{
  "slug": "extraction-schema-versioning-for-document-pipelines",
  "title": "Versioning a Document Extraction Schema: Migrations, Manifests and Rollback",
  "description": "How to version a document extraction schema: which edits break stored records, absent versus null fields, lazy upcast versus backfill, reprocessing from stored parse output, a version manifest, dual-read cutover, correction migration and rollback.",
  "kind": "sub",
  "order": 5,
  "target_query": "how to version a document extraction schema",
  "secondary_queries": [],
  "tags": [
    "document-extraction",
    "schema-versioning",
    "migration",
    "rollback",
    "production"
  ],
  "published": "2026-10-27",
  "updated": "2026-10-27",
  "words": 1627,
  "estimated_tokens": 2164,
  "premium": false,
  "rights": {
    "access": "free",
    "note": "Editorial guides are always free and never part of the licensed corpus.",
    "license": "https://changegamer.ai/license.xml",
    "pricing": "https://changegamer.ai/api/pricing.json",
    "payment": "https://changegamer.ai/api/payment.json"
  },
  "license": "https://changegamer.ai/license.xml",
  "citation": "ChangeGamer (2026-10-27). Versioning a Document Extraction Schema: Migrations, Manifests and Rollback. ChangeGamer. https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines (updated 2026-10-27).",
  "bibtex": "@misc{changegamer_extraction_schema_versioning_for_document_pipelines, title = {Versioning a Document Extraction Schema: Migrations, Manifests and Rollback}, publisher = {ChangeGamer}, year = {2026}, url = {https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines}, note = {Updated 2026-10-27}}",
  "canonical": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines",
  "markdown": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines.md",
  "takeaways": [
    "A document extraction schema should be versioned as one manifest of schema, parser, model and prompt, because a parser upgrade or model pin change alters stored values even when the schema file is untouched (author design).",
    "Editing a field description changes extraction behaviour without changing the schema shape, so a description or prompt edit needs a new version and a regression run just like a renamed field (author inference).",
    "Old extraction records should distinguish a field that did not exist yet (absent) from a field the extractor looked for and did not find (null), otherwise a later required field cannot be backfilled safely (author design).",
    "Reprocessing stored documents from the saved parse artifact, scoped by version and keyed idempotently, is safer than re-parsing the originals, and the superseded record should be kept until the new version passes review (author design).",
    "The structured-outputs reference states that on Anthropic a schema change invalidates the grammar cache and adds first-request latency, so a schema cutover should include a warm-up request before traffic moves."
  ],
  "outline": [
    {
      "depth": 2,
      "text": "Which schema changes are breaking beyond a rename?",
      "anchor": "which-schema-changes-are-breaking-beyond-a-rename",
      "url": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines#which-schema-changes-are-breaking-beyond-a-rename"
    },
    {
      "depth": 2,
      "text": "How should old records represent absent versus null fields?",
      "anchor": "how-should-old-records-represent-absent-versus-null-fields",
      "url": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines#how-should-old-records-represent-absent-versus-null-fields"
    },
    {
      "depth": 2,
      "text": "Lazy upcast or eager backfill: how do you migrate stored records?",
      "anchor": "lazy-upcast-or-eager-backfill-how-do-you-migrate-stored-records",
      "url": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines#lazy-upcast-or-eager-backfill-how-do-you-migrate-stored-records"
    },
    {
      "depth": 2,
      "text": "How do you reprocess documents from the stored parse artifact?",
      "anchor": "how-do-you-reprocess-documents-from-the-stored-parse-artifact",
      "url": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines#how-do-you-reprocess-documents-from-the-stored-parse-artifact"
    },
    {
      "depth": 2,
      "text": "What belongs in a version manifest?",
      "anchor": "what-belongs-in-a-version-manifest",
      "url": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines#what-belongs-in-a-version-manifest"
    },
    {
      "depth": 2,
      "text": "How do you cut over without losing records?",
