# noCV engineering task library Content version 5 Fictional engineering practice briefs. Starter repositories, fixtures, automated grading, and verified ownership are not included. Tests, patches, and runbooks are requested deliverables. They become Outcome Evidence only through a qualified Mission and immutable Evidence IDs. Independent adaptation must be observed under a declared verification policy and cite immutable Evidence IDs. Completing a planning ticket establishes no Ownership Evidence. ## BEXTRACT — Reviewable invoice field extraction A fictional procurement team receives inconsistent supplier documents. It wants draft records that staff can review without treating model output as authoritative accounting data. **Field:** Applied AI. **Suggested stack:** TypeScript, JSON Schema, AiEvaluationProvider. **Engineer value:** Practice schema constraints, numerical checks, review workflows, and safe model boundaries. **Company value:** Produce an inspectable extraction prototype that can reduce manual transcription while preserving review control. **Delivery agreement:** Deliver draft extraction and correction behavior; no real invoices, payment execution, or accounting advice. ### Setup prerequisites - Author synthetic text invoices with ambiguous dates, currencies, and line items; use a deterministic provider double. ### Define extraction rules Represent fields, provenance, and ambiguity explicitly. #### BEXTRACT-101 — Define invoice fields with explicit missing and ambiguous states **Task · Medium priority · Foundational** noCV practice brief v5 · BEXTRACT-101 · Reviewable invoice field extraction Fictional engineering practice briefs. Starter repositories, fixtures, automated grading, and verified ownership are not included. Phase: Define extraction rules. Depends on: No preceding ticket. Difficulty: Foundational. Estimated focused work: 60 minutes; setup and prerequisite tickets are additional. Estimated field mix: Applied AI 80% · Data engineering 20%. Field percentages are editorial estimates of the ticket's engineering focus. They total 100%; they are not measured time, proficiency scores, or ownership evidence. Empty strings currently mean both absent values and extraction failures. Acceptance criteria - Represent value, missing, and ambiguous states distinctly. - Declare required fields before a draft is reviewable. - Preserve the original textual date and amount references. Implementation constraints - Do not guess missing tax or currency values. Verification - Parse a complete synthetic invoice. - Represent missing currency and an ambiguous date without invented values. Deliverables - Extraction schema. Rollout and recovery: Version the schema before creating drafts; retain raw synthetic source references. Project prerequisites: Author synthetic text invoices with ambiguous dates, currencies, and line items; use a deterministic provider double. Engineer value: Practice schema constraints, numerical checks, review workflows, and safe model boundaries. Company value: Produce an inspectable extraction prototype that can reduce manual transcription while preserving review control. AI tools are welcome during implementation. Record assumptions, review the result, and verify its behavior. Planning status does not create Outcome Evidence or Ownership Evidence. #### BEXTRACT-102 — Bind extraction runs to immutable document digests **Task · High priority · Intermediate** noCV practice brief v5 · BEXTRACT-102 · Reviewable invoice field extraction Fictional engineering practice briefs. Starter repositories, fixtures, automated grading, and verified ownership are not included. Phase: Define extraction rules. Depends on: BEXTRACT-101. Difficulty: Intermediate. Estimated focused work: 120 minutes; setup and prerequisite tickets are additional. Estimated field mix: Applied AI 50% · Data engineering 30% · Storage systems 20%. Field percentages are editorial estimates of the ticket's engineering focus. They total 100%; they are not measured time, proficiency scores, or ownership evidence. A corrected document is uploaded under the same filename and the old result appears current. Acceptance criteria - Identify runs by document digest and extractor version. - Preserve earlier runs when content changes. - Reject a result bound to another document digest. Implementation constraints - Filenames are display labels, not identity. Verification - Replay an unchanged document deterministically. - Change one byte and prevent reuse of the old result. Deliverables - Run identity contract. Rollout and recovery: Append new runs on document changes; never rewrite accepted extraction history. Project prerequisites: Author synthetic text invoices with ambiguous dates, currencies, and line items; use a deterministic provider double. Engineer value: Practice schema constraints, numerical checks, review workflows, and safe model boundaries. Company value: Produce an inspectable extraction prototype that can reduce manual transcription while preserving review control. AI tools are welcome during implementation. Record assumptions, review the result, and verify its behavior. Planning status does not create Outcome Evidence or Ownership