/settings, then a frozen extracted body with character count and raw/normalized digests and proof that that frozen body itself clears the **300,000-character** corpus threshold.

corpus/candidates/source-edition-traces/bely-petersburg-1916-ru.extraction-preflight.json

@@ -0,0 +1,58 @@
+{
+  "candidate_id": "bely-petersburg-1916-ru",
+  "carrier": {
+    "byte_count": 3621459,
+    "current_file_history_upload_timestamp": "2016-03-20T18:31:00Z",
+    "encrypted": false,
+    "page_count": 632,
+    "pdf_format_version": "1.6",
+    "provider": "Wikimedia Commons",
+    "sha1": "682476934dd6ed49c6bbcdb0720127c1812ff477",
+    "sha256": "b08820ad1339894c6d20fbaa2367c11385492bf376d199f8de23c859ef6ddef5"
+  },
+  "decision": {
+    "direct_pdf_text_extraction_admissible": false,
+    "embedded_text_layer_status": "unverified",
+    "next_required_evidence": "Run a source-free exact-carrier text-layer probe with a version-recorded extractor; retain only aggregate counts, output digest and page-segment coverage. If a text layer exists, correctness still requires separate evidence before body promotion.",
+    "provider_metadata_proves_embedded_text_layer": false,
+    "raster_ocr_required": false,
+    "visual_facsimile_preview_proves_absence_of_text_layer": false
+  },
+  "evidence_version": "scriptorium-petersburg-1916-extraction-preflight-v1",
+  "existing_promotion_contract": {
+    "body_outputs_frozen": false,
+    "canonical_sha256": "d338d8dc4d800b0fe28a848e383265ba9a0e720e75962f7f9e5733ad727a20d2",
+    "contract_version": "scriptorium-petersburg-1916-ocr-body-contract-v1",
+    "ocr_toolchain_bound": false,
+    "page_selection_bound": false,
+    "rasterization_toolchain_bound": false
+  },
+  "gates": {
+    "admitted_for_calibration": false,
+    "diagnostic_ready": false,
+    "extraction_profile_frozen": false,
+    "fantlab_source_edition_match": "unknown",
+    "literary_body_frozen": false,
+    "literary_page_selection_bound": false,
+    "m2_parity_admissible": false,
+    "minimum_300k_proved_from_frozen_body": false
+  },
+  "observed_on": "2026-09-24",
+  "provider_surface": {
+    "current_file_cannot_be_overwritten": true,
+    "first_book_edition_facsimile": true,
+    "metadata_short_title": "Андрей Белый. Петербург. 1916. EBook 2008",
+    "public_domain_notice": true,
+    "visual_samples": [
+      {"carrier_page": 1, "observation": "facsimile title-page image visible in the provider PDF preview"},
+      {"carrier_page": 21, "observation": "facsimile prose page visible in the provider PDF preview; printed folio 19 is visible"}
+    ]
+  },
+  "publication": {
+    "extracted_text_committed": false,
+    "literary_source_text_committed": false,
+    "page_images_committed": false,
+    "pdf_bytes_committed": false,
+    "source_free": true
+  }
+}

project/CHANGELOG.d/2026-09-24-scrip-corpus-087.md

@@ -0,0 +1,9 @@
+## 2026-09-24 — Petersburg 1916 extraction-strategy preflight
+
+- Recovery assigns this Petersburg work to **SCRIP-CORPUS-087**; White Guard remains SCRIP-CORPUS-086.
+- Issue #263 and Draft PR #264 recover the staged Petersburg preflight onto fresh current-master branch `scrip-corpus-087-petersburg-preflight-current`.
+- Exact stale source handoff `a24bbaeff4054762640694b8025500815853d4a5` was used only to transfer five Petersburg-specific files; stale project state was not merged.
+- Source-free evidence remains `scriptorium-petersburg-1916-extraction-preflight-v1`, physical SHA-256 `093f251cb8d03846a5158a1ac29d4b66e1796a7cc2bc65a77ade3d9e3cebcff1`.
+- Embedded-text-layer status remains `unverified`; neither direct PDF extraction nor raster OCR is selected.
+- Literary-page selection, body identity, >=300k admission, FantLab analyzer-input identity, diagnostics, M2 admission and parity remain closed; M2 remains 0/5.
+- PR #264 remains Draft for a later independent exact-head review after fresh hosted checks.

