What is happening now? 现在发生了什么？
Need quote time, exchange, delay class, session state, currency, and bid/ask context. 需要报价时间、交易所、延迟类别、交易时段、币种及买卖盘语境。
Task-first buyer's guide
按任务选型指南
Best Financial Data APIs for Doubao 适合豆包的最佳金融数据 API
There is no single best feed. Alpha Vantage is approachable for prototypes, Twelve Data is a practical multi-asset candidate, Massive (formerly Polygon.io) is strong for detailed U.S. market retrieval, Financial Modeling Prep fits normalized company research, and FRED plus SEC EDGAR provide authoritative public evidence. 没有一个数据源能在所有任务中胜出。Alpha Vantage 适合快速原型，Twelve Data 值得用于多资产时序，Massive（原 Polygon.io）适合细粒度美国市场数据，Financial Modeling Prep 偏向标准化公司研究，而 FRED 与 SEC EDGAR 可提供权威公共证据。
Choose the question before the vendor. Doubao should receive a compact, typed evidence packet—not a raw stream—and every answer should retain timestamps, units, source identity, and usage constraints. 先确定问题，再选服务商。豆包应接收精简、类型明确的证据包，而不是原始数据流；答案还应保留时间戳、单位、来源身份与使用限制。
Start with the job
从任务出发
Four questions require four different data shapes 四类问题，需要四种不同的数据形态
“Financial data” hides incompatible clocks and semantics. A quote, an adjusted price series, a restated filing, and a revised macro observation cannot share the same freshness rule. “金融数据”背后是不同的时钟与语义。最新报价、复权价格序列、重述财报和修订后的宏观观测值，不能共用同一套时效规则。
How did it change? 它如何变化？
Need interval, timezone, corporate-action adjustment, missing bars, and a fixed date range. 需要周期、时区、公司行动复权、缺失 K 线和固定日期范围。
How strong is the business? 公司经营状况如何？
Need fiscal period, reported currency, units, filing form, accession, and restatement status. 需要财政期间、报告币种、单位、文件类型、申报编号与重述状态。
What does the economy show? 宏观经济说明什么？
Need series ID, observation date, release date, frequency, units, seasonal adjustment, and vintage. 需要序列 ID、观测日、发布日期、频率、单位、季调方式与数据版本。
Evidence-led shortlist
基于证据的候选清单
Best fit by Doubao workflow 按豆包工作流选择最佳匹配
This is a use-case map, not an absolute ranking. Plan limits, licensing, coverage, and regional availability change; verify the exact endpoint and intended downstream use before production. 这是按场景划分的匹配表，并非绝对排名。套餐限制、许可、覆盖和区域可用性都会变化，上线前应核查具体端点与下游用途。
| Source 数据源 | Best-fit task 适合任务 | Why it belongs 入选理由 | Verify before use 使用前核查 |
|---|---|---|---|
| Alpha Vantage | Prototype & broad exploration 原型与广泛探索 | One documented surface spans time series, fundamentals, FX, crypto, commodities, economic indicators, and technical indicators. 一套文档覆盖时序、基本面、外汇、加密资产、大宗商品、经济指标与技术指标。 | Endpoint tier, request limits, delay entitlement, history depth, and commercial rights. 端点套餐、请求限额、延迟授权、历史深度与商业使用权。 |
| Twelve Data | Multi-asset time series 多资产时序 | A unified API and WebSocket-oriented product surface suits workflows that cross equities, FX, and crypto. 统一 API 与 WebSocket 产品面适合跨股票、外汇和加密资产的工作流。 | Symbol mapping, exchange coverage, credit consumption, streaming access, and redistribution. 标的映射、交易所覆盖、积分消耗、流式权限与再分发。 |
| Massive | Detailed U.S. market retrieval 细粒度美国市场检索 | REST, WebSocket, and flat-file access can support trades, quotes, aggregates, and reference-data pipelines. REST、WebSocket 与平面文件可支撑成交、报价、聚合和参考数据管道。 | Feed source, SIP entitlement, delay class, corporate actions, and historical availability. Feed 来源、SIP 授权、延迟类别、公司行动与历史可用性。 |
