Dated project state
Each answer reflects the project facts and timing in effect at that moment.
Method
How records turn into reviewable answers.
Olivine connects current production records, standardized language, deterministic validation, visible gaps, and auditable outputs inside one shared record.
- Structured inputs Records, governing language, and applicability read as explicit inputs from the start.
- Deterministic checks Validation runs against project facts and explicit logic, with gaps left visible when they matter.
- Reviewable outputs Basis, cited values, ambiguity, and human review remain first-class in the final answer.
What goes in
Records and governing language connect end to end.
The method starts with clear source material, not guesswork.
Production state
Budgets, line items, purchase orders, licenses, policies, permits, timecards, call sheets, safety records, and deliverable packages arrive as records.
Standardized language
AICP sections, guild classifications, fringe tiers, insurance minimums, incentive programs, media rights, and deliverable specs sit beside the work they apply to.
Relationships
Entities, dates, dependencies, and applicability links remain explicit so the chain can be inspected when the answer matters.
What every answer shows
Each result carries basis, cited data, and open gaps.
The useful answer includes the conclusion, the supporting facts, and what still needs review.
Relevant entities and values
Line items, policies, licenses, deliverables, programs, officers, and other cited values remain visible.
Missing records stay visible
Unloaded files, incomplete classifications, absent deliverables, and unresolved mappings should remain gaps, not be smoothed away.
Ambiguity stays reviewable
Clarifications, approvals, and trust boundaries remain first-class instead of being hidden behind a single confident summary.
Where it runs
Library, Studio, API routes, and MCP.
Teams should get the same answer whether they browse, integrate, or run agents against the same records and rules.
Library
Browse coverage, examples, business areas, and record walks.
Studio, API, and MCP
Search, checks, and automation run through Studio, HTTP APIs, and MCP servers so internal tools and agents read from the same records and rules.
Workflow tools
Questions, checks, and actions can move into approval and audit flows without losing context.
Known limits
Confidence should match the facts on hand.
The method is strongest when its boundaries are explicit.
Deepest where coverage is strongest
Coverage is strongest today in commercials and operating production workflows.
No hidden update assumption
If a record is old, stale, or missing, the result should show that condition instead of implying silent freshness.
Humans still own the edge cases
Approvals, exceptions, and unresolved ambiguities remain review steps rather than magic automation.