technical-analysis

Technical analysis

Turn stored OHLCV bars into a defensible read of trend, momentum, volume, and volatility. The difficulty is not calling an indicator; it is choosing the window, respecting warm-up, keeping price bases consistent, and not smuggling a forecast into a description.

Read references/data-policy.md first, then the indicator playbook for window selection and the reporting shape.

MCP tools used

• list_indicators - the catalog: each indicator's group, parameters with defaults and ranges, output channels, required bar fields, and defaultLookback
• get_bars - raw OHLCV series for inspection, validation, and anything the indicator catalog does not cover
• get_indicators - computed, date-aligned indicator series plus a latest value per channel
• get_intraday_bars - stored minute bars, for intraday questions
• sync_stock - when history is too short or the daily series stops; a full backfill is proposed, not run unilaterally
• search_symbol - resolve an ambiguous ticker before analysis

Workflow

1. Frame the question as a window. Decide the observation period from the user's horizon before touching tools, and say which window you are using. A 20-day trend question is not answered with a 200-bar series, and a "long-term trend" question is not answered with 30 bars. When the horizon is unstated, use daily bars, state the window you chose, and explain the choice in one line.
2. Confirm the data exists and how fresh it is. Call get_bars for the symbol at the interval you intend to use. Check the last trade_date, the number of rows returned, and whether the series has gaps or nulls. If the series is shorter than the window you need, or clearly stale, stop and route to stock-data-setup - do not compute indicators over an inadequate series, and do not turn a partial series into a confident verdict.
3. Ask the catalog, not your memory. Call list_indicators and choose indicators by group and by the question being asked. Read each one's params (defaults and valid ranges), outputs, and requires from the catalog instead of hardcoding parameter values or channel names.
4. Compute with enough warm-up. Call get_indicators with the indicators you selected, the same interval, and a limit that covers the window you want to show. get_indicators internally fetches limit plus the maximum lookback of the requested indicators, so warm-up is handled for you - but it can only use bars that are actually stored.
5. Read the warnings before the values. If warnings reports that fewer bars were available than the warm-up needed, or that no value could be computed, then every channel that is null across the window was not computed. That is not a neutral reading, not a weak signal, and not evidence for a verdict. Exclude those channels from the trend/momentum/volume/volatility conclusions entirely and list them as not computable. If every requested channel is uncomputed, there is no technical read to give: say so, and offer a full-history sync as the next step instead of a conclusion. A verdict may only be stated from indicators that are fully warmed up over the stated window.
6. Write the read. Follow the reporting shape in the playbook. Separate what the series shows from what you expect.