Follow a research question across two stories

This journey follows Ada, a character shared by Mystery and Voyage. Focus on one story, work independently, review selected changes, and continue after acceptance. Rust owns the behavior; Python, Node and the CLI expose the same research operations. An agent or application can explain each step using the returned fields.

The executable corpus is analyst-journey-v1. It contains synthetic scan and OCR inputs and an analyst assertion. Supplying OCR text does not execute an OCR engine. The supporting acceptance matrix covers richer lineage, competing claims, ontology changes, failures and retention through the owning native regression tests.

Explore and focus

Discover the Project by its title, then select Mystery with a one-hop traversal. Ada is included because of the FEATURES relationship. Voyage appears as outside context: sharing Ada does not silently add the second story to the selection. Evidence dependencies have their own output and do not become active graph membership.

Question	Native result	Fields to inspect
What did I open?	discovery.arrow, project-metadata.json	title and Project metadata
What is selected?	included.arrow	object_kind, object_uuid
What remains outside?	boundary.arrow	Outside object identity and boundary context
Why is Ada here?	explanations.arrow	reason, root_uuid, predecessor_uuid, via_edge_uuid, depth
What supports the claim?	source.arrow, ocr-artifact.arrow, assertion.arrow, claim-evidence.arrow	Source/Artifact/assertion identities and explicit evidence links
Is this evidence or a judgment?	claim-context.arrow	Explicit classification, status and context; these are separate from raw graph membership

Create a frozen Slice from one exact Version, then create Branch A for Mystery and Branch B for Voyage. Both preserve Ada’s identity, but each has an independent current Version. Local edits in A do not modify B or the Project. The native fixtures read the selected scan/OCR bytes from immutable Versions. The Rust capture also selects an external-only Artifact. external-artifact.arrow and imported-external-artifact.arrow expose its native availability and reference metadata. Reading its historical payload fails with GF_RESULT_NOT_RETAINED before and after import and cleanup: Core does not fetch replacement bytes. This supporting Rust case adds the external limitation to the shared stage sequence; the thin runners exercise the local scan/OCR route.

Analyze, compare and review

In A, set Ada’s score to 1 and ending to detective. The Project independently changes x and y. Inspect upstream-preview.arrow, select only property:x, and commit against that preview. A now has score 1, x 1 and y 0. The unselected y update remains unapplied. branch-analysis.arrow is the actual query result; upstream-receipt.json records the native operation.

comparison.arrow exposes changed object/field units. Submit only property:score and property:ending from A’s frozen Version. proposal-preview.arrow identifies each review item with item_uuid, field and preview_sha256. Accept the score and defer the ending. The Project’s score becomes 1; its ending and private annotation stay unchanged. The private annotation is absent from the submitted field set and review preview.

acceptance-receipt.json and accepted-history.arrow identify the committed operation and exact accepted contributions. decision-history.arrow records research decisions. canonical-choices-after-acceptance.arrow remains empty: acceptance does not implicitly promote a claim to canonical status. Explicit promotion and pre-existing canonical choices have separate owner regressions listed in the acceptance matrix.

Continue and revisit

Continue A to score 2, B to score 73, and the Project to score 99. The continued-live-reference.json resolves A’s new head. The earlier immutable-reference.json identifies the frozen source Version and its content identity. A later resolution may report a different resolved_generation_uuid; that is current resolution authority, not a change to frozen content.

Restore A’s accepted Version. A returns to score 1, while B stays 73 and the Project stays 99. Repeating the original review returns its receipt; reproposing already accepted work does not apply it again. Reusing the operation identity with changed content fails with GF_IDEMPOTENCY_CONFLICT. The journey repeats receipt/conflict and graph/ontology/evidence readbacks after cleanup and reopen; the Proposal restoration owner test also reproposes after cleanup. restoration-outcome.json is a derived assertion summary, not a public response schema.

Share with explicit scope

Restoration creates a fresh Version. export-source-reference.json captures that Version separately from the earlier accepted source. A complete Version export preserves the restored Version’s immutable identity and historical genealogy. complete-verification.json describes the verified package; imported-reference.json identifies the imported Version. Importing historical Branch genealogy does not create a live Branch with that identity.

A score-only export creates a distinct projected Version. selected-reference.json distinguishes its content identity from version.content.source_version, which records provenance. The private field is absent from the imported graph. Disjoint exports and identity/content conflicts are covered by the interchange owner tests.

A Fork creates an independent Project with explicit governance and metadata. fork-reference.json distinguishes the new Project from its origin. Editing the Fork leaves A unchanged. Governance and access-policy metadata describe intent; a hosted application owns authentication and access enforcement.

Run and inspect the evidence

From a source checkout with the pinned toolchain and an admitted durable filesystem, run the native facade journey:

CARGO_TARGET_DIR=/path/to/isolated-target \
  cargo test -p graphforge-api --test research_journey -- --nocapture

To capture a fresh set of results, also set GRAPHFORGE_JOURNEY_CAPTURE_DIR=/path/to/new-output-directory. The manifest lists file sizes, SHA-256 digests, reader qualification and generated identity roles. Arrow files preserve native schemas and values; JSON files serialize native control results except the explicitly labeled assertion summaries. Generated UUIDs and temporary discovery paths differ between runs. They are not stable identifiers for another Project.

The same corpus is executed by:

uv run python crates/graphforge-bindings-py/tests/research_journey.py
node --test crates/graphforge-bindings-node/tests/research-journey.test.mjs
cargo test -p graphforge-cli --test research_journey -- --nocapture

Build native bindings from the same source revision before running the Python and Node commands. The CLI test invokes its same-build binary. It uses a facade setup call for the evidence link because that constructor has no CLI command; the research stages themselves execute through the CLI.

These fixtures establish Core behavior and consumer-visible contracts. They do not establish application rendering quality, observed human comprehension, remote access enforcement, or perpetual compatibility with future readers.