| Financial Modeling Prep | Normalized company research 标准化公司研究 | Useful candidate for statements, ratios, estimates, profiles, and repeatable company briefs. 适合作为财务报表、比率、预期、公司资料与标准化研究简报的候选源。 | Field definitions, fiscal alignment, currency, restatements, plan access, and source provenance. 字段定义、财政期对齐、币种、重述、套餐权限与来源追溯。 |
| Finnhub | Mixed market + research feed 行情与研究混合数据 | A broad surface can reduce integration count when a brief combines prices, company data, estimates, and news. 当简报同时需要价格、公司数据、预期与新闻时，较宽的产品面可减少集成数量。 | Plan-specific endpoints, geography, news rights, rate limits, and citation metadata. 套餐端点、地域、新闻权利、限流与引用元数据。 |
| FRED + SEC EDGAR | Authoritative public evidence 权威公共证据 | FRED identifies U.S. economic series; EDGAR provides primary company filings and filing metadata. FRED 提供美国经济序列，EDGAR 提供公司原始申报文件与元数据。 | Revision vintage, release calendar, filing amendments, fair-access guidance, and parsing logic. 修订版本、发布日历、文件修正、公平访问规则与解析逻辑。 |
Provider profiles服务商画像
Where each financial data API earns its place每种金融数据 API 为什么值得入选
The table gives the quick answer. These profiles explain the role each source should play inside a Doubao workflow and what must be tested before production.表格用于快速决策。下面说明每个数据源在豆包工作流中适合承担的角色，以及上线前必须测试的内容。
PROTOTYPE / BROAD COVERAGE
Alpha Vantage
A sensible starting point while a team is still discovering user questions. Its documented surface spans stock time series, fundamentals, currencies, commodities, economic indicators, news sentiment, and technical indicators.团队仍在探索用户问题时，这是实用起点。其文档覆盖股票时序、基本面、外汇、大宗商品、经济指标、新闻情绪与技术指标。
- Best for prototypes, educational assistants, and bounded research.适合原型、教育型助手与边界明确的研究。
- Test endpoint entitlement, adjusted versus raw series, history depth, and error payloads.测试端点授权、复权与原始序列、历史深度和错误载荷。
MULTI-ASSET / CONSISTENT SERIES
Twelve Data
Worth testing when Doubao must answer similarly shaped questions across equities, forex, and crypto. A consistent surface can simplify symbol resolution, intervals, and normalization; live state should remain outside the model.如果豆包需要跨股票、外汇与加密资产回答结构相近的问题，Twelve Data 值得测试。一致的产品面有助于统一标的、周期与规范化；实时状态仍应留在模型之外。
- Best for cross-asset watchlists, technical explainers, and portfolio snapshots.适合跨资产观察列表、技术分析解释与投资组合快照。
- Test exchange-qualified symbols, timezones, credit accounting, reconnects, and streaming rights.测试交易所标识、时区、积分计费、重连和流式权限。
U.S. MARKET / GRANULAR RETRIEVAL
Massive
Keep its delivery paths separate: REST for bounded snapshots and history, WebSocket for an application-side live stream, and Flat Files for bulk ingestion. Massive documents that stock Flat Files are unadjusted; adjusted analysis needs REST settings or separate corporate-action processing.其交付方式应保持独立：REST 用于有边界的快照与历史查询，WebSocket 用于应用侧实时流，Flat Files 用于批量入库。Massive 文档说明股票平面文件未经复权，复权分析需使用 REST 设置或另行处理公司行动。
- Best for U.S. trades, quotes, aggregates, reference data, and backtesting.适合美国市场成交、报价、聚合、参考数据与回测。
- Test feed source, timestamps, delay entitlement, splits, and history.测试 Feed 来源、时间戳、延迟授权、拆股与历史数据。
COMPANY RESEARCH / NORMALIZATION
Financial Modeling Prep