      "anchor": "how-do-you-cut-over-without-losing-records",
      "url": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines#how-do-you-cut-over-without-losing-records"
    },
    {
      "depth": 2,
      "text": "What happens to reviewer corrections when the schema changes?",
      "anchor": "what-happens-to-reviewer-corrections-when-the-schema-changes",
      "url": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines#what-happens-to-reviewer-corrections-when-the-schema-changes"
    },
    {
      "depth": 2,
      "text": "How do you roll back and prove nothing regressed?",
      "anchor": "how-do-you-roll-back-and-prove-nothing-regressed",
      "url": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines#how-do-you-roll-back-and-prove-nothing-regressed"
    },
    {
      "depth": 2,
      "text": "Does ChangeGamer version anything like this?",
      "anchor": "does-changegamer-version-anything-like-this",
      "url": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines#does-changegamer-version-anything-like-this"
    },
    {
      "depth": 2,
      "text": "Schema versioning checklist",
      "anchor": "schema-versioning-checklist",
      "url": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines#schema-versioning-checklist"
    }
  ],
  "faq": [
    {
      "question": "When does a document extraction schema change need a new version?",
      "answer": "Any change that can alter a stored value or how a downstream consumer reads it needs a new version: removing, renaming, splitting or merging a field, making an optional field required, changing nesting or units, and editing a field description or prompt. A pure additive nullable field may stay compatible, but it is still worth recording (author design)."
    },
    {
      "question": "Should I backfill old documents when the extraction schema changes?",
      "answer": "Backfill only the fields a consumer actually needs from old documents. Lazy upcast, which converts a record to the new shape when it is read, is cheaper for rarely read data, while eager backfill suits fields that reports or decisions depend on. Both choices are author design, not a published standard."
    },
    {
      "question": "Does upgrading the OCR parser or the model count as a schema change?",
      "answer": "A parser or model upgrade counts as a change to the effective version even though the schema file stays identical. Parsed text, extracted values and confidence signals can all shift, so the parser and model pin belong in the same version manifest as the schema (author inference)."
    },
    {
      "question": "How do I roll back a bad extraction schema release?",
      "answer": "Point readers back at the previous manifest and keep the previous records, which is possible only if new records were written alongside the old ones instead of over them. Then run the labelled sample against both versions to find what regressed (author design)."
    }
  ],
  "body": "A document extraction schema should be versioned as part of a manifest that also names the parser, the model pin and the prompt text, with records written beside their predecessors rather than over them. The [AI document extraction pipeline](/articles/ai-document-extraction-pipeline-in-production) pillar gives the short rule and a small breaking-change table; this piece covers the operations it does not: migrating stored records, reprocessing, cutover, correction migration and rollback. Facts cite the [Document Extraction and Parsing for Agents](/resources/document-extraction-for-agents) reference for the two-stage split, the [Structured Outputs and JSON Mode](/resources/structured-outputs-and-json-mode) reference (updated 2026-09-26) and the [Shipping Agents to Production](/resources/shipping-agents-to-production) checklist. Everything else is author design or inference, labelled.\n\n## Which schema changes are breaking beyond a rename?\n\nAny change that can alter a stored value, or how a consumer reads it, is breaking, and several of them leave the schema looking almost identical. The list below covers changes the pillar's table does not (adding an optional field, renaming and enum widening stay there); all classifications are author inference.