Record source scope mismatch against original Oracle XML

.project-docs/30-worklog/tasks/20261008-production-review-9e7b.md

@@ -105,3 +105,14 @@ Read: memory-index, project-positioning, current-state latest September sections
 - Version comparison: the desktop primary checkout remains at `ad2c592`; localhost runs the isolated repair checkout with the new source-review/date-switching UI. Nevertheless `process_daily.py`, `validate_daily.py` and `ARR_XML_RAW_FIELDS.md` are byte-identical in both checkouts. `git diff ad2c592` for those files is empty. No new room-required rule or fixed-price/deduplication change was introduced in this repair.
 - Verification: authenticated local review/config inspection, exact file comparisons, saved-capture rule replay, Git history and an independent read-only audit. No test rerun, service restart, field confirmation, XML rewrite or Oracle request was needed; all observations and business values are preserved.
 - Follow-up/promotion candidate: distinguish mandatory missing values from optional source uncertainty in product copy; assess evidence for treating particular omitted optional API structures as empty before changing that behavior. Preserve the separate unresolved cancellation/report-scope question. Do not silently convert unknown/failed observations to empty or claim all26 fields are required by the original XML rules. Integrate the repair branch into the primary project through the established integration workflow when requested.
+
+## Same-task Follow-up: Original Oracle XML Comparison
+
+- User reports confirmation300755250 cannot be found in the downloaded table, clarifies it is Oracle's original report, and supplies `/Users/chillishark/Downloads/res_detail_10.7.XML`. Same-task feature ownership and existing planning context remain verified. Scope is read-only comparison, not permission to change source eligibility or finalize real data.
+- Original file is191356 bytes, SHA256 `bf1d1905e068f7c496a3caf63e3682d2aeee2b62c30ce6c87e14e2411a9d6d84`; confirmed unchanged after inspection. Full raw bytes contain neither300755250 nor300007418. Its hotel is57106. It has40 distinct reservations in two arrival-date groups:2026-10-07 has13,2026-10-08 has27. Frozen `read_xml` correctly refuses the full file as multiple business dates; in-memory structural parsing was used only for diagnosis, without relaxing that rule or producing artifacts for ingestion.
+- Target300755250 is genuinely returned by saved OHIP getReservation: exact typed Confirmation ID, internal reservation5984613, InHouse, room4406, arrival10/7/departure10/10, rateGRPA1. It is in the existing whitelist and is the only4406 row in the58-row API source, so whitelist/room-date deduplication does not explain its absence from Oracle's XML. It only needs a block-source decision; its ABFN package was returned.
+- XML10/7's13 confirmation IDs exactly equal the13 CheckedOut IDs in the saved58-row API source. API-only45 rows are InHouse38, Cancelled6 and NoShow1. No XML10/7 confirmation is absent from the API. XML10/8's27 confirmations are outside the single-day API source. This proves different observed source populations; it does not prove Oracle export parameters, export instant, or a universal CheckedOut-only report policy.
+- Of the26 pending field decisions,15 concern reservations absent from XML10/7 (room1,block12,packages2). The remaining11 concern XML10/7's6 whitelisted reservations (block6,packages5); every one of those11 values is explicitly blank in the original XML and allowed blank by the existing business rules. Staff should not be asked to blindly fill all26 before report scope is reconciled. No actual decisions were imported or changed during this audit.
+- Additional common13-row comparison found normalized display/text differences in company12,full-name13,comments4. Their semantics have not all been resolved; do not claim exact source/report equivalence merely from matching confirmation IDs or infer that all differences are cosmetic. Preserve the evidence for a separate field-mapping review.
+- Verification: original XML parsing/hashes, frozen rule normalization, exact confirmation-set/status join against immutable captured details, and independent read-only review. No Oracle requests, real report generation, database writes, service restart, source rewriting or rule changes occurred. Private comparison summary: `/Users/chillishark/Library/Application Support/ARR2.0/production-validation-20261007/oracle-xml-scope-comparison-20261007.json`; original XML/PII stay outside Git.
+- Follow-up/promotion candidate: align the API query's date/status/report-selection scope with the intended Oracle ARR export settings before accepting the real10/7 daily/monthly result. Inspect original export settings or authoritative report criteria; do not silently adopt CheckedOut-only from one observed file or filter by this file's ID list. Optional empty-field evidence can support explicit auditable resolution for matching records, but the user's request here is comparison only. The previous claim that only business-field completion remained for real10/7 acceptance is incomplete: observed report scope and some text mappings also need alignment.