Useful when Doubao needs company briefs rather than prices alone. The stable API documents symbol search, quotes, historical prices, profiles, financial statements, estimates, and filing endpoints. Normalized fields help comparison, but filing identity and fiscal context must remain attached.当豆包需要公司简报而不仅是价格时，它很实用。稳定版 API 包含标的搜索、报价、历史价格、公司资料、财务报表、预期与申报端点。标准化字段便于比较，但仍需保留文件身份与财政语境。
- Best for screening, peer comparison, earnings briefs, and ratios.适合筛选、同业比较、财报简报与比率。
- Test fiscal periods, currency, units, amendments, definitions, and nulls.测试财政期间、币种、单位、修正文件、定义与空值。
MIXED RESEARCH / FEWER CONNECTORS
Finnhub
A candidate when one Doubao workflow combines quotes, fundamentals, estimates, calendar events, and news. A broad surface reduces connector count only when results carry enough source, time, and entitlement context.当一个豆包流程需要组合报价、基本面、预期、日历事件与新闻时，它可作为候选。只有结果包含足够的来源、时间与授权语境，较宽的产品面才能真正减少连接器数量。
- Best for monitoring briefs and mixed-source company research.适合监控简报与混合来源的公司研究。
- Test plan endpoints, geography, news attribution, throttling, and partial responses.测试套餐端点、地域、新闻署名、限流与部分响应。
PRIMARY EVIDENCE / PUBLIC DATA
FRED + SEC EDGAR
Use primary sources when authority matters more than convenience. FRED centers identifiable economic series and metadata. SEC EDGAR provides company submissions and XBRL facts as JSON without API-key authentication, plus nightly bulk archives. EDGAR is filing evidence—not a low-latency price feed.当权威性比便利性更重要时，应使用一手来源。FRED 以可识别的经济序列与元数据为核心；SEC EDGAR 无需 API Key 即可提供公司申报和 XBRL JSON，并提供夜间批量归档。EDGAR 是文件证据，不是低延迟行情 Feed。
- Best for macro explanations, filing verification, cited facts, and audit trails.适合宏观解释、申报核验、有引用的事实与审计链。
- Test data vintage, accession, amendments, XBRL units, fair access, and CORS.测试数据版本、申报编号、修正、XBRL 单位、公平访问与 CORS。
The three-station test
三站式评测台
Evaluate outputs, not marketing pages 评估真实输出，而不是营销页面
Identity & time 身份与时间
Run ambiguous tickers, delisted names, extended-hours quotes, holidays, daylight-saving changes, and revised macro releases. Reject silent guessing. 测试歧义代码、退市标的、盘前盘后报价、节假日、夏令时变化和宏观修订值；拒绝静默猜测。
Meaning & arithmetic 语义与计算
Compare raw versus adjusted prices, fiscal versus calendar periods, currencies, scales, nulls, and restatements. Recalculate every ratio outside the model. 比较原始价与复权价、财政期与自然期、币种、数量级、空值和重述；所有比率均在模型外重算。
Failure & recovery 失败与恢复
Force rate limits, timeouts, partial pages, stale caches, disconnects, and entitlement errors. Confirm that Doubao reports uncertainty instead of filling gaps. 主动触发限流、超时、分页不全、缓存过期、断连和授权错误，确认豆包会披露不确定性，而非补写缺失信息。
Model-ready evidence
面向模型的证据
Normalize before Doubao reasons 豆包推理前先做规范化
Keep the full vendor response in logs, but send the model only fields needed to answer. This reduces token waste and makes contradictions visible. 完整供应商响应应保存在日志中，但只向模型发送回答所需字段。这样既减少 Token 浪费，也更容易发现冲突。
{
"instrument": {"symbol":"…","exchange":"…"},
"metric": "close | revenue | cpi",
"value": 0,
"unit": "…", "currency": "…",
"period": {"start":"…","end":"…"},
"observed_at": "…",
"retrieved_at": "…",
"adjustment": "raw | split_dividend",
"source": {"provider":"…","endpoint":"…"},
"status": "current | delayed | revised | partial"