\n\n| Change | Class | Why it bites |\n|---|---|---|\n| Remove a field | Breaking | Consumers that read it get absent data they treat as an error or as zero |\n| Optional to required | Breaking for old records | Old records lack a value the new contract demands |\n| Edit a field description or prompt | Behavioural | The shape is unchanged but the extracted values can shift |\n| Split one field into two | Breaking | Old records carry a combined value that must be parsed apart |\n| Merge two fields into one | Breaking | The old pair must be recombined, and the rule must be written down |\n| Change nesting depth | Breaking | Every path expression in a consumer breaks |\n| Normalise units or locale (dates, decimal marks) | Breaking | Old and new values look alike but mean different things |\n\nThe behavioural row is easy to overlook (author inference). A description is an instruction to the model, which is why [writing the schema](/articles/schema-constrained-field-extraction-from-documents) treats descriptions as part of the contract. Treat a tightened validation rule the same way: as [table and form validation](/articles/table-and-form-validation-for-document-extraction) explains, rules decide what reaches review, so the rule set carries a version too (author design).\n\n## How should old records represent absent versus null fields?\n\nOld records should leave a field missing when it did not exist in their schema version and hold null when the extractor looked and the document had nothing, which is the null convention that [schema-constrained field extraction](/articles/schema-constrained-field-extraction-from-documents) uses for \"not on the page\". The two mean different things: missing says \"never asked\", null says \"asked, not on the page\". Collapsing them is what makes a later backfill dangerous, because a null you cannot tell from a missing field might be re-asked or might be a true negative (author design).\n\nThe practical rule is to stamp every record with its schema version and let the reader decide: a field missing from a record whose version predates the field is \"never asked\"; the same gap in a record whose version includes the field is null. If a field later becomes required, only records at or above the introducing version can be validated against it. Older ones either get backfilled or are exempt by version, never silently failed.\n\n## Lazy upcast or eager backfill: how do you migrate stored records?\n\nChoose lazy upcast for data read rarely or by one consumer, and eager backfill for fields that feed reports, decisions or joins. Neither is a published standard; both are author design.\n\n- **Lazy upcast.** A reader function converts a record from version N to the current shape on read, using pure, tested conversion code (rename, unit conversion, default for absent). No document is reprocessed, so it is cheap and reversible, but every consumer must go through the reader.\n- **Eager backfill.** A job rewrites or re-extracts records ahead of time so every row has the new shape. Queries stay simple, but you pay in compute and review load, and you risk overwriting good data.\n\nA field that conversion cannot derive from an old record can only be filled by asking the document again, which is the next section.\n\n## How do you reprocess documents from the stored parse artifact?\n\nReprocess by re-running only the extraction stage on the saved parse output, selecting records by version, and keying the job so a retry cannot duplicate or overwrite. The pillar's two-stage design keeps the Stage A artifact for this reason; this section adds the job rules (author design).\n\n1. **Scope by version.** Select records where the manifest id is below the target. Never \"all documents\".\n2. **Write beside, not over.** Each output is a new record carrying the new manifest id, linked to the old one by document hash.\n3. **Make it idempotent.** Key each result by document hash plus manifest id, so a crashed job can restart without double-writing.\n4. **Route normally.** Results pass the same checks and the same review routing as live traffic.\n5. **Keep refusals visible.** The structured-outputs reference states that `refusal` and `max_tokens` stop reasons bypass the schema guarantee, so a batch must count those separately, not discard them.\n\nRe-parsing the original files instead of the stored artifact changes two variables at once, parser and schema, and the diff is unreadable (author inference). Do that only when the parser itself is what you are migrating, and treat it as a new manifest.\n\n## What belongs in a version manifest?