upstream: ask ARA maintainer to add a per-node narrative field to the schema

[EYH0602]

Reminder / action

Ask the ARA format maintainer whether/when a per-node narrative field will be added to trace/exploration_tree.yaml. This is an upstream question, not our runtime work — filing it so we don't forget to raise it.

Why this matters

The old ARA-Labs/ARA-Demo viewer (*/trajectory.html) was a static, upstream-baked HTML page, so it could embed rich per-node prose (a written narrative of each step). Our viewer renders the YAML directly and deterministically and, by design, never calls an LLM at view time (see plans/ara-runtime-impl-plan.md:7).

Consequence: a node narrative can only appear if it's precomputed upstream and stored on the node. Today the schema has no such field.

Current behavior (correct, not a bug)

• No narrative field exists in RawNode (crates/ara-core/src/schema.rs) or in DetailModel.
• The detail pane omits absent fields — no placeholder, no error (graceful degradation, crates/ara-viewer/src/detail.rs).
• So a missing narrative is simply not displayed. Nothing to fix here.

What we need from upstream

1. Confirm whether a canonical per-node narrative field is planned, and its key name + type.
2. Confirm it's optional (viewer must keep degrading to structured-fields-only when absent).

This reinforces the existing format asks in docs/ara-format-feedback.md (esp. items 1 & 4: versioned schema + schema_version, guaranteed keys).

Once upstream answers

• Widen the schema under T-REAL-CORPUS to carry the narrative field.
• Render it in the detail pane + as part of the tree-list/layer-panel work in viewer: add published DOM tree-list as an alternate display mode (match ARA-Demo interaction/display) #7 (currently blocked on this).

Links

• Related: viewer: add published DOM tree-list as an alternate display mode (match ARA-Demo interaction/display) #7 (tree-list / layer panels — blocked on schema widening)
• Plan: plans/ara-runtime-impl-plan.md (narrative-text precondition, lines 7, 199, 215)
• Format feedback: docs/ara-format-feedback.md

[EYH0602]

Decision from the #7 tree-list planning: the tree-list is not blocked on this field.

Issue lists #7 as blocked on schema widening, but that only applies to per-node prose (detail-pane narrative + part-4 layer panels). The tree-list rows label off title ?? body ?? "(untitled)", not a narrative field, so #7 parts 1–3 (display-mode toggle, DOM tree-list, replay stepper) ship without it.

For the missing narrative we keep graceful omission — absent field renders nothing, as it does today. We considered shipping an empty placeholder box and rejected it: it would appear on every node until the schema widens (the demo has no narrative anywhere), reads as broken UI, contradicts the plan's "no dead chrome" stance, and breaks the detail pane's existing omit-when-absent pattern. This issue already confirms graceful omission is correct behaviour, not a bug.

When upstream lands the field (under T-REAL-CORPUS, same schema widening as #7 part 4) the add is strictly additive because None/absent and a future Some(text) differ only in whether prose renders:

1. schema.rs — add narrative: Option<String> to RawNode, #[serde(default)].
2. parse.rs — pass it through.
3. manifest.rs — add to Node, #[serde(default, skip_serializing_if = "Option::is_none")] (mirrors the isolated field being added in viewer: add published DOM tree-list as an alternate display mode (match ARA-Demo interaction/display) #7).
4. detail.rs — add to DetailModel + render in DetailPane: None → nothing (as today), Some(text) → prose.
5. docs/manifest-schema.md — document it.
6. Close/reference this issue in that PR.

Still need from upstream before implementing: exact key name; flat string vs. structured markdown (decides whether DetailPane needs a markdown renderer); and any schema_version guarantee.

Recorded in plans/issue-7-tree-list-mode.md under "Future: per-node narrative field (#12)".