}
Tool selection stays explicit 工具选择必须明确
Expose narrow tools such as quote lookup, historical bars, filing retrieval, and macro series—not one unconstrained “finance” function. 提供报价、历史 K 线、文件检索和宏观序列等窄工具，不要只暴露一个无约束的“金融”函数。
Application code owns controls 控制权属于应用代码
Credentials, validation, retries, caching, deterministic math, and audit logs remain outside Doubao. 凭据、校验、重试、缓存、确定性计算与审计日志均保留在豆包之外。
QVeris can route capabilities QVeris 可路由能力
Use QVeris to discover financial tools, inspect schemas and operational signals, then call a selected capability through a structured interface. 可使用 QVeris 发现金融工具、检查 Schema 与运行信号，再通过结构化接口调用选定能力。
Answers carry provenance 答案携带来源
Require the provider, retrieval time, relevant period, adjustment method, and caveats beside each consequential conclusion. 每项重要结论旁都应保留服务商、获取时间、相关期间、调整方式与限制。
Production gate
生产上线门槛
Ship only when these boundaries are explicit 只有边界明确，才适合上线
Freshness policy 时效策略
Define maximum age per question, distinguish delayed from real-time data, and show the relevant timestamp in the final answer. 为每类问题定义最大数据年龄，区分延迟与实时数据，并在最终答案中显示相关时间戳。
Rights policy 权利策略
Confirm internal analysis, display, caching, derived work, model context, and redistribution separately. API access alone grants none of them automatically. 分别确认内部分析、展示、缓存、衍生内容、模型上下文与再分发权利。获得 API 访问并不自动获得这些权利。
Cost policy 成本策略
Measure the real query mix: symbols per request, pagination, refresh interval, retries, WebSocket fan-out, and stored history. 按真实查询组合测算：每次请求标的数、分页、刷新周期、重试、WebSocket 扇出与历史存储。
Action policy 动作策略
Keep research tools read-only. Separate brokerage or payment credentials, validate every action deterministically, and require approval for consequential transactions. 研究工具保持只读。交易或支付凭据应隔离，对每个动作做确定性校验，并为高影响交易设置审批。
Financial-data output is evidence, not financial
advice.
A fluent model response can still contain selection, timing, unit,
or calculation errors. Use qualified human review for
consequential decisions.
金融数据输出是证据，不是投资建议。流畅的模型回答仍可能出现标的、时间、单位或计算错误；重大决策应由具备资质的人员审核。
Doubao integration path豆包接入路径
From a financial question to a reviewable answer从金融问题到可审核答案
Keep Doubao as planner and synthesizer while deterministic services own identity, retrieval, calculation, and policy. This separation prevents a fluent answer from hiding an ambiguous ticker, stale evidence, or mismatched units.让豆包负责规划与总结，由确定性服务负责身份解析、数据获取、计算与策略。这样可避免流畅回答掩盖代码歧义、过期证据或单位不一致。
Parse intent解析意图
Extract asset, exchange, metric, period, freshness, and comparison.提取资产、交易所、指标、期间、时效与比较目标。
Resolve identity解析身份
Map names and tickers to canonical instruments; ask when ambiguity remains.映射到规范标的；仍有歧义时必须追问。
Route a source路由数据源
Select by task, entitlement, geography, and evidence—not popularity.按任务、授权、地域与证据要求选择，而非热度。
Validate & compute校验与计算
Normalize units and periods; calculate ratios outside the model.统一单位与期间；比率在模型外计算。
Answer with lineage携带来源回答
Return source, observation time, retrieval time, method, and caveats.返回来源、观测时间、获取时间、方法与限制。
QVeris role: its documented Discover → Inspect →
Call pattern helps an application find a financial capability,
inspect parameters, then execute through a structured interface.