\n\nA version manifest names schema, parser, model pin, prompt and validation rules as a single identifier, and every record stores that identifier. The [shipping agents to production](/resources/shipping-agents-to-production) checklist uses a composite tag of code, prompt template, model pin and tool contracts; carrying that idea over to extraction gives a five-part manifest once validation rules are added (author design). An illustrative example (all ids invented), using the `s-vN`, `p-vN` naming from the [review fallback article](/articles/human-review-fallback-for-document-extraction):\n\n```json\n{\n  \"manifest_id\": \"m-2026-10-27-a\",\n  \"schema\": \"s-v8\",\n  \"parser\": \"p-v3\",\n  \"model\": \"model-pin-2026-09\",\n  \"prompt\": \"pr-v5\",\n  \"validation_rules\": \"r-v4\",\n  \"supersedes\": \"m-2026-09-12-b\"\n}\n```\n\nThe consequence is easy to miss: a parser upgrade or a model pin change moves the effective version even when `s-v8` is untouched. Parsed text changes, extracted values change, and any confidence signal changes meaning. Bump the manifest, run the regression sample, and re-check review cut-offs. The model id is a placeholder.\n\n## How do you cut over without losing records?\n\nCut over in stages: dual-write, dual-read, switch, then retire. This is a standard migration pattern applied here as author design, with no source specific to extraction.\n\n1. **Dual-write.** New documents are extracted under both manifests; the old one stays authoritative.\n2. **Compare.** Diff the two outputs per field, and send disagreements to review.\n3. **Dual-read.** Consumers read the new record where present and fall back to the old one, behind a flag.\n4. **Switch.** Make the new manifest authoritative; keep dual-write on for a defined window.\n5. **Retire.** Stop writing the old manifest only after the window passes with no regression; keep stored records per your retention rules.\n\nWarm the new path first. The structured-outputs reference states that on Anthropic a schema change invalidates the grammar cache and adds first-request latency, so send a few synthetic requests before moving traffic, rather than letting the first live documents absorb the delay.\n\n## What happens to reviewer corrections when the schema changes?\n\nStore each correction with the manifest id it was made under, and migrate it with the same upcast rules as the record it corrects. A correction made against a field that has since been split or renamed is otherwise orphaned, or worse, applied to the wrong field (author design). How corrections are collected is covered in [the review fallback article](/articles/human-review-fallback-for-document-extraction); here the point is only that they carry a version.\n\nWhen a correction cannot be converted exactly, mark it unmigrated instead of guessing, and exclude it from the regression set until a reviewer re-confirms it against the new shape.\n\n## How do you roll back and prove nothing regressed?\n\nRoll back by pointing readers at the previous manifest and keeping its records, then compare both manifests on a labelled sample. Rollback is only a pointer change if you wrote beside, not over (author inference), which is the reason for the write rules above.\n\nThe regression run needs three things (author design):\n\n- A labelled sample that includes past reviewer corrections and the hardest document families.\n- Per-field accuracy for both manifests, not one overall score, because a gain on one field can hide a loss on another.\n- A comparison of review-routing rate, since a stricter rule or looser prompt shows up there first.\n\nFix the acceptance rule, such as \"no field class may get worse\", before the run.\n\n## Does ChangeGamer version anything like this?\n\nAs of October 2026 ChangeGamer runs no document extraction, so none of the above is measured on its data. One verifiable parallel: `src/data/corpus-version.ts` holds `CORPUS_VERSION` (1.27.1, whose PATCH bump is dated 2026-10-25) with a dated comment for every bump, MINOR for a resource addition and PATCH for content edits. Separately, the build (`src/lib/articles.ts`) fails if the article layer breaks cluster rules, so a structural change cannot ship silently.\n\n## Schema versioning checklist\n\n- Every record stores its manifest id; the manifest names schema, parser, model, prompt and rules.\n- Changes classed as compatible, behavioural or breaking before merge.\n- Absent and null kept distinct in old records.\n- Upcast code is pure and tested; backfill is scoped by version and idempotent.\n- New records written beside old ones; old ones kept until the regression run passes.\n- A warm-up request before cutover; a flag for dual-read.\n- Corrections stored with their manifest id.\n- Acceptance rule written before the regression run.",
  "cluster": {
    "id": "document-pipelines",
    "title": "Document extraction pipelines in production",
