One connection. A real first result.
Add the server, sign in and approve, then verify your workspace connection. No key to copy for an OAuth-capable agent.
Prevent cross-context false matches in an evidence join
Require compatible source context and object type before accepting an identifier match.
Read guidePrevent spreadsheet formulas from executing in research exports
Use a tested export representation for untrusted text without altering the source quotation.
Read guidePreserve commas, quotes and line breaks through CSV exports
Test logical field round trips instead of splitting research text on delimiters manually.
Read guideNormalize display text without merging distinct evidence keys
Keep Unicode comparison rules explicit and preserve source identifiers unchanged.
Read guideDetect encoding problems before a research table is imported
Check decoding and header interpretation without guessing lost characters from context.
Read guideKeep returned source text inert in operational artifacts
Treat apparent markup as evidence text rather than executable instructions or formatting authority.
Read guideLog observed response shapes without inventing a universal schema
Track field paths and types as observations linked to specific authorized receipts.
Read guideStop a transformation when a required field disappears
Distinguish an absent required key from an empty value or legitimate optional omission.
Read guideHandle a newly nullable field without fabricating defaults
Propagate unknown values explicitly when a returned field no longer always contains data.
Read guideQuarantine unfamiliar status values until their meaning is known
Preserve new response states without silently mapping them to success or failure.
Read guideMigrate a local projection while keeping old artifacts readable
Version field mappings and compatibility checks instead of silently renaming the only schema.
Read guideValidate citation pointers after a table transformation
Ensure every retained citation still resolves to the correct permitted evidence row.
Read guideNormalize source links without collapsing distinct evidence objects
Preserve original URLs and justify each comparison-only normalization rule.
Read guideRecord redirects without silently replacing original citations
Keep observed redirect destinations and original evidence references as separate observations.
Read guideKeep excerpt locations traceable after text transformations
Preserve the basis of local offsets when formatting changes an evidence excerpt.
Read guideTrace a derived table back through its permitted inputs
Build a minimal lineage chain for a table that combines several local operations.
Read guideSeparate requested, returned and admitted membership counts
Keep operational counts tied to distinct stages rather than treating one number as total available supply.
Read guidePreserve unavailable values in operational count columns
Prevent missing metrics from becoming false zeroes during table cleanup or aggregation.
Read guideAccount for overlap when combining returned evidence sets
Report union membership and duplicate overlap without adding page or branch counts as independent supply.
Read guideMaintain a category coverage matrix beside a returned table
Show which requested result kinds returned evidence, remained pending or were unavailable.
Read guideKeep retrieval limits separate from available source supply
Label bounded pages and accepted subsets without inventing a population total.
Read guideInventory permitted evidence copies before a retention review
Locate working tables, exports and derivative artifacts without spreading their content into the inventory.
Read guideAttach purpose-specific review deadlines to retained artifacts
Track when an evidence artifact needs policy review without inventing a service retention promise.
Read guideMap a withholding decision to affected derivative artifacts
Use lineage references to identify outputs requiring review without reproducing withdrawn content.
Read guide