See the
QVeris documentation
and
REST API reference.QVeris 的作用：其文档中的 Discover → Inspect →
Call
模式可帮助应用发现金融能力、检查参数，再通过结构化接口执行。可查看
QVeris 文档与
REST API 参考。
Reference stacks参考组合
Three practical API combinations for Doubao三种适合豆包的实用 API 组合
These are architecture examples, not claims that vendors are bundled or formally integrated.以下是架构示例，并不表示这些厂商被捆绑销售或存在官方集成。
STARTUP RESEARCH ASSISTANT
Broad prototype stack广覆盖原型组合
Use Alpha Vantage for bounded price, history, indicator, and company questions; use SEC EDGAR when a material claim needs its underlying filing.使用 Alpha Vantage 回答有边界的价格、历史、指标与公司问题；重要结论则回到 SEC EDGAR 原始文件。
- Cache symbol search and daily history.缓存标的搜索与日线历史。
- Show adjusted/raw status beside returns.收益率旁显示复权/原始状态。
- Link filing claims to accession IDs.文件结论关联申报编号。
MULTI-ASSET MONITOR
Cross-market stack跨市场组合
Use Twelve Data as a candidate for equity, FX, and crypto series. Maintain streams in the application, aggregate snapshots, and send Doubao only the window needed for explanation.可用 Twelve Data 统一股票、外汇与加密资产序列。实时流由应用维护并聚合成快照，只把解释所需窗口发送给豆包。
- Partition caches by market and timezone.按市场和时区划分缓存。
- Detect closed sessions before calling.调用前识别休市状态。
- Never pour tick streams into context.不要把 Tick 流直接塞入上下文。
U.S. EQUITY COPILOT
Market + fundamentals + filings行情 + 基本面 + 文件
Use Massive for detailed U.S. market retrieval, FMP for normalized statements and estimates, and EDGAR for primary verification. Resolve conflicts by meaning—not by averaging values.使用 Massive 获取细粒度美国行情，FMP 获取标准化报表与预期，EDGAR 用于原始核验。出现冲突时按数据语义解决，不能取平均。
- Keep vendor timestamps independently.分别保留各来源时间戳。
- Reconcile fiscal periods before comparison.比较前对齐财政期间。
- Prefer filing facts for sourced disclosure.披露型事实优先采用申报文件。
Transport decision传输方式决策
REST, WebSocket, or bulk files?REST、WebSocket 还是批量文件？
REST
Best default for Doubao tools豆包工具的默认选择
Use for on-demand quotes, historical windows, fundamentals, filings, and macro observations. Calls are bounded and map naturally to structured parameters.适合按需获取报价、历史窗口、基本面、文件与宏观观测。调用边界清晰，也易映射为结构化参数。
WEBSOCKET
Application state, not prompt state应用状态，而非 Prompt 状态
A stream processor should deduplicate, sequence, aggregate, and detect gaps before Doubao receives a snapshot or alert.流处理器应先去重、排序、聚合并检测缺口，再把快照或告警交给豆包。
FLAT FILE / BULK
Backfills and research stores回填与研究库
Use for large historical ranges and reproducible datasets. Record date, adjustment state, schema version, and checksum before deriving features.适合大范围历史与可复现数据集。派生特征前应记录日期、复权状态、Schema 版本与校验值。
Failure modes常见故障
What breaks financial answers first金融答案最先会在哪里出错
Ambiguous ticker代码歧义
A symbol maps to multiple exchanges. Require exchange or canonical ID instead of guessing.一个代码对应多个交易所。必须要求交易所或规范 ID，不要猜测。
False “real time”伪实时
Retrieval time is shown as market time. Preserve observation, publication, ingestion, and retrieval times separately.把获取时间误当市场时间。应分别保存观测、发布、入库与获取时间。
Adjustment mismatch复权不一致
Raw and adjusted prices are mixed. Lock the adjustment method and corporate-action version.混用原始价与复权价。必须锁定复权方法与公司行动版本。
Fiscal-period drift财政期间漂移
Quarterly and trailing values are compared. Preserve period type, dates, currency, units, and filing form.把季度值与滚动值直接比较。应保留期间类型、日期、币种、单位与文件类型。
Silent partial response静默部分响应