    "description": "How to run AI document extraction as a production pipeline — OCR vs vision-model vs hybrid parsing, schema-constrained field extraction, table and form validation, human-review fallback, schema versioning, cost per page, open-source vs managed tooling and feeding a system of record — not chunking and retrieval (chunking-documents-for-rag, rag-ingestion-pipeline), not generic structured-output and tool-call mechanics (structured-outputs-vs-tool-calling, tool-calling-contracts-for-ai-agents) and not injection defense in depth (prompt-injection-defense-in-depth-for-agents).",
    "status": "complete",
    "pillar": {
      "slug": "ai-document-extraction-pipeline-in-production",
      "title": "AI Document Extraction in Production: Parse, Extract, Validate, Review",
      "description": "How to run AI document extraction as a production pipeline: separate parsing from schema-constrained field extraction, choose OCR, a vision model or a hybrid, validate values, route doubtful ones to a human, version the schema, and work out cost per page.",
      "kind": "pillar",
      "order": 0,
      "html": "https://changegamer.ai/articles/ai-document-extraction-pipeline-in-production",
      "markdown": "https://changegamer.ai/articles/ai-document-extraction-pipeline-in-production.md",
      "json": "https://changegamer.ai/api/articles/ai-document-extraction-pipeline-in-production.json"
    },
    "articles": [
      {
        "slug": "ocr-vs-vision-model-vs-hybrid-document-parsing",
        "title": "OCR vs Vision Language Model for Document Parsing: How to Choose Stage A",
        "description": "OCR vs vision language model for document parsing: a decision order keyed on residency, layout variance, volume and known schema, how a hybrid arbitrates disagreements, a per-field bake-off, failure-mode mapping, page batching and privacy.",
        "kind": "sub",
        "order": 1,
        "html": "https://changegamer.ai/articles/ocr-vs-vision-model-vs-hybrid-document-parsing",
        "markdown": "https://changegamer.ai/articles/ocr-vs-vision-model-vs-hybrid-document-parsing.md",
        "json": "https://changegamer.ai/api/articles/ocr-vs-vision-model-vs-hybrid-document-parsing.json"
      },
      {
        "slug": "schema-constrained-field-extraction-from-documents",
        "title": "Schema-Constrained Field Extraction From Documents: Settings and Failure Branches",
        "description": "Schema constrained field extraction from documents: nullable fields, enums with an escape value, evidence fields, a branch table for refusal, truncation, parse failure and empty output, a one-retry policy, and field confidence as inference.",
        "kind": "sub",
        "order": 2,
        "html": "https://changegamer.ai/articles/schema-constrained-field-extraction-from-documents",
        "markdown": "https://changegamer.ai/articles/schema-constrained-field-extraction-from-documents.md",
        "json": "https://changegamer.ai/api/articles/schema-constrained-field-extraction-from-documents.json"
      },
      {
        "slug": "table-and-form-validation-for-document-extraction",
        "title": "Table and Form Validation for Document Extraction: Row Counts, Totals and Required Fields",
        "description": "Validating table and form extraction from documents: rows split across pages, subtotal rows, exact money reconciliation, column sums, required-field maps per form type, cross-field checks and a failed-check-to-route table.",
        "kind": "sub",
        "order": 3,
        "html": "https://changegamer.ai/articles/table-and-form-validation-for-document-extraction",
        "markdown": "https://changegamer.ai/articles/table-and-form-validation-for-document-extraction.md",
        "json": "https://changegamer.ai/api/articles/table-and-form-validation-for-document-extraction.json"
      },
      {
        "slug": "human-review-fallback-for-document-extraction",
        "title": "Human Review Fallback for Document Extraction: Triggers, Capacity and the Correction Loop",
        "description": "Designing the human review fallback for document extraction: a routing-trigger table, cross-checking high-stakes fields, review capacity and backlog policy, the review item record and a correction loop that feeds regression sets.",
        "kind": "sub",
        "order": 4,
        "html": "https://changegamer.ai/articles/human-review-fallback-for-document-extraction",
        "markdown": "https://changegamer.ai/articles/human-review-fallback-for-document-extraction.md",
        "json": "https://changegamer.ai/api/articles/human-review-fallback-for-document-extraction.json"
      },
