dataverse-payload-preflight

name	dataverse-payload-preflight
description	Validate a Dataverse create/update payload against LIVE entity metadata in ONE pass — before you POST/PATCH — so you never fix-one-field-and-retrigger. Catches nonexistent columns, invalid option-set values, malformed/missing lookup binds, missing required fields, and the owner-not-provided (SPN-create) trap together. Run it on the FIRST create failure, or up front when a human re-fire or long run gates each test.
allowed-tools	Bash, Read

dataverse-payload-preflight — one-pass payload validation against live metadata

Why this exists. The Contoso assumption-rework retro (2026-06-24): a Create_BalanceSheet failed four times in a row, each on a different field — empty lookup bind /accounts() → invalid sourcechannel option value → undeclared extractionrun columns → "Owner was not provided" — each needing a costly human re-fire. A single metadata-vs-payload sweep finds them all at once. This is the deterministic instrument of default-to-verification: don't reason about the schema, read it.

When to run it (the discipline — expensive-test front-loading)

• Before the first create/update of an entity you didn't author the payload for (an inherited payload, a prior session's, a new entity). Don't assume the schema — validate it.
• On the FIRST create/update failure — not after the second. One pass finds every remaining issue.
• Whenever each test costs a human re-fire or a long run: those are scarce resources; spend cheap static-validation tokens to conserve them (see ../../../ravenclaude-core/best-practices/expensive-test-front-loading.md).

Run it

# 1. Acquire a Dataverse token (see knowledge/dataverse-token-acquisition.md for the decision tree).
export DATAVERSE_TOKEN="$(az account get-access-token --resource https://yourorg.crm.dynamics.com --query accessToken -o tsv)"

# 2. Validate the whole payload against LIVE metadata, in one pass:
python3 preflight.py --org https://yourorg.crm.dynamics.com --entity contoso_balancesheet --payload payload.json

# Offline / in a test (skip the live fetch — feed metadata you captured):
python3 preflight.py --entity contoso_balancesheet --metadata metadata.json --payload payload.json

Output is JSON {entity, ok, error_count, warning_count, violations[]} (each violation has field · kind · severity · detail · fix) plus a human summary. Exit 0 = no error-severity violations; exit 3 = fix-these-before-you-POST.

What it checks (against the entity's live EntityDefinitions)

kind	catches
nonexistent-column	a payload key that isn't an attribute (the extractionrun undeclared-fields trap)
missing-required	an ApplicationRequired/SystemRequired attribute absent from the payload
invalid-option-set	a Picklist value not in the attribute's declared options (the sourcechannel trap)
malformed-lookup-bind	an @odata.bind whose value isn't /<entityset>(<non-empty-id>) (the /accounts() trap)
lookup-needs-bind	a Lookup attribute given a raw value instead of a <nav>@odata.bind
unknown-lookup-target (warn)	a bind to an entity-set that isn't a known target of the entity's lookups
owner-not-provided (warn)	a User/Team-owned entity with no ownerid — the SPN-create "Owner was not provided" trap

Honest limits

• The live-fetch query shapes are version-sensitive (…/Attributes/Microsoft.Dynamics.CRM.<Type>AttributeMetadata, the OptionSet expand) — if a sub-query 400s the script degrades that one check class (a note on stderr) rather than failing; confirm against your org's $metadata [verify-at-use].
• Lookup navigation-property names (the X in [e-mail]) often differ from the lookup's logical name; the validator verifies the bind value shape + target entity-set but treats an unknown nav-name as a warning, not an error (no false-fail). It does not (yet) probe prvCreate/prvAssign privileges — owner-not-provided is a warning pointing you to verify the SPN's role live (don't design around an assumed privilege; test it — the retro's incident #4).
• The pure validate(payload, metadata) is deterministic + unit-tested (hooks/tests/test-preflight.sh); the live fetch is best-effort.

Cross-references

• Token acquisition: ../../knowledge/dataverse-token-acquisition.md.
• REST-first debugging (read the real error envelope, not the portal): ../../knowledge/pbir-fabric-rest-debugging.md.
• The advisory hooks/nudge-dataverse-preflight.sh reminds you to run this when it sees a Dataverse create/update Bash command.