Evidence. ### Validate drafts Reject unsupported or inconsistent provider output. #### BEXTRACT-103 — Validate source spans for every extracted material field **Task · High priority · Advanced** noCV practice brief v5 · BEXTRACT-103 · Reviewable invoice field extraction Fictional engineering practice briefs. Starter repositories, fixtures, automated grading, and verified ownership are not included. Phase: Validate drafts. Depends on: BEXTRACT-102. Difficulty: Advanced. Estimated focused work: 180 minutes; setup and prerequisite tickets are additional. Estimated field mix: Applied AI 80% · Data engineering 20%. Field percentages are editorial estimates of the ticket's engineering focus. They total 100%; they are not measured time, proficiency scores, or ownership evidence. The provider returns a supplier address that appears nowhere in the document. Acceptance criteria - Require bounded source spans for material values. - Resolve spans against the exact source version. - Flag transformed values with their documented normalization rule. Implementation constraints - Provider prose cannot substitute for source support. Verification - Accept a supported normalized date. - Reject out-of-range spans and invented supplier values. Deliverables - Source-grounding validator. Rollout and recovery: Keep unsupported drafts blocked for review; retain safe failure categories. Project prerequisites: Author synthetic text invoices with ambiguous dates, currencies, and line items; use a deterministic provider double. Engineer value: Practice schema constraints, numerical checks, review workflows, and safe model boundaries. Company value: Produce an inspectable extraction prototype that can reduce manual transcription while preserving review control. AI tools are welcome during implementation. Record assumptions, review the result, and verify its behavior. Planning status does not create Outcome Evidence or Ownership Evidence. #### BEXTRACT-104 — Check line-item arithmetic using exact decimal rules **Task · High priority · Intermediate** noCV practice brief v5 · BEXTRACT-104 · Reviewable invoice field extraction Fictional engineering practice briefs. Starter repositories, fixtures, automated grading, and verified ownership are not included. Phase: Validate drafts. Depends on: BEXTRACT-101, BEXTRACT-103. Difficulty: Intermediate. Estimated focused work: 150 minutes; setup and prerequisite tickets are additional. Estimated field mix: Data engineering 60% · Applied AI 40%. Field percentages are editorial estimates of the ticket's engineering focus. They total 100%; they are not measured time, proficiency scores, or ownership evidence. Binary floating point produces an apparent mismatch on an otherwise consistent invoice. Acceptance criteria - Use declared currency scale and rounding rules. - Compare quantity, price, subtotal, and total consistently. - Represent unresolved tax or discount differences as discrepancies. Implementation constraints - Never silently alter document totals to make arithmetic balance. Verification - Validate a rounding-boundary invoice. - Flag inconsistent totals and unsupported precision. Deliverables - Arithmetic consistency checks. Rollout and recovery: Show discrepancies to reviewers; do not promote inconsistent drafts automatically. Project prerequisites: Author synthetic text invoices with ambiguous dates, currencies, and line items; use a deterministic provider double. Engineer value: Practice schema constraints, numerical checks, review workflows, and safe model boundaries. Company value: Produce an inspectable extraction prototype that can reduce manual transcription while preserving review control. AI tools are welcome during implementation. Record assumptions, review the result, and verify its behavior. Planning status does not create Outcome Evidence or Ownership Evidence. #### BEXTRACT-105 — Prevent document text from overriding extraction policy **Bug · High priority · Advanced** noCV practice brief v5 · BEXTRACT-105 · Reviewable invoice field extraction Fictional engineering practice briefs. Starter repositories, fixtures, automated grading, and verified ownership are not included. Phase: Validate drafts. Depends on: BEXTRACT-103. Difficulty: Advanced. Estimated focused work: 180 minutes; setup and prerequisite tickets are additional. Estimated field mix: Applied AI 50% · Security 50%. Field percentages are editorial estimates of the ticket's engineering focus. They total 100%; they are not measured time, proficiency scores, or ownership evidence. An invoice note contains instructions to mark every field verified. Acceptance criteria - Treat document text only as extraction input. - Require the same schema and grounding rules for all responses. - Keep document instructions unable to change review status. Implementation constraints - No external tools or supplier contact is permitted from model output. Verification - Extract a normal note field. - Inject policy-changing text and verify review remains required. Deliverables - Untrusted-document regression cases. Rollout and recovery: Fail closed on invalid output; retain the deterministic provider as a safe development mode. Project prerequisites: Author