Anthropic CCA-F Questions & Answers

Delivery notices require a different extraction schema. The invoice schema lacks fields that are meaningful for delivery status and notice metadata. Incoming documents may be invoices or delivery notices and one extractor must run. Instead, require selection of an appropriate schema tool without fixing its name.

The service needs machine-readable output for each form. Unstructured prose cannot be reliably consumed as typed extraction by the downstream service. Incoming documents may be invoices or delivery notices and one extractor must run. Instead, require selection of an appropriate schema tool without fixing its name.

Order is not document classification. Document order in a queue says nothing about whether the current record is a notice or invoice. Incoming documents may be invoices or delivery notices and one extractor must run. Instead, require selection of an appropriate schema tool without fixing its name.

One version legitimately lacks the other's fields. A shared all-required schema would reject valid earlier versions with fewer legitimate fields. A discriminated or tagged union validates one explicitly identified form variant with only that version's mandatory fields. A packet contains one of two recognized form versions with different mandatory fields. Instead, use a tagged union schema to validate each version without fabricating absent fields.

Length is not a reliable discriminator. The number of characters does not reliably distinguish the form edition or its required keys. A packet contains one of two recognized form versions with different mandatory fields. Instead, use a tagged union schema to validate each version without fabricating absent fields.

The form's actual version determines its contract. Selecting the least demanding schema could silently omit fields mandated by the actual version. A discriminated or tagged union validates one explicitly identified form variant with only that version's mandatory fields. A packet contains one of two recognized form versions with different mandatory fields.

Schema pressure cannot create missing source evidence. Multiple retries cannot reveal information the source never contained; the schema needs optionality. A document legitimately omits counterparty phone and the model invents one to satisfy a required string. Instead, represent the phone as optional or nullable.

Valid documents should not fail on nonessential fields. Rejecting a valid invoice over an optional attribute would lose legitimate transaction details. Requiring a nonempty phone string when the source has none creates fabricated contact data that downstream systems may misuse.

A fabricated contact becomes downstream misinformation. A fake phone number may be contacted, causing operational and privacy harm. Requiring a nonempty phone string when the source has none creates fabricated contact data that downstream systems may misuse. A document legitimately omits counterparty phone and the model invents one to satisfy a required string.

Invented identifiers violate source fidelity. A fabricated identifier cannot be traced to the source and may collide with legitimate records. Downstream settlement cannot process a record without invoice ID, but a permissive schema omits it silently. Instead, keep invoice ID required and reject missing values.

A blank required identifier is not meaningful data. An empty string is not a usable primary reference for reconciliation or dispute handling. A required invoice identifier is a processing precondition rather than a cosmetic field. Downstream settlement cannot process a record without invoice ID, but a permissive schema omits it silently.

Settlement cannot identify the record without it. Treating the ID as optional breaks downstream joins between settlement and original documents. Downstream settlement cannot process a record without invoice ID, but a permissive schema omits it silently. Instead, keep invoice ID required and reject missing values. A required invoice identifier is a processing precondition rather than a cosmetic field.

Omitting real transactions distorts totals. Dropping the item hides actual spending and prevents the totals from reconciling. Forcing an unseen expense into travel contaminates reporting totals and hides a taxonomy extension need. Instead, provide an other value plus descriptive detail field.

Uncontrolled schema variation breaks consumers. Mutating an enum on each record produces incompatible outputs and unreviewed consumer behavior. A new expense type is not in the existing enum and the model labels it travel. Instead, provide an other value plus descriptive detail field.

Travel would be a false classification. Misclassifying the unknown expense as travel contaminates category-level reporting. Forcing an unseen expense into travel contaminates reporting totals and hides a taxonomy extension need. A new expense type is not in the existing enum and the model labels it travel.

The document itself might be wrong and needs evidence. Changing the source-stated value without review could conceal an actual supplier billing error. A schema-valid invoice reports line items totaling 900 but a stated total of 800. Instead, validate JSON schema/types and independently verify totals and cross-field business constraints. JSON-schema validation checks field shapes but does not establish arithmetic consistency or documentary truth.

Type changes do not reconcile business contradictions. Converting amounts to strings weakens type checks while leaving the 900-versus-800 mismatch intact. A schema-valid invoice reports line items totaling 900 but a stated total of 800. Instead, validate JSON schema/types and independently verify totals and cross-field business constraints. Recomputing line totals and comparing to the stated 800/900 discrepancy exposes a semantic mismatch that must be reconciled.

