SaaS pricing page change tracker (Stream A · Actor 2)

Spec (10 lines)

1. What: give it a list of public SaaS pricing pages; every run stores a structured snapshot (plans, prices, periods, per-seat/billing hints, feature bullets) and reports exactly what changed since the last run.
2. Buyer: teams and analysts who pay for or track many tools and want to know the week a price, tier or limit moves — without re-reading twenty pricing pages.
3. Input: pricingPages (required), renderJs (default on), onlyChanged, snapshotStoreName, maxFeaturesPerPlan, requestDelaySecs — .actor/input_schema.json.
4. Output: one row per page: vendor, url, pageTitle, fetchedAt, firstSeen, previousFetchedAt, changed, diff{addedPlans, removedPlans, priceChanges[], periodChanges[], featureChanges[], mentionsChanged}, plans[], priceMentions[], hash.
5. Pricing: one PPE event page-tracked at $0.003 ($3 / 1,000 pages), charged after the row is saved; with onlyChanged on, unchanged pages cost nothing.
6. How it parses: structure, not vendor rules — repeated sibling "cards" with a heading and a price, header-labelled comparison tables, Free/Custom marker tiers, hidden monthly/annual toggles kept as altPrices; every numeric price on the page is also listed in priceMentions, so a change is caught even when a layout defeats the plan parser.
7. History: previous snapshots live in a named key-value store in the buyer's account (snapshotStoreName), keyed by URL hash; the actor never sees another buyer's data.
8. Legal floor (from the template): public logged-out pages, robots.txt honoured, ≤1 request/second per host, no cookies, no proxy/fingerprint rotation, no CAPTCHA handling.
9. Tests (28, no network): price parsing, period/billing detection, three golden fixtures (cards+toggle, table, prose), diff logic, snapshot truncation, a smoke test running the real crawler three times through a local server, an opt-in Playwright run, brand consistency against ops/brands.json, and site-owner opt-outs.
10. Non-goals: logged-in pricing, per-customer quotes, currency conversion, judging whether a change is "good" — it reports; interpretation is the reader's.

Input example

{
  "pricingPages": [
    { "url": "https://vendor-a.example/pricing" },
    { "url": "https://vendor-b.example/plans" }
  ],
  "renderJs": true,
  "onlyChanged": false,
  "snapshotStoreName": "ai-tools-watchlist",
  "maxFeaturesPerPlan": 12,
  "requestDelaySecs": 1
}

Output example (one dataset row, second run, price moved)

{
  "url": "https://vendor-a.example/pricing",
  "vendor": "vendor-a.example",
  "pageTitle": "Vendor A — Pricing",
  "fetchedAt": "2026-09-12T06:00:04.120Z",
  "firstSeen": false,
  "previousFetchedAt": "2026-09-11T06:00:03.900Z",
  "changed": true,
  "diff": {
    "changed": true,
    "addedPlans": [],
    "removedPlans": [],
    "priceChanges": [{ "plan": "Plus", "from": "$10", "to": "$11", "fromAmount": 10, "toAmount": 11, "currency": "USD", "deltaPct": 10 }],
    "periodChanges": [],
    "featureChanges": [],
    "mentionsChanged": true
  },
  "plans": [
    { "name": "Starter", "price": { "kind": "free", "amount": 0, "currency": null, "raw": "Free" }, "period": null, "perUnit": null, "billing": null, "altPrices": [], "features": ["Up to 3 notebooks", "Basic search"] },
    { "name": "Plus", "price": { "kind": "numeric", "amount": 11, "currency": "USD", "raw": "$11" }, "period": "month", "perUnit": "user", "billing": "annual", "altPrices": [{ "amount": 12, "currency": "USD", "raw": "$12" }], "features": ["Unlimited notebooks", "Advanced search & filters"] },
    { "name": "Enterprise", "price": { "kind": "custom", "amount": null, "currency": null, "raw": "Custom" }, "period": null, "perUnit": null, "billing": null, "altPrices": [], "features": ["Dedicated success manager"] }
  ],
  "priceMentions": ["$12", "$11", "$5", "$0.50"],
  "hash": "…sha256…",
  "scrapedAt": "2026-09-12T06:00:04.120Z"
}

price.kind is numeric, free (amount 0 or the word Free) or custom (Custom / Contact sales). period is month, year, one-time or null; billing says how a displayed per-month price is actually billed (annual / monthly). RUN_SUMMARY in the default key-value store holds { saved (pages processed), charged, pagesOpened, pagesFailed, stopReason }.

Pricing (pay-per-event)

Event	When charged	Price
page-tracked	after a page's row is written to the dataset	$0.003 ($3 / 1,000 pages)

A weekly watch-list of 60 tools costs about $0.18 per run — or less with onlyChanged, which charges only for pages whose pricing actually moved (and first sightings). PPE events are defined in Apify Console from .actor/pay_per_event.json.

Lawful and responsible use

• Fetches public pricing pages exactly as a logged-out visitor sees them; it never logs in, never keeps cookies, and never works around a block, CAPTCHA or rate limit — a blocked page shows up as a failed request, not as a workaround.
• robots.txt is fetched for every host and obeyed (QuietfetchBot, then *). Pages a site disallows are skipped and reported in the log.
• Rate: at most one request per second per host, two per second overall, one page per vendor per run.
• Data: published list prices and plan descriptions only — factual, non-personal information. No personal data is collected. Buyers are responsible for their own use of the output and for the target sites' terms.
• Identification: the user agent names the bot and a contact address so site owners can reach the operator. Bot token QuietfetchBot; contact page https://quietfetch.com/bots, mailbox hello@quietfetch.com (from ops/brands.json → streams.A). Site owners who want the bot to stop can block QuietfetchBot in robots.txt (honoured on the next request) or email the address — the Builder adds the domain to src/optout.json in every actor, after which that host and its subdomains are never fetched, whatever a buyer's input says (test/optout.test.js).

Local development

npm install
npm test                                  # 20 tests, no network
FOUNDRY_TEST_BROWSER=1 npm test           # + the Playwright path (21)
npm run start:ppe                         # local run with a simulated PPE charging log

Local input: storage/key_value_stores/default/INPUT.json. Snapshots land in storage/key_value_stores/<snapshotStoreName>/.

Maintenance notes for the Sentinel

• The nightly health run fetches one page (health.json) and expects ≥1 row with plans.length ≥ 1. A row with plans: [] but non-empty priceMentions means the layout changed and the card/table heuristics need a look — open a P1, not a P0, unless priceMentions is empty too.
• To add a layout case: save the page HTML into test/fixtures/, write the golden JSON, and extend src/extract.js. Never special-case a vendor by name.

Built and maintained with AI assistance under human review.