Agent review synthesis for Data Horizon strategic whitepaper

Scope

This file records the first-pass local personal-domain review of projects/data-horizon/strategic-whitepaper.md.

The review was intentionally read-only. No source document was changed as part of this evidence pass.

Review lens:

• Whether a human project reviewer, project owner, internal team reader, Controller, or downstream document author can understand what Data Horizon / 数据视界 is.
• Whether the document works as a project-level strategic baseline without becoming a product definition, MVP, field list, API contract, schema, database model, or engineering plan.
• Whether the document preserves Data Horizon as a financial information sensing and governance system, not a trading, strategy, execution, or cognition system.
• Whether the document is ready to be used as an upstream baseline for later product experience, V1 definition, reference evaluation, and controlled rewrite-lite work.

External review verdicts

Reviewer	Route	Verdict	Summary
OOSO	oh-my-opencode run --agent "Momus - Plan Critic" --json	accept-with-minor-rewrite	The document can serve as an initial strategic baseline. Main risks are status over-read, first-stage scope being treated as frozen, risky signal wording, and high first-read load.
Hermes	hermes chat -q -Q	accept-with-conditions	The strategic baseline is acceptable, but downstream work must treat missing alpha/runtime evidence, empty reference shortlist, human-review role, and cost-quality path as hard constraints.

Common findings

Both reviewers independently accepted the whitepaper as a valid first-pass strategic baseline.

Shared strengths:

• The document clearly frames Data Horizon as an independent financial information sensing and governance system.
• It separates sensing, information organization, quality/provenance, and collaboration from financial cognition, strategy, trading, or execution.
• It gives the team a stronger reason for Data Horizon than "we need news sources": AI reduces traditional information gaps but amplifies non-structured information overload, noise, duplication, conflict, source ambiguity, and human attention limits.
• It defines strategic information product forms such as Perception Record, Financial Information Feed, Evidence Package, Quality / Provenance Metadata, and Dataset Package without freezing fields or schemas.
• It keeps first-stage value focused on an independent internal information consumption loop before machine feeds, B-side data services, or external consumer products.
• It preserves FinClaw and AI Trading Matrix as downstream collaborators rather than letting their integration needs define Data Horizon's first-stage value.

Shared concerns:

• The phrase canonical / Current Strategic Baseline can be over-read by readers who miss the caveat that product definition, MVP, runtime evidence, reference evaluation, fields, APIs, schemas, and engineering plans remain unfinished.
• Current alpha facts remain under-mapped: running state, actual sources, sample outputs, users, consumption paths, quality evidence, and failure examples are still not registered clearly enough in the governance repo.
• signal remains a risky word in financial context. Even when the document says it is not a trading signal, names such as Event / Signal Digest may invite downstream misread.
• First-stage information range is strategically plausible but could be mistaken for an approved MVP scope.
• Human review / quality labeling is strategically present but its role is not yet decided: core capability, quality fallback, operating workflow, or later-stage enhancement.
• Cost-quality routing is a first-class strategic constraint, but still needs a later concrete model before engineering readers default to a full AI pipeline.
• Dataset Package is correctly deferred, but its maturity threshold remains vague and could be pulled forward by training-data or B-side pressure.

Ranked risks

Priority	Risk	Synthesis
P0	Alpha/runtime evidence gap	current-state.md still says alpha code, running state, capability range, sources, output examples, users, and consuming systems are not mapped clearly enough. The strategic baseline is acceptable, but downstream product/V1 work should not proceed as if runtime proof already exists.
P1	signal terminology drift	In financial systems, signal can be read as trade signal. Future rewrite-lite should prefer information signal, market state change, or information change unless the text explicitly means a non-execution information cue.
P1	Human review role unresolved	The document correctly includes human review and quality labeling, but later product experience work must decide whether it is a core first-stage workflow, a quality fallback, or a later operating surface.
P1	Cost-quality path needs downstream hardening	Strategy says cost control is first-class, but later V1 work needs an explicit layered path such as rule/lightweight handling, model-assisted enhancement, and human review.
P2	First-stage scope can be mistaken as frozen	Public financial events, crypto, global English finance, Chinese flash news, macro and market state coverage should stay framed as candidate slicing direction until reference evaluation and runtime evidence confirm the scope.
P2	Dataset Package maturity threshold	The whitepaper defers dataset productization, but later documents should define when perception records, provenance, review, authorization, and quality evidence are stable enough for datasets.
P3	FinClaw collaboration shape unresolved	Collaboration should remain strategic until reference evaluation and first-stage scope decide whether Data Horizon supplies information packages, topic packages, source context, retrieval results, or another non-execution form.

Revision-lite recommendations

Recommended lightweight edits before the whitepaper is promoted beyond draft strategic baseline:

1. Add a short status note near the top: "This is a strategic baseline, not product definition, MVP, API contract, schema, database design, or engineering plan; runtime evidence, reference evaluation, and first-stage closure remain pending."
2. Rename or qualify Event / Signal Digest to reduce trade-signal ambiguity.
3. Change first-stage market/source range wording from fixed scope to candidate slicing direction.
4. Add a compact "current usable / still undecided / do not over-read" block for first-time team readers.
5. Keep all changes at strategy level: positioning, boundary, maturity, terminology, and reading path only. Do not add fields, data structures, endpoints, tables, task packages, or V1 feature lists.

Constraints for downstream documents

These constraints should survive product experience, V1 definition, and reference evaluation:

• Sensing outputs must not be packaged as financial cognition conclusions, investment advice, trading instructions, or execution triggers.
• First-stage validation should prioritize internal operators, researchers, project owners, and internal agents before external productization.
• Quality/provenance is a baseline capability, not a late enhancement.
• Capability mapping should be driven by information product value, not by the current engineering module tree.
• Reference evaluation should challenge and calibrate strategic downflow; it should not directly become product definition.
• Cost-quality routing must be explicit before engineering design.

Questions to keep open

• Which first-stage internal consumer is primary: operator, researcher, project owner, or internal agent?
• Which initial information range best reuses current practice assets without promising the full financial information universe?
• Which information product form should anchor first-stage validation: Financial Information Feed, Perception Record, Evidence Package, Quality / Provenance Metadata, or a constrained combination?
• What role should human review play in first-stage operations?
• What evidence is required before Dataset Package becomes more than a second-order candidate?
• How should Data Horizon collaborate with FinClaw without becoming FinClaw's preprocessing or execution-support appendage?

Current recommendation

Accept projects/data-horizon/strategic-whitepaper.md as an initial project-level strategic baseline, but keep it in evidence-backed draft posture until the next constrained pass completes:

1. Runtime evidence补录到 projects/data-horizon/current-practice-profile.md。
2. Reference evaluation shortlist 建立到 commons/references/data-horizon.md。
3. signal 术语、人工复核角色、成本-质量路径三个风险点进入 rewrite-lite 或后续 V1 问题清单。

Do not proceed from this whitepaper directly into product definition, MVP definition, schema/API contract, or engineering implementation plan.