      {
        "slug": "extraction-schema-versioning-for-document-pipelines",
        "title": "Versioning a Document Extraction Schema: Migrations, Manifests and Rollback",
        "description": "How to version a document extraction schema: which edits break stored records, absent versus null fields, lazy upcast versus backfill, reprocessing from stored parse output, a version manifest, dual-read cutover, correction migration and rollback.",
        "kind": "sub",
        "order": 5,
        "html": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines",
        "markdown": "https://changegamer.ai/articles/extraction-schema-versioning-for-document-pipelines.md",
        "json": "https://changegamer.ai/api/articles/extraction-schema-versioning-for-document-pipelines.json"
      },
      {
        "slug": "cost-per-page-for-document-extraction",
        "title": "Document Extraction Cost per Page: Tokens, Retries, Review and Budget Alarms",
        "description": "How to model document extraction cost per page beyond the parse fee: page-count mix versus per-page pricing, image tokens versus per-call overhead, retries by failure class, review share, batch versus sync, and a budget kill switch on cost per accepted document.",
        "kind": "sub",
        "order": 6,
        "html": "https://changegamer.ai/articles/cost-per-page-for-document-extraction",
        "markdown": "https://changegamer.ai/articles/cost-per-page-for-document-extraction.md",
        "json": "https://changegamer.ai/api/articles/cost-per-page-for-document-extraction.json"
      },
      {
        "slug": "open-source-vs-managed-document-parsing",
        "title": "Open Source vs Managed Document Parsing: A Decision Order, Operating Burden and Exit Plan",
        "description": "How to decide between self-hosted open-source parsers and a managed parsing service: residency and licence first, volume and operations capacity second, reversibility third, plus the cost of operating models, lock-in, a single-interface fallback design and a trial protocol on your own documents.",
        "kind": "sub",
        "order": 7,
        "html": "https://changegamer.ai/articles/open-source-vs-managed-document-parsing",
        "markdown": "https://changegamer.ai/articles/open-source-vs-managed-document-parsing.md",
        "json": "https://changegamer.ai/api/articles/open-source-vs-managed-document-parsing.json"
      },
      {
        "slug": "feeding-extraction-output-into-a-system-of-record",
        "title": "Writing Extracted Document Data to a System of Record: Matching, Keys, States and Replay",
        "description": "How to write extracted document data to a system of record: matching an extracted record to an existing one, retry keys versus business keys, a staged-to-committed state machine, per-record versus per-document atomicity, dead-letter queues, replay from the saved parse artifact and the write-boundary log.",
        "kind": "sub",
        "order": 8,
        "html": "https://changegamer.ai/articles/feeding-extraction-output-into-a-system-of-record",
        "markdown": "https://changegamer.ai/articles/feeding-extraction-output-into-a-system-of-record.md",
        "json": "https://changegamer.ai/api/articles/feeding-extraction-output-into-a-system-of-record.json"
      },
      {
        "slug": "multimodal-document-understanding-for-extraction-pipelines",
        "title": "Multimodal Document Understanding for Extraction: Charts, Diagrams, Photos, Stamps and Video Frames",
        "description": "How to handle non-text visuals in a document extraction pipeline: when OCR is not enough, extracting a value versus a description versus a class, region-level routing, figure-aware page packing, poor scans, video frames, image-borne injection and per-visual evaluation.",
        "kind": "sub",
        "order": 9,
        "html": "https://changegamer.ai/articles/multimodal-document-understanding-for-extraction-pipelines",
        "markdown": "https://changegamer.ai/articles/multimodal-document-understanding-for-extraction-pipelines.md",
        "json": "https://changegamer.ai/api/articles/multimodal-document-understanding-for-extraction-pipelines.json"
      },
      {
        "slug": "monitoring-document-extraction-accuracy-in-production",
        "title": "Monitoring Document Extraction Accuracy in Production: Audit Samples, Field Metrics and Drift",
        "description": "How to monitor document extraction accuracy in production: a stratified audit sample with independent labels, per-field metrics with intervals, extraction-specific drift triggers, thresholds from your own variance, and regression intake.",