project/STATE_AND_QUEUE.md

@@ -1,23 +1,23 @@
 # STATE_AND_QUEUE — Scriptorium
 
-STATE_REVISION: 287
+STATE_REVISION: 288
 PHASE: M1 — Deterministic FantLab surface
-LAST_COMMITTED_RUN_AT: 2026-09-29T13:08:28Z
-LAST_RESULT: SCRIP-CORPUS-086 / Issue #261 / PR #262 completed. Independent exact-head review `5353031421` re-read all 7 changed files at `64d405809c400a7f7294b1308b5d394fa61ee2aa`, found no merge blockers, and confirmed 26/26 pull-request-triggered workflows completed successfully with no unresolved review threads. PR #262 was marked Ready and squash-merged with expected-head protection as `fa9ead6e5a2489c9eea604ddbdbe73b0093c9eb1`; Issue #261 closed completed. No live 20-row provider manifest, literary-body identity, FantLab analyzer-input identity, benchmark or M2 movement is claimed.
+LAST_COMMITTED_RUN_AT: 2026-09-29T22:39:45Z
+LAST_RESULT: SCRIP-CORPUS-087 / Issue #263 / Draft PR #264 recovered the Petersburg 1916 extraction-strategy preflight from exact stale handoff `a24bbaeff4054762640694b8025500815853d4a5` onto current master without merging stale state. The five Petersburg-specific files were transferred onto fresh branch `scrip-corpus-087-petersburg-preflight-current`; source-free preflight SHA-256 remains `093f251cb8d03846a5158a1ac29d4b66e1796a7cc2bc65a77ade3d9e3cebcff1`. Embedded text-layer status remains unverified; literary body, >=300k admission, FantLab-input identity, diagnostics, M2 and parity remain closed. PR #264 is intentionally Draft for later independent exact-head judgement.
 LAST_VERIFIED_PROGRESS: M1 deterministic metric families remain executable inferred candidates. Public corpus/showcase evidence includes exact frozen/admitted works and source-free derived outputs, while no retained work has an independently established FantLab analyzer-input identity. Perelman now has independently reviewed exact PDF/DjVu identities, pagination, current Wikisource/FantLab surface boundaries, zero-PDF-PageLabels evidence, front-matter anchors, transition/tail anchors, restored-page interior observations, and canonical broad mid-body exact-PDF anchors that expose an unlabelled carrier page while keeping global mapping closed. Darwin/Rachinsky retains canonical reviewed dependency/replay/provider evidence through SCRIP-CORPUS-084. M2 stays 0/5.
 
 ## Current unit
 
 ```text
-UNIT_ID:        SCRIP-CORPUS-086
-ISSUE:          #261 (closed completed)
-STATUS:         COMPLETE
-PR:             #262 (squash-merged as fa9ead6e5a2489c9eea604ddbdbe73b0093c9eb1)
-NEXT_ACTION:    Resume normal-flow corpus/provenance selection.
-                A separate bounded White Guard unit may perform authoritative 20-page identity-only capture/replay
-                of page ID, revision ID, UTC timestamp and MediaWiki SHA-1.
-                Keep literary-body, composite/single-edition identity, FantLab-input, benchmark, M2 and parity gates closed
-                until separately evidenced.
+UNIT_ID:        SCRIP-CORPUS-087
+ISSUE:          #263 (open)
+STATUS:         REVIEW_PENDING
+PR:             #264 (Draft)
+AUTHORED_HEAD:  e6e93a4639f5986a85f87b5341f684ba6808dda7
+NEXT_ACTION:    Independently review PR #264 on its final exact head after this bookkeeping reconciliation.
+                Require fresh exact-head checks, no blocking reviews and no unresolved review threads before Ready/merge.
+                Keep embedded text-layer status unverified and do not select direct PDF extraction or raster OCR.
+                Literary-page selection, body identity, >=300k admission, FantLab-input identity, diagnostics, M2 and parity remain closed.
 ```
 
 ## Current milestone gate

project/run-receipts/SCRIP-CORPUS-087.md

@@ -0,0 +1,29 @@
+# Run receipt — SCRIP-CORPUS-087
+
+Status: REVIEW_PENDING
+
+## Recovery state
+
+- Canonical recovery base: `master@27f54858d20000244bb8e80342038d80ea510b45`.
+- Issue: #263.
+- Draft PR: #264.
+- Fresh recovery branch: `scrip-corpus-087-petersburg-preflight-current`.
+- Exact stale Petersburg handoff used only as a five-file source: `a24bbaeff4054762640694b8025500815853d4a5`.
+- Recovery semantic head before bookkeeping: `e6e93a4639f5986a85f87b5341f684ba6808dda7`.
+- The stale branch itself was not merged and its old `STATE_AND_QUEUE.md` was not copied.
+
+## Recovered evidence
+
+- Source-free preflight physical SHA-256: `093f251cb8d03846a5158a1ac29d4b66e1796a7cc2bc65a77ade3d9e3cebcff1`.
+- Exact frozen carrier: 632 pages / 3,621,459 bytes / SHA-1 `682476934dd6ed49c6bbcdb0720127c1812ff477` / SHA-256 `b08820ad1339894c6d20fbaa2367c11385492bf376d199f8de23c859ef6ddef5`.
+- Existing OCR/body promotion contract SHA-256 remains `d338d8dc4d800b0fe28a848e383265ba9a0e720e75962f7f9e5733ad727a20d2`.
+- Embedded text-layer status remains `unverified`; neither direct PDF text extraction nor raster OCR is selected.
+- No PDF bytes, page images, extracted text or literary source text are committed.
+
+## Gate boundary
+
+Literary-page selection, extraction profile, body identity, >=300k admission, FantLab analyzer-input identity, diagnostics, M2 admission and parity remain closed. M2 remains 0/5.
+
+## Handoff
+
+PR #264 remains intentionally Draft. A later independent wake must re-read the final exact head, inspect fresh hosted checks/reviews/threads and only then decide Ready/merge.