JSON shape cannot guarantee arithmetic consistency. A parser cannot determine whether invoice arithmetic is consistent with reported totals. Recomputing line totals and comparing to the stated 800/900 discrepancy exposes a semantic mismatch that must be reconciled. A schema-valid invoice reports line items totaling 900 but a stated total of 800. Instead, validate JSON schema/types and independently verify totals and cross-field business constraints.

Runtime locale may not match document origin. The processing server's locale is incidental and may disagree with the document's originating jurisdiction. A source uses 03/04/2026 and downstream systems expect ISO dates. Instead, specify source-locale interpretation and normalization before output acceptance.

The downstream contract expects normalized dates. A raw slash-delimited date violates the agreed ISO output and may fail validators. A source uses 03/04/2026 and downstream systems expect ISO dates. Instead, specify source-locale interpretation and normalization before output acceptance.

Ambiguous numeric order is not universally resolved. March 4 and April 3 are both possible interpretations depending on documented locale. A source uses 03/04/2026 and downstream systems expect ISO dates. Instead, specify source-locale interpretation and normalization before output acceptance.

The consumer expects a typed number. A numeric calculation cannot be trusted if the input remains an unvalidated string. A model returns an amount as a string although the calculator expects a number. Instead, reject type mismatch with structured feedback.

A required business value cannot disappear silently. Omitting an essential amount can result in an apparently complete record without usable value. A model returns an amount as a string although the calculator expects a number. Instead, reject type mismatch with structured feedback.

Defaulting hides source and type errors. Zero is a real amount and replacing invalid text with it fabricates a transaction. A model returns an amount as a string although the calculator expects a number. Instead, reject type mismatch with structured feedback.

Changing labels does not fix source attribution. Renaming fields after extraction still preserves the values in their incorrect relationships. The record is syntactically valid but assigns customer ID to the order-id field. Instead, use source-grounded semantic checks in addition to schema rules.

Removing constraints makes the bug less detectable. Removing identifier constraints makes accidental or malicious misattribution harder to discover. Cross-checking identifiers against source labels and identity relationships catches swapped customer and order IDs. Instead, use source-grounded semantic checks in addition to schema rules.

A semantic ID swap can pass structural checks. A string type check cannot distinguish a customer identifier from an order identifier. Cross-checking identifiers against source labels and identity relationships catches swapped customer and order IDs. The record is syntactically valid but assigns customer ID to the order-id field. Instead, use source-grounded semantic checks in addition to schema rules.

The schema cannot reliably reconcile line-item ownership. An overloaded ID string prevents reconciliation of individual invoice totals and line ownership. Multiple source invoices should yield multiple individually identifiable records, each with its own totals and validation receipts. A document contains two invoices but extraction produces one merged object.

The batch loses legitimate transactions. Ignoring the second invoice violates the required coverage of the source packet. A document contains two invoices but extraction produces one merged object. Instead, use an array of individually identified records and per-record validation. Multiple source invoices should yield multiple individually identifiable records, each with its own totals and validation receipts.

Distinct source records require independent identities. Combining invoices merges independent payment obligations and can double-count or omit items. Multiple source invoices should yield multiple individually identifiable records, each with its own totals and validation receipts. Instead, use an array of individually identified records and per-record validation.

Consumers cannot know which contract applies. A union of undocumented old and new shapes leaves consumers unable to apply the right validation. A new field becomes mandatory but old consumers still deserialize the prior schema. Instead, publish a versioned schema contract and provide a compatible migration path to consumers. Making a new schema field mandatory changes the contract consumed by deployed readers.

Old deserializers may fail on the new required field. The new producer could emit a field older deserializers reject or misinterpret. A new field becomes mandatory but old consumers still deserialize the prior schema. Instead, publish a versioned schema contract and provide a compatible migration path to consumers. Versioning and a compatibility migration let old applications continue decoding while new producers and consumers coordinate.

Fabrication undermines provenance and business semantics. Default text masquerades as an observed value when the new field has no source. Versioning and a compatibility migration let old applications continue decoding while new producers and consumers coordinate. A new field becomes mandatory but old consumers still deserialize the prior schema.