[Product Slice] Make one Copilot answer fully traceable from question to financial fact to citation #101

Problem

Folio already has most of the components required for a trustworthy financial Copilot, but they are still being developed as separate subsystems. A user can ask a financial question, the Agent can call tools, financial data can be normalized, evidence can be persisted, and citation/source-inspector work is underway — yet there is not one explicit product acceptance path proving that these pieces work together as a single user experience.

The risk is that each subsystem passes its own tests while the final answer still loses provenance, time/currency semantics, evidence identity, or citation navigation somewhere between runtime and UI.

Related: #22 #24 #26 #27 #29 #30 #100

Current state

The repository already contains or is actively developing:

• provider-agnostic financial data routing;
• canonical instrument identity;
• FinancialFact timing/currency/unit semantics;
• structured financial evidence persistence;
• typed answer blocks;
• Stream Event work;
• inline citations / Source Inspector work;
• provider conflict/reconciliation work.

What is missing is a single end-to-end Copilot vertical slice that treats these as one product contract.

Expected direction

Establish one canonical Copilot acceptance scenario such as:

“What is Apple’s current valuation / price / key metric, what time is this data from, and where did it come from?”

The exact query may vary, but the path must be:

User question
  ↓
Agent/tool selection
  ↓
Provider gateway
  ↓
FinancialFact / normalized semantics
  ↓
Persisted evidence
  ↓
Answer generation
  ↓
Inline citation / typed block
  ↓
Source Inspector

The user should be able to inspect the answer and determine:

• which provider supplied the value;
• the canonical instrument and metric;
• currency/unit;
• as_of versus retrieved_at;
• live/delayed/EOD/stale state when applicable;
• whether multiple providers conflicted or fallback occurred;
• the evidence identity used by the answer;
• the run/tool identity that produced it.

This Issue should reuse existing contracts. Do not create a parallel Copilot-only evidence model.

Non-goals

• No new orchestration framework.
• No broad UI redesign.
• No requirement to finish all of [Finance Data] Reconcile conflicting values across financial data providers #26 if the selected scenario does not naturally produce a provider conflict.
• No requirement for a large benchmark suite.
• Do not block incremental prerequisite PRs; this Issue is the integration/acceptance target.

Validation / acceptance criteria

• A real Copilot question executes through the production Agent/tool path and real financial provider gateway.
• The final answer includes at least one real structured financial fact with preserved provider, instrument, metric, unit/currency and timing semantics.
• The answer binds that fact to a stable evidence/citation identity rather than only rendering the numeric value as free text.
• The user can open the corresponding source/evidence inspector and see the same value and provenance used by the answer.
• Reloading the conversation preserves the citation/evidence relationship and does not silently recompute it from current data.
• If the selected provider reports delayed/stale/fallback state, that state remains visible to the answer/UI rather than being discarded.
• If conflicting provider candidates are present, the answer does not silently average or hide the conflict.
• Provide a reproducible test report with exact commands, Bun/OS, focused test results, and the real provider used.
• Because this is a visible product flow, attach screenshots showing: (1) the completed answer and citation/typed fact; (2) the opened Source Inspector/evidence detail.

Maintainer note

Success is not “all underlying Issues are closed.” Success is that one ordinary financial question is traceable end to end with the current architecture. Prefer integration and contract fixes over adding new abstractions.