# TrydingDay full guide for AI agents Canonical entry point: https://trydingday.com/agents/ Short guide: https://trydingday.com/llms.txt Discovery document: https://trydingday.com/.well-known/trydingday.json Agent data sitemap: https://trydingday.com/data/agents/v1/sitemap.xml Contract and schema: https://trydingday.com/data/agents/v1/schema.json ## Purpose and scope TrydingDay publishes public, read-only editorial research about selected U.S. equities. An episode records the original thesis, public evidence, dated updates and observed outcomes. Public coverage is not a personal portfolio, trade execution, a guaranteed fill or individualized investment advice. The public data does not expose private account positions, capital or performance. The machine-readable API consists of direct, standalone JSON resources. A URL ending in `.json` returns JSON as its response body; it is not an HTML page containing embedded JSON and does not require JavaScript or browser navigation. The `/agents/` page is a human-readable directory of these resources. ## Canonical resources - Discovery: `https://trydingday.com/.well-known/trydingday.json` identifies canonical resource URLs and the current API version. - Active index: `https://trydingday.com/data/agents/v1/index.json` lists episodes that remain under editorial tracking and links each summary to its episode record. This index is the authority for active tracking membership; it is not an entry signal. - Episode record: each active or archived episode has its own URL under `https://trydingday.com/data/agents/v1/episodes/`. The episode ID in the index and archive is the stable key for retrieving that record. A record contains the current public projection, including its thesis, technical references when published, public updates, sources, levels and observed outcomes. - Archive index: `https://trydingday.com/data/agents/v1/archive/index.json` lists years with archived records. Fetch only the year files needed from `https://trydingday.com/data/agents/v1/archive/` to discover historical episode summaries. Closed episodes remain available at their episode URLs. - Schema: `https://trydingday.com/data/agents/v1/schema.json` describes the API documents and their versioned fields. - Live snapshot: `https://trydingday.com/data/open-publications.json` provides frequently refreshed quote and publication observations. It is separate from the episode history and is not the source of truth for active membership. - Paper Live snapshot: `https://trydingday.com/data/paper-live-publications.json` is a separate read-only feed with current-session 5-minute VWAP observations for the published episodes it covers. The agent sitemap lists the directory, documentation, discovery document, indexes, schema, snapshots, archive files and episode records. Use it to enumerate the published resources rather than guessing episode URLs. ## Suggested retrieval sequence 1. Read the discovery document and schema to learn the canonical URLs and document contract. 2. Read the active index to identify episodes still under editorial tracking. 3. Fetch only the episode records relevant to the task. 4. Read the Live or Paper Live snapshot separately when current observations are relevant; use each snapshot's own timestamps and status fields. 5. For historical research, read the archive index, select the needed year files and then fetch the corresponding episode records. The active index stays compact. It does not grow by appending every historical episode or every update. An episode remains active until its public tracking closure is recorded; closed and observation episodes are discoverable through the archive. `lifecycle.active_for_tracking` describes editorial tracking status only. It does not mean that a security is currently suitable for entry. ## Reading episode records Use the schema as the field-level contract. The public episode projection contains editorial narrative and evidence, not a broker order. The `levels` object and `technical_reference` reflect levels and technical values present in the public source. The exporter does not invent a missing stop or target, recalculate published levels, or replace them with new values. Treat absent or null values as unavailable, not as zero. Update `timestamp` identifies when the editorial update was recorded. `price_as_of` and `vwap_as_of`, when present, identify the observations used for those values. Machine timestamps are UTC ISO 8601. Prefer `return_vs_entry_pct` with its `price` and `reference_price` for the calculated episode return. The legacy `pct` field is preserved from the source and may refer to a different capture, so it is not a substitute for that calculation. `market_comparison` is a direct projection of the canonical `winner.market_comparison` node. It contains the shared quote baseline captured after publication confirmation and regular-close observations through S30. Each baseline instrument keeps its own provider timestamp, capture time and market phase. A return from that baseline is available only when its provider timestamp is at or after publication and before the compared close. Each session has a `market_day`, which is independent of editorial update time. Session returns use official regular closes, and benchmark excess is calculated only for candidate and benchmark observations from the same U.S. market date. `candidate_comparison_history` is retained only for older publications without the canonical node and is omitted when `market_comparison` is present to avoid duplicating the same session series. ## Reading the Live snapshot The Live snapshot is a point-in-time view that can refresh more often than an episode record. Check its generation and quote timestamps before interpreting the values. `return_vs_today_pct` is the return from the current price to `previous_regular_close`. `return_vs_prior_pct` is a compatibility alias with that same daily meaning in every market phase. `change_vs_previous_capture_pct_points` measures the additive change in cumulative return versus episode entry since the latest recorded episode capture. It is not the daily return. Therefore it can be zero while the daily return is positive when the quote has not moved from the last episode capture. The older `change_vs_prior_session_pct` is a compatibility alias for this capture-to-capture measure. Prefer the explicitly named field in new clients. ## Reading the Paper Live snapshot Paper Live is a separate feed. Its top-level `generated_at` is when that feed was produced; `source_generated_at` identifies the Live source capture when available. Each row carries its own observation timestamps and freshness or capture status. Do not infer freshness from the feed refresh interval alone. The published Paper entry-check limits are `stale_after_seconds: 1800` for quote age and `vwap_stale_after_seconds: 14400` for the latest 5-minute VWAP bar age. The VWAP is session-specific. If a capture has no usable bars, the feed can carry forward the last successful VWAP from the same session and reports the capture status and data age; inspect those fields before using the value. A carried-forward observation is not a newly captured bar. The McpTrading Paper consumer also rejects a Paper snapshot whose `generated_at` is more than 2700 seconds old. That is a consumer-side feed-age threshold, not a field or freshness guarantee supplied by the endpoint. External clients should apply their own maximum feed age. These are data freshness limits in the feed contract, not promises that a quote or VWAP exists for every row at every time. The consumer should reject or treat as degraded observations that are missing, stale, future-dated, or marked with a non-healthy status. The JSON Schema is authoritative for exact field names and nullable values. ## Returns, time and missing data - Machine timestamps use UTC ISO 8601; market-session dates identify U.S. exchange sessions and are not timestamps. - A return is meaningful only with its stated reference price and observation time. Do not combine prices from different captures as if simultaneous. - Daily change, cumulative episode return and capture-to-capture change are distinct measures. Use the named fields above and the schema descriptions. - Missing or `null` means no usable value was published. It does not mean zero. - Editorial updates and market observations have separate timestamps and should remain separate in any downstream history. ## Access and indexing All API resources are public, read-only and support unauthenticated HTTP GET. The JSON endpoints return `application/json`, allow cross-origin GET, and send no-store cache directives so intermediaries should request the current resource. The documentation files `/llms.txt` and `/llms-full.txt` are public plain text. General crawlers are allowed to discover the public agent resources. The robots policy excludes the data paths and guide files only for the named Google advertising crawlers, to keep them out of AdSense processing; it does not restrict other agents or direct public access. Follow the canonical links and sitemap above rather than relying on search-engine indexing for current data. ## Source of truth and limits Published episode records preserve the public editorial record. The active index describes tracking status; the archive describes historical discovery; Live snapshots provide separate market observations. A more recent snapshot does not rewrite the historical update timeline. The schema defines the published contract, and the episode source remains authoritative for what the editorial team recorded. These resources are research data, not a recommendation to enter or exit a position. They do not include account-specific sizing, order routing, fees, slippage or guaranteed execution prices. Any external analysis must make its own risk and execution decisions and preserve the observation timestamps and source distinctions.