START WITH THE PRODUCT STEPS
Read START-HERE.txt first. It states the actual source acquisition, access/privacy limits, manual field mapping and exact next prompt. The original file templates and synthetic examples below are supporting review aids. Some paths need separately supplied documents or source text; a blank worksheet is not a native export or completed integration.

Q02 - Check the final PDF for residual approved-redaction targets

Status: Proposed / not tested

WHAT THIS KIT CONTAINS
Exact input scaffolds and synthetic examples for an offline pilot. Header, JSON-shape and fixture-consistency checks validate these files only; they do not test an assistant, product integration, legal decision, PDF inspection or live workflow.

PREREQUISITES
1. Approved final derivative, original preserved, lawyer-approved target IDs/tokens or coordinates, Acrobat Pro hidden-information inspection and approved local text extraction/rendering. Do not upload real target values to a general-purpose public assistant. Never claim text extraction alone proves safe redaction; image-only data and embedded content require separate checks.

SETUP
1. Use the separate synthetic example inputs first. Fill a copy of the blank template or input checklist for your own preparation; do not mix fixture information with real records.
2. The fixture fingerprints are hashes of synthetic labels, not attached PDFs. This kit validates metadata logic only; PDF extraction, rendering, hidden-content inspection and sanitization have not been run.
3. In a negative-control copy, remove the finding and set visual_checked=unknown: output must remain incomplete.
4. Preserve exact CSV headers. CSV templates contain only a header row. Import ID, hash, page-label and timestamp columns as text; inspect leading zeros and do not enable formulas from untrusted input.
5. Give only the selected synthetic input sections/files and the prompt below to an existing firm-approved private workspace, or perform the clerical comparison manually. Keep expected-result notes separate from the input. For planning checklists, leave unknown requirements explicit instead of inventing them.
6. Compare the draft output with the expected fixture result. A reviewer checks original evidence and all exceptions before any separately approved action. Retain the version, scope and source references with the review.

INPUT RULES
sha256 is 64 hexadecimal characters for the exact final bytes. page and page_count are positive integers. Check fields contain literal true, false or unknown. layer: visible_text, hidden_text, metadata, attachment, annotation or image_review. Do not place real sensitive token values in these files.

FILES
- final_files.template.csv: Freeze the exact final derivative identifier, fingerprint and page count
- final_files.example.csv: Freeze the exact final derivative identifier, fingerprint and page count
- targets.template.csv: Lawyer-approved token aliases and regions; no real target values
- targets.example.csv: Lawyer-approved token aliases and regions; no real target values
- extraction_findings.template.csv: Evidence from separate approved PDF inspection tools
- extraction_findings.example.csv: Evidence from separate approved PDF inspection tools
- inspections.template.csv: Inspection coverage tied to file hash; the older row is intentionally stale
- inspections.example.csv: Inspection coverage tied to file hash; the older row is intentionally stale

COPYABLE OFFLINE PROMPT
Using only these synthetic metadata tables, check that every inspection refers to the supplied final sha256. Join residual extraction_findings to lawyer-approved targets by file_id and token ID. Report residual target matches without reproducing sensitive values. Flag missing or stale inspections and unchecked layers. Return file_id,sha256,target_id,page,layer,exception,source_ref and a coverage checklist. Do not decide what should be redacted, alter a PDF, remove metadata, disclose targets, or certify confidentiality. Unknown and unchecked stay unresolved.

INPUTS
final_files.csv = file_id,sha256,page_count. targets.csv = target_id,file_id,page,approved_token_id,approved_region_ref. extraction_findings.csv = finding_id,file_id,sha256,page,layer,matched_token_id,source_ref where layer is visible_text|hidden_text|metadata|attachment|annotation|image_review. inspections.csv = file_id,sha256,text_checked,hidden_checked,attachments_checked,visual_checked,reviewer_id (checks true|false|unknown).

OUTPUT
A residual-target exception report tied to the exact final file hash and page, plus a manual review checklist. Missing inspections remain unresolved.

Use only these synthetic or firm-approved inputs and preserve source row/page IDs. Keep missing or conflicting evidence unresolved. Return a draft for the named reviewer. Do not send messages, connect accounts, change records or permissions, move money, or make legal decisions. Keep source files unchanged. List missing prerequisites before continuing.

EXPECTED SYNTHETIC FIXTURE RESULT
D1/T1 has a residual hidden_text target match citing X1. The current-hash inspection leaves hidden_checked=false. The different-hash inspection is stale and cannot cover the final version. This is an unresolved release review, never a safe-redaction finding.

LIMITS
- Lawyer supplies redaction scope and checks release copy. Qualified staff independently inspect rendered pages, extractability, attachments and hidden data in the approved environment. Preserve originals and signatures; this recipe never performs destructive sanitization or release.
- Synthetic data only in the delivered files. No credentials, real client records, live-action scripts, API connection or Make blueprint is included.
- Real use requires firm-approved access, confidentiality, processing terms, retention, product access and reviewer authority. Pseudonymous IDs alone do not anonymize real records.