Pagination or throttling returns an incomplete universe. Validate coverage and surface partial status.分页或限流导致范围不完整。应校验覆盖并明确返回 partial 状态。
License mismatch许可不匹配
Technical access does not guarantee rights for display, caching, redistribution, or model context.技术访问不代表拥有展示、缓存、再分发或模型上下文使用权。
Procurement checklist采购检查表
Questions to answer before naming a “best” API选出“最佳”API 前必须回答的问题
A provider passes only when its documentation, test responses, and contract agree for the exact Doubao questions you support.只有文档、测试响应与合同对计划支持的豆包问题达成一致，服务商才算通过。
- Which exchanges, assets, identifiers, currencies, and delisted instruments are covered?覆盖哪些交易所、资产、标识符、币种与退市标的？
- Which endpoints are real-time, delayed, end-of-day, revised, or restricted?哪些端点属于实时、延迟、日终、修订值或受限？
- How are splits, dividends, raw prices, and adjusted prices represented?拆股、分红、原始价与复权价如何表示？
- Can values retain provider, endpoint, parameters, period, and retrieval time?数值能否保留服务商、端点、参数、期间与获取时间？
- How are null, unavailable, unauthorized, stale, revised, and partial states distinguished?如何区分空值、不可用、未授权、过期、修订与部分状态？
- What does the real symbol count, refresh rate, pagination, retry, and history mix cost?真实标的数、刷新率、分页、重试与历史组合成本是多少？
- Does the license allow model context, caching, display, derived data, and redistribution?许可是否允许模型上下文、缓存、展示、衍生数据与再分发？
- What fallback or honest failure message is used when the source fails?数据源失败时使用什么备用方案或诚实的失败提示？
FAQ
Questions about financial APIs for Doubao 豆包金融数据 API 常见问题
Can Doubao access real-time stock data? 豆包可以访问实时股票数据吗？
Yes, when an application connects it to a licensed market-data provider through function calling or another controlled tool layer. Doubao itself is not a market-data feed. Confirm exchange entitlement, delay class, and permitted use. 可以，但应用需要通过函数调用或其他受控工具层连接有许可的行情服务商。豆包本身不是市场数据 Feed；仍需确认交易所授权、延迟类别与允许用途。
What is the best free financial data API for Doubao? 最适合豆包的免费金融数据 API 是哪个？
There is no universal winner. Alpha Vantage is an approachable prototype candidate; FRED and SEC EDGAR are authoritative public sources for U.S. macro series and filings. Free access may be rate-limited and may not permit production display or redistribution. 没有通用冠军。Alpha Vantage 是易于上手的原型候选；FRED 与 SEC EDGAR 分别是美国宏观序列与监管文件的权威公共来源。免费访问可能受限流，也未必允许生产展示或再分发。
Should Doubao receive WebSocket ticks directly? 应该把 WebSocket Tick 直接发送给豆包吗？
Usually no. Aggregate, filter, and validate high-frequency events in application code, then give Doubao a bounded snapshot or computed feature set. This is cheaper, easier to audit, and less likely to overload context. 通常不应该。应先在应用代码中聚合、过滤并校验高频事件，再向豆包提供有边界的快照或计算后特征。这样成本更低、更易审计，也不易挤占上下文。
Can Doubao make trading decisions from API data? 豆包可以基于 API 数据做交易决策吗？
It can assist research, but consequential financial decisions need deterministic checks, explicit risk limits, isolated credentials, audit logs, and qualified human review. Do not connect a general research tool directly to order execution. 它可以辅助研究，但重大金融决策需要确定性校验、明确风险限额、隔离凭据、审计日志与专业人员审核。不要把通用研究工具直接连接到下单执行。
Test one real question before choosing a stack 选定技术栈前，先测试一个真实问题
Use a representative, read-only question. Inspect candidate schemas, compare returned timestamps and fields, then validate the answer against the primary source. 选择一个有代表性的只读问题，检查候选 Schema，比较返回的时间戳与字段，再用一手来源核验答案。