        "kind": "sub",
        "order": 10,
        "html": "https://changegamer.ai/articles/monitoring-document-extraction-accuracy-in-production",
        "markdown": "https://changegamer.ai/articles/monitoring-document-extraction-accuracy-in-production.md",
        "json": "https://changegamer.ai/api/articles/monitoring-document-extraction-accuracy-in-production.json"
      },
      {
        "slug": "splitting-long-documents-for-document-extraction",
        "title": "How to Split Long Documents for Extraction: Batches, Overlap and Reassembly",
        "description": "How to split long documents for extraction: size batches from measured tokens per page, choose section or page windows, overlap pages, carry table headers, reconcile conflicts across batches and reassemble with provenance.",
        "kind": "sub",
        "order": 11,
        "html": "https://changegamer.ai/articles/splitting-long-documents-for-document-extraction",
        "markdown": "https://changegamer.ai/articles/splitting-long-documents-for-document-extraction.md",
        "json": "https://changegamer.ai/api/articles/splitting-long-documents-for-document-extraction.json"
      }
    ]
  },
  "navigation": {
    "pillar": {
      "slug": "ai-document-extraction-pipeline-in-production",
      "title": "AI Document Extraction in Production: Parse, Extract, Validate, Review",
      "description": "How to run AI document extraction as a production pipeline: separate parsing from schema-constrained field extraction, choose OCR, a vision model or a hybrid, validate values, route doubtful ones to a human, version the schema, and work out cost per page.",
      "kind": "pillar",
      "order": 0,
      "html": "https://changegamer.ai/articles/ai-document-extraction-pipeline-in-production",
      "markdown": "https://changegamer.ai/articles/ai-document-extraction-pipeline-in-production.md",
      "json": "https://changegamer.ai/api/articles/ai-document-extraction-pipeline-in-production.json"
    },
    "previous": {
      "slug": "human-review-fallback-for-document-extraction",
      "title": "Human Review Fallback for Document Extraction: Triggers, Capacity and the Correction Loop",
      "description": "Designing the human review fallback for document extraction: a routing-trigger table, cross-checking high-stakes fields, review capacity and backlog policy, the review item record and a correction loop that feeds regression sets.",
      "kind": "sub",
      "order": 4,
      "html": "https://changegamer.ai/articles/human-review-fallback-for-document-extraction",
      "markdown": "https://changegamer.ai/articles/human-review-fallback-for-document-extraction.md",
      "json": "https://changegamer.ai/api/articles/human-review-fallback-for-document-extraction.json"
    },
    "next": {
      "slug": "cost-per-page-for-document-extraction",
      "title": "Document Extraction Cost per Page: Tokens, Retries, Review and Budget Alarms",
      "description": "How to model document extraction cost per page beyond the parse fee: page-count mix versus per-page pricing, image tokens versus per-call overhead, retries by failure class, review share, batch versus sync, and a budget kill switch on cost per accepted document.",
      "kind": "sub",
      "order": 6,
      "html": "https://changegamer.ai/articles/cost-per-page-for-document-extraction",
      "markdown": "https://changegamer.ai/articles/cost-per-page-for-document-extraction.md",
      "json": "https://changegamer.ai/api/articles/cost-per-page-for-document-extraction.json"
    }
  },
  "resources": [
    {
      "slug": "document-extraction-for-agents",
      "html": "https://changegamer.ai/resources/document-extraction-for-agents",
      "markdown": "https://changegamer.ai/resources/document-extraction-for-agents.md",
      "json": "https://changegamer.ai/api/resources/document-extraction-for-agents.json"
    },
    {
      "slug": "structured-outputs-and-json-mode",
      "html": "https://changegamer.ai/resources/structured-outputs-and-json-mode",
      "markdown": "https://changegamer.ai/resources/structured-outputs-and-json-mode.md",
      "json": "https://changegamer.ai/api/resources/structured-outputs-and-json-mode.json"
    },
    {
      "slug": "shipping-agents-to-production",
      "html": "https://changegamer.ai/resources/shipping-agents-to-production",
      "markdown": "https://changegamer.ai/resources/shipping-agents-to-production.md",
      "json": "https://changegamer.ai/api/resources/shipping-agents-to-production.json"
    }
  ]
}