synthetic text invoices with ambiguous dates, currencies, and line items; use a deterministic provider double. Engineer value: Practice schema constraints, numerical checks, review workflows, and safe model boundaries. Company value: Produce an inspectable extraction prototype that can reduce manual transcription while preserving review control. AI tools are welcome during implementation. Record assumptions, review the result, and verify its behavior. Planning status does not create Outcome Evidence or Ownership Evidence. #### BEXTRACT-106 — Deduplicate concurrent extraction requests without mixing versions **Bug · High priority · Advanced** noCV practice brief v5 · BEXTRACT-106 · Reviewable invoice field extraction Fictional engineering practice briefs. Starter repositories, fixtures, automated grading, and verified ownership are not included. Phase: Validate drafts. Depends on: BEXTRACT-102, BEXTRACT-104. Difficulty: Advanced. Estimated focused work: 210 minutes; setup and prerequisite tickets are additional. Estimated field mix: Applied AI 50% · Database engineering 30% · Distributed systems 20%. Field percentages are editorial estimates of the ticket's engineering focus. They total 100%; they are not measured time, proficiency scores, or ownership evidence. Two upload retries consume duplicate provider budget and occasionally attach the older result to the newer run. Acceptance criteria - Deduplicate exact document and extractor identities. - Reject conflicting reuse of an operation key. - Commit one terminal result and preserve failed-attempt metadata. Implementation constraints - Access providers through AiEvaluationProvider with bounded timeout and retries. Verification - Race identical requests and observe one accepted result. - Change extractor version and ensure a distinct run. Deliverables - Extraction operation coordinator. Rollout and recovery: Adopt one document type first; retry terminal failures through a new explicit operation. Project prerequisites: Author synthetic text invoices with ambiguous dates, currencies, and line items; use a deterministic provider double. Engineer value: Practice schema constraints, numerical checks, review workflows, and safe model boundaries. Company value: Produce an inspectable extraction prototype that can reduce manual transcription while preserving review control. AI tools are welcome during implementation. Record assumptions, review the result, and verify its behavior. Planning status does not create Outcome Evidence or Ownership Evidence. #### BEXTRACT-107 — Reserve extraction cost before processing a document batch **Task · High priority · Intermediate** noCV practice brief v5 · BEXTRACT-107 · Reviewable invoice field extraction Fictional engineering practice briefs. Starter repositories, fixtures, automated grading, and verified ownership are not included. Phase: Validate drafts. Depends on: BEXTRACT-106. Difficulty: Intermediate. Estimated focused work: 150 minutes; setup and prerequisite tickets are additional. Estimated field mix: Applied AI 50% · Database engineering 30% · Backend 20%. Field percentages are editorial estimates of the ticket's engineering focus. They total 100%; they are not measured time, proficiency scores, or ownership evidence. A large upload queue can exceed the tenant's configured processing budget. Acceptance criteria - Estimate a conservative per-document reservation. - Enforce tenant and batch ceilings before dispatch. - Settle actual usage and return unused reservation exactly once. Implementation constraints - Record model, prompt/schema version, timeout, retries, cost, and trace without document text. Verification - Process a batch within the ceiling. - Retry settlement and verify no double charge or negative reservation. Deliverables - Budget accounting tests. Rollout and recovery: Start with a small ceiling; pause new dispatch when accounting state is uncertain. Project prerequisites: Author synthetic text invoices with ambiguous dates, currencies, and line items; use a deterministic provider double. Engineer value: Practice schema constraints, numerical checks, review workflows, and safe model boundaries. Company value: Produce an inspectable extraction prototype that can reduce manual transcription while preserving review control. AI tools are welcome during implementation. Record assumptions, review the result, and verify its behavior. Planning status does not create Outcome Evidence or Ownership Evidence. ### Support correction Preserve human review and evaluate bounded behavior. #### BEXTRACT-108 — Preserve corrections as append-only reviewer decisions **Story · High priority · Advanced** noCV practice brief v5 · BEXTRACT-108 · Reviewable invoice field extraction Fictional engineering practice briefs. Starter repositories, fixtures, automated grading, and verified ownership are not included. Phase: Support correction. Depends on: BEXTRACT-104, BEXTRACT-106. Difficulty: Advanced. Estimated focused work: 210 minutes; setup and prerequisite tickets are additional. Estimated field mix: Backend 50% · Applied AI 30% · Database engineering 20%. Field percentages are editorial estimates of the ticket's engineering focus. They total 