tests/test_petersburg_extraction_preflight.py

@@ -0,0 +1,77 @@
+from __future__ import annotations
+
+import hashlib
+import json
+from pathlib import Path
+import unittest
+
+
+ROOT = Path(__file__).resolve().parents[1]
+EVIDENCE = ROOT / "corpus/candidates/source-edition-traces/bely-petersburg-1916-ru.extraction-preflight.json"
+CANDIDATE_PAGE = ROOT / "corpus/candidates/bely-petersburg-1916-ru.md"
+EXPECTED_PHYSICAL_SHA256 = "093f251cb8d03846a5158a1ac29d4b66e1796a7cc2bc65a77ade3d9e3cebcff1"
+EXPECTED_SCAN_SHA256 = "b08820ad1339894c6d20fbaa2367c11385492bf376d199f8de23c859ef6ddef5"
+
+
+class PetersburgExtractionPreflightTests(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls) -> None:
+        cls.raw = EVIDENCE.read_bytes()
+        cls.data = json.loads(cls.raw)
+
+    def test_source_free_artifact_is_frozen(self) -> None:
+        self.assertEqual(hashlib.sha256(self.raw).hexdigest(), EXPECTED_PHYSICAL_SHA256)
+        self.assertTrue(self.data["publication"]["source_free"])
+        self.assertFalse(self.data["publication"]["pdf_bytes_committed"])
+        self.assertFalse(self.data["publication"]["page_images_committed"])
+        self.assertFalse(self.data["publication"]["extracted_text_committed"])
+        self.assertFalse(self.data["publication"]["literary_source_text_committed"])
+
+    def test_exact_carrier_identity_is_preserved(self) -> None:
+        carrier = self.data["carrier"]
+        self.assertEqual(carrier["page_count"], 632)
+        self.assertEqual(carrier["byte_count"], 3621459)
+        self.assertEqual(carrier["sha1"], "682476934dd6ed49c6bbcdb0720127c1812ff477")
+        self.assertEqual(carrier["sha256"], EXPECTED_SCAN_SHA256)
+
+    def test_preflight_does_not_guess_text_layer_or_ocr_strategy(self) -> None:
+        decision = self.data["decision"]
+        self.assertEqual(decision["embedded_text_layer_status"], "unverified")
+        self.assertFalse(decision["provider_metadata_proves_embedded_text_layer"])
+        self.assertFalse(decision["visual_facsimile_preview_proves_absence_of_text_layer"])
+        self.assertFalse(decision["direct_pdf_text_extraction_admissible"])
+        self.assertFalse(decision["raster_ocr_required"])
+        self.assertIn("text-layer probe", decision["next_required_evidence"])
+
+    def test_existing_promotion_contract_remains_unbound(self) -> None:
+        contract = self.data["existing_promotion_contract"]
+        self.assertEqual(
+            contract["canonical_sha256"],
+            "d338d8dc4d800b0fe28a848e383265ba9a0e720e75962f7f9e5733ad727a20d2",
+        )
+        self.assertFalse(contract["page_selection_bound"])
+        self.assertFalse(contract["rasterization_toolchain_bound"])
+        self.assertFalse(contract["ocr_toolchain_bound"])
+        self.assertFalse(contract["body_outputs_frozen"])
+
+    def test_downstream_gates_remain_closed(self) -> None:
+        gates = self.data["gates"]
+        self.assertFalse(gates["literary_page_selection_bound"])
+        self.assertFalse(gates["extraction_profile_frozen"])
+        self.assertFalse(gates["literary_body_frozen"])
+        self.assertFalse(gates["minimum_300k_proved_from_frozen_body"])
+        self.assertEqual(gates["fantlab_source_edition_match"], "unknown")
+        self.assertFalse(gates["diagnostic_ready"])
+        self.assertFalse(gates["m2_parity_admissible"])
+        self.assertFalse(gates["admitted_for_calibration"])
+
+    def test_public_candidate_exposes_preflight_without_promotion(self) -> None:
+        page = CANDIDATE_PAGE.read_text(encoding="utf-8")
+        self.assertIn("Extraction-strategy preflight", page)
+        self.assertIn(EXPECTED_PHYSICAL_SHA256, page)
+        self.assertIn("embedded text layer remains **unverified**", page)
+        self.assertIn("does not prove that the PDF lacks a hidden text layer", page)
+
+
+if __name__ == "__main__":
+    unittest.main()