zhongkaifu/tensorsharp/market-data

Market data

One script. It asks a JSON endpoint for rows that are already typed and ranked, so nothing has to be parsed out of a page or remembered.

python3 scripts/market_movers.py --movers gainers --count 10
python3 scripts/market_movers.py --movers losers --count 5
python3 scripts/market_movers.py --movers actives --count 10
python3 scripts/market_movers.py --quote AAPL MSFT NVDA

It prints a heading naming the source and the data's own timestamp, then a markdown table. That output is the answer: copy the rows as printed. Do not round them, re-order them, or add a column the table does not have.

Putting it in a deck

--spec writes a spec the documents skill accepts as it stands, so the whole request is two commands:

python3 scripts/market_movers.py --movers gainers --count 10 --spec spec.json --title "Top 10 gainers today"
python3 ../documents/scripts/make_pptx.py --spec spec.json --out gainers.pptx

The spec splits at twelve rows per slide, which is what a table slide holds. Return the .pptx to the user as the downloadable artifact.

What it will not do

It never fills a gap. A row has to carry a symbol, a name, a price, a change, a percentage and a volume, with numbers where numbers belong. A row missing any of them is dropped, so asking for ten can return eight — and eight real rows is the answer, not a reason to invent two. If nothing usable comes back the script exits non-zero and prints why; say that, and do not answer the question from memory. Prices you remember are wrong by definition.

It does not say why a price moved. The response carries prices, not reasons. A number here supports "ROIV is up 18.75% today"; it supports no sentence containing "because", "on news of", or "driven by". If the user asks why, say the data does not carry it.

Read the Currency column. --movers is US shares and prints USD, but --quote takes any symbol the source knows, and those are priced in their own market's unit — a London line comes back in GBp, which is PENCE, so 1,574.80 is £15.75 and not £1,574.80. The unit is a column in the table for that reason. Never compare two rows' prices without it, and never drop it when you quote a row.

Read the heading too. It says when the rows were stamped, and it says "rows stamped between" when they do not share a moment, which they will not across exchanges. If the session is not open it says so — "last market closed price" — and then "today" in a title means the last close, not a live price.

A long `--quote` can run out of time. Each symbol is its own request, and the run keeps a budget inside the tool's timeout so a few slow symbols cannot cost you the whole answer. If it runs out, the table holds the symbols that answered and a note names the ones left out; ask for those in a second run rather than repeating the whole list.

One moment, not a history. These are prices as the source last stamped them. It is not a portfolio, a history, a forecast, or advice, and nothing here should be presented as any of those.

Where the data comes from. Two JSON endpoints Yahoo publishes for its own front end. They are undocumented and unofficial: they can change or start refusing without notice, which is what the non-zero exits are for. Attribute the source in your answer, as the printed heading does. Do not present the figures as a licensed market feed, and do not build anything that depends on them staying available.