CDIP cheat sheet

Scan resolution for a 1:1 bitonal image equals dpi; for grayscale or color, sample at 300 dpi minimum and 400 dpi for OCR-sensitive or small text.

Bit depth decision rule: 1-bit bitonal for clean text with high contrast; 8-bit grayscale for faded or low-contrast text; 24-bit color when color carries meaning (maps, photos, stamps).

JPEG is lossy and suited to continuous-tone color; TIFF (uncompressed or LZW) is the preservation master; PDF/A is the archival delivery format.

Compression ratio and quality: JPEG quality below ~75 introduces visible artifacts around text; never use JPEG for bitonal text masters.

Deskew, despeckle, and border removal are capture-stage image enhancements; autocrop and blank-page detection reduce downstream index errors.

OCR accuracy depends on resolution, contrast, and skew; a 1-degree skew can degrade OCR accuracy measurably, so deskew before recognition.

Index fields captured at scan time (barcode, forms recognition, zone OCR) must be validated against a confidence threshold before commit.

Image QC sampling: inspect a statistically valid sample per batch for skew, missing pages, double feeds, and legibility before release.

Retention schedule drives disposition: active, semi-active, and final disposition (destroy or permanent/archival) each map to distinct storage tiers.

WORM (write once, read many) media and immutable object-lock policies satisfy legal-hold and audit requirements that ordinary file shares do not.

Indexing strategy: metadata fields that users search on must be indexed; full-text index is separate from metadata index and has its own refresh cost.

Retrieval latency tiers: online (milliseconds), nearline (seconds to minutes), offline/archive (hours); match tier to access frequency, not document age alone.

Backup versus archive: backups are for recovery of operational state; archives are the authoritative long-term record and must be independently retrievable.

Migration and format obsolescence: refresh media on a defined cycle and verify checksums after every migration to prove bit-level integrity.

Legal hold suspends normal retention deletion; the hold must be tracked per matter and released explicitly before disposition resumes.

Deduplication and hashing: SHA-256 content hash supports duplicate detection and integrity verification across stores.

Workflow types: sequential (linear approval), parallel (concurrent tasks), ad hoc (user-defined routing); choose based on whether steps are mandatory and ordered.

A queue/work-queue model assigns tasks by skill, priority, and workload; escalation rules fire when a task exceeds its SLA timer.

Exception handling must define a path for every rejection: return to prior step, route to supervisor, or terminate with a recorded reason.

Business process modeling notation (BPMN) distinguishes gateways (exclusive, parallel, inclusive) which control branching logic in the process model.

Straight-through processing rate is the percentage of items completed without human touch; measure it to find automation candidates.

SLA measurement starts at task creation and stops at completion; clock-pause rules for awaiting-customer states must be defined explicitly.

Audit trail requirements: every state change records user, timestamp, prior state, and new state; the trail must be tamper-evident.

Version control of process definitions: changes to routing rules require change control and re-testing before production deployment.

Print output resolution: 300 dpi for office laser printing, 600 dpi or higher for forms with small type or barcodes; barcodes require adequate quiet zones.

PDF/A for archival output, PDF/X for print production; PDF/A forbids encryption, external fonts, and transparency that break long-term rendering.

Distribution methods: email (size limits, encryption), secure portal (authenticated download), fax (regulated in some jurisdictions), API/EDI (system-to-system).

Access control on distribution: apply need-to-know at the document level, not just the folder level, and log every download and view.

Redaction must remove underlying text and image layers, not just draw a black box; verify by extracting text from the released file.

Batch printing and mail merge require a control record to reconcile documents produced against documents mailed.

Digital signature and timestamping prove integrity and signer identity; a scanned wet signature image alone provides no cryptographic proof.

Output reconciliation: count of items sent must equal count of items processed minus exceptions, with exceptions logged and resolved.

ISO 15489 governs records management principles; ISO 30301 specifies a management system for records; DoD 5015.02 defines records management application functional requirements.

PDF/A (ISO 19005) is the archival PDF standard; PDF/UA (ISO 14289) addresses accessibility; PDF/A-1 uses PDF 1.4, PDF/A-2 and -3 use later versions.

MoReq2010 and ICA-Req are records management specification frameworks; Dublin Core is a metadata element set with 15 core elements.

Security triad: confidentiality, integrity, availability; apply least privilege and separation of duties to records administration roles.

Encryption at rest (AES-256) and in transit (TLS 1.2+) are baseline; key management and rotation are separate controls from encryption itself.

Retention and privacy: GDPR-style rules require lawful basis, data minimization, and deletion at end of purpose; retention schedules must not conflict with privacy limits.

Audit logging must be immutable and retained for the audit period; access reviews should be periodic and documented.

Classification scheme: assign security labels at capture and enforce them through storage, retrieval, and distribution so labels travel with the record.

Forms processing uses template registration and dropout color; the dropout color must not match the ink color of the data to be captured.

Barcode reading requires sufficient module width and quiet zone; Code 39 and Code 128 differ in density and character set.

Scanner types: sheet-fed (loose pages, fast), flatbed (bound or fragile originals), overhead/book (bound volumes), and production rotary (high volume).

Color management: calibrate scanners with an IT8 target so captured color is consistent across devices and time.

Image file size formula: bytes ≈ (width px × height px × bit depth) / 8, before compression; use it to size storage and estimate throughput.

RAID levels: RAID 1 mirrors, RAID 5 stripes with single parity, RAID 6 stripes with double parity; RAID is availability, not a backup.

Storage media life: magnetic tape and optical media have defined shelf lives; verify readability on a scheduled cycle rather than trusting the vendor claim.

Object storage uses a unique identifier plus metadata; it scales better than hierarchical file systems for very large document counts.

Search recall versus precision: broad full-text search raises recall and lowers precision; fielded metadata search does the reverse.

Checksum verification on read detects silent corruption; schedule periodic integrity scans of the archive tier.

Task routing rule example: if amount > threshold, route to second approver; if confidence < 90%, route to manual index queue.

Workflow state machine needs explicit terminal states; documents must not remain in an in-process state indefinitely.

Cycle time equals elapsed time from receipt to completion; touch time is the subset spent actively worked, and the gap is waiting waste.

Role-based assignment uses a group queue with claim/lock semantics to prevent two users processing the same item.

Notification and escalation timers should be configurable per process step, not hard-coded, to survive policy changes.

Print production: imposition arranges pages on a press sheet; bleed is the area beyond trim size that prevents white edges after cutting.

Color modes: RGB for screen, CMYK for process printing, spot color for exact brand or security ink; converting CMYK to RGB is lossy.

Secure distribution channels must authenticate the recipient and log delivery; plain email attachment is the weakest common channel.

Output formats for delivery: PDF (fixed layout), TIFF (image), XML/CSV (data), and native formats when downstream editing is required.

Print release and pull printing hold jobs until the user authenticates at the device, reducing confidential output left on trays.

Chain of custody documents each transfer of a record; a gap weakens admissibility and audit defensibility.

Access control models: RBAC assigns permissions by role, ABAC evaluates attributes; records systems commonly layer both.

Legal admissibility factors include authenticity, integrity, completeness, and reliable audit trail; a documented system description supports them.

Incident response for records: detect, contain, assess scope, notify per regulation, remediate, and record lessons; notification deadlines are jurisdiction-specific.

Vital records protection plan identifies records needed to resume operations and stores copies off-site or in a separate system.