Practical habits that keep learning evenings calm:

1. Download once per symbol/range into data/ (parquet or CSV).
2. Load from disk on the next run; only refresh when you change the window.
3. Cap concurrency. One ticker at a time beats a 200-name loop that looks like a scrape.
4. Treat adjusted closes, dividends, and splits as “good enough for study,” not audit-grade.

Do not build live order logic on this layer. Unofficial chart sources are not fills. If an AI-written notebook claims perfect overnight equity from Yahoo bars alone, read AI-written backtests and look-ahead bias before you trust the chart.

Layer 2 — Research: keyed REST

When you need the same history tomorrow, pay for (or register) a vendor that issues API keys, documents rate limits, and owns the redistribution license. Massive (formerly Polygon.io) is one common choice for US equities and related markets; others exist. The point is the contract: a key in an environment variable, retries you control, and logs you can read.

Keep keys out of git. Prefer os.environ["MASSIVE_API_KEY"] (or your vendor’s name) over hard-coded strings. Pin the client library version so a silent SDK default change does not rewrite your base URL mid-project.

Polygon → Massive migration checklist

On 30 October 2025 at 4 PM ET, Polygon.io renamed to Massive.com (official announcement). Product behaviour stayed the same that day; branding and domains moved.

Use this checklist; do not invent a big-bang cutover:

1. Read the post. Keys, logins, REST, WebSockets, and flat files kept working; no forced code change on day one.
2. Note the parallel bases. api.polygon.io continued for an extended period; api.massive.com is the new base. Plan a controlled switch, not a surprise outage.
3. Website redirects. polygon.io redirects to massive.com; update bookmarks and docs links you own.
4. SDK path. Official guidance: polygon clients keep routing to the old base; new massive SDKs default to api.massive.com. Both were maintained in parallel; upgrades may need import tweaks only if you choose to switch packages.
5. Billing labels. Invoices and card statements may show Massive.com / MASSIVE.COM — useful when finance asks why the merchant name changed.
6. Email. @polygon.io addresses forward to @massive.com; prefer the new domain in new tickets (support@massive.com per the post).
7. Deprecation watch. Old endpoints were slated for later phase-out with notice — track vendor mail; do not assume forever on the legacy host.

Re-run one golden script after migration: same symbol and range, check row counts and a close checksum.

Layer 3 — Macro and filings: FRED and EDGAR

Price bars alone do not teach macro or disclosure hygiene. Two free official stacks cover that gap without a paid market-data bill.

FRED (St. Louis Fed). The FRED® API requires an API key. You request keys from a fredaccount.stlouisfed.org login; Version 1 passes api_key as a query parameter (32-character lowercase alphanumeric). See API Keys, FRED API Version 1. Version 2 uses a Bearer token in the Authorization header. One key per application; do not share a single key across unrelated tools.

SEC EDGAR. Filings and JSON submissions APIs are free; there is no SEC API key. Fair-access rules matter: declare a descriptive User-Agent (company/app name + contact email), and stay within the published 10 requests per second ceiling (Accessing EDGAR Data). Missing or generic agents often yield HTTP 403. Cache submissions locally; do not crawl the archive like a search engine.

For context only: QuantConnect’s docs change log added Fama–French Factors docs on 12 August 2026 and European Central Bank datasets (Euro Yield Curve, Policy Rates, Systemic Stress) on 25 August 2026 (change log). The after-work stack is broader than one equity CSV; you do not need any platform tonight.

Stooq as failover — without unofficial wrappers

When Yahoo throttles, a simple failover is Stooq historical CSV for many equities and FX symbols: fetch the CSV yourself, parse with pandas, store beside your Yahoo cache. Align columns and calendars explicitly; do not assume identical corporate-action handling.

Do not reach for unofficial “yfinance-compatible” mega-wrappers that hop Yahoo → Stooq → SEC → crypto behind one import. Learning needs visible primary and backup paths.

User-Agent and civility

Official HTTP surfaces care who you are:

• EDGAR: required identifying User-Agent.
• FRED / Massive / others: follow each vendor’s docs; keys already identify you on keyed APIs.
• Any public CSV host: moderate rate, cache, and identify your script if the host asks.

A polite agent string and a local cache beat rotating proxies. Fair-access blocks waste more evenings than typing a header once.

What this stack is not

• Not a live SIP feed and not a substitute for your broker’s market data entitlement.
• Not proof that a strategy works — only that your download path is honest.
• Not a reason to enroll in unpublished quant or macro courses. If you want structured Python after work, start with published lessons.

Next evening

Install or confirm Python, then write one script that (1) loads a cached daily series, (2) pulls one FRED series with your own key, or (3) fetches one EDGAR submissions JSON with a proper User-Agent. Stop there. The enrollable path for language foundations is Introduction to Python Programming — short lessons for evenings, not a trading-signal product.

Disclaimer: Educational content only. Not investment, trading, or legal advice. Data vendors change terms; re-read primary docs before you ship anything that touches money.