100%; they are not measured time, proficiency scores, or ownership evidence. Staff corrections overwrite the extracted values, making later disagreement impossible to investigate. Acceptance criteria - Record original value, corrected value, reviewer, and reason. - Require optimistic revision control for concurrent review. - Keep source and extractor output immutable. Implementation constraints - Review actions must be tenant-scoped. Verification - Correct a draft and inspect both versions. - Reject stale revisions and cross-tenant review attempts. Deliverables - Correction history workflow. Rollout and recovery: Enable on draft records first; retain prior review state when a correction fails. Project prerequisites: Author synthetic text invoices with ambiguous dates, currencies, and line items; use a deterministic provider double. Engineer value: Practice schema constraints, numerical checks, review workflows, and safe model boundaries. Company value: Produce an inspectable extraction prototype that can reduce manual transcription while preserving review control. AI tools are welcome during implementation. Record assumptions, review the result, and verify its behavior. Planning status does not create Outcome Evidence or Ownership Evidence. #### BEXTRACT-109 — Design a field-level extraction evaluation with abstention costs **Task · High priority · Expert** noCV practice brief v5 · BEXTRACT-109 · Reviewable invoice field extraction Fictional engineering practice briefs. Starter repositories, fixtures, automated grading, and verified ownership are not included. Phase: Support correction. Depends on: BEXTRACT-105, BEXTRACT-107, BEXTRACT-108. Difficulty: Expert. Estimated focused work: 360 minutes; setup and prerequisite tickets are additional. Estimated field mix: Applied AI 60% · Quality engineering 40%. Field percentages are editorial estimates of the ticket's engineering focus. They total 100%; they are not measured time, proficiency scores, or ownership evidence. Document-level success hides whether the extractor frequently invents totals or simply leaves them for review. Acceptance criteria - Report supported correct, incorrect, missing, and abstained fields separately. - Include layout, date, currency, and injection variations. - Compare manual-review volume against unsupported-field risk under declared assumptions. Implementation constraints - Deterministic doubles validate plumbing; live model accuracy remains unmeasured. Verification - Detect a fabricated total in the evaluation. - Show an appropriate abstention separately from an incorrect value. Deliverables - Evaluation protocol and synthetic results. Rollout and recovery: Expand document coverage only after reviewing failure categories and unresolved assumptions. Project prerequisites: Author synthetic text invoices with ambiguous dates, currencies, and line items; use a deterministic provider double. Engineer value: Practice schema constraints, numerical checks, review workflows, and safe model boundaries. Company value: Produce an inspectable extraction prototype that can reduce manual transcription while preserving review control. AI tools are welcome during implementation. Record assumptions, review the result, and verify its behavior. Planning status does not create Outcome Evidence or Ownership Evidence. #### BEXTRACT-110 — Prepare a reviewer checklist for ambiguous supplier documents **Chore · Low priority · Foundational** noCV practice brief v5 · BEXTRACT-110 · Reviewable invoice field extraction Fictional engineering practice briefs. Starter repositories, fixtures, automated grading, and verified ownership are not included. Phase: Support correction. Depends on: BEXTRACT-109. Difficulty: Foundational. Estimated focused work: 60 minutes; setup and prerequisite tickets are additional. Estimated field mix: Applied AI 100%. Field percentages are editorial estimates of the ticket's engineering focus. They total 100%; they are not measured time, proficiency scores, or ownership evidence. Reviewers need a consistent way to handle ambiguous dates and unsupported fields. Acceptance criteria - Explain how to inspect source spans and arithmetic discrepancies. - Show how to request correction without editing source history. - Document when a draft must remain unresolved. Implementation constraints - Do not treat review completion as proof of invoice authenticity. Verification - Review one supported synthetic invoice. - Leave an ambiguous supplier document unresolved with a reason. Deliverables - Reviewer guide. Rollout and recovery: Publish alongside the prototype; revise examples when extraction policy changes. Project prerequisites: Author synthetic text invoices with ambiguous dates, currencies, and line items; use a deterministic provider double. Engineer value: Practice schema constraints, numerical checks, review workflows, and safe model boundaries. Company value: Produce an inspectable extraction prototype that can reduce manual transcription while preserving review control. AI tools are welcome during implementation. Record assumptions, review the result, and verify its behavior. Planning status does not create Outcome Evidence or Ownership Evidence.