Keep a startup-shutdown tracker honest: check every row has a live source with Python

#python #webscraping #tutorial #beginners

A list of Indian startups that shut down in 2026 is only as trustworthy as the links behind it. Two things quietly break such a tracker: pages stop loading, and pages load fine but don't actually mention the company they are cited for. Both happened while assembling one. A YourStory URL about the AI fashion startup Alle returned 403 to curl behind Cloudflare, so the Alle facts had to come from an Entrepreneur India report and Inc42 instead.

Both problems are machine-checkable. This post builds a small checker with only the Python standard library: it reads a CSV of shutdown entries, fetches each source URL with urllib, flags non-200 responses and pages that don't mention the company name, and writes a report CSV.

The tracker CSV

The tracker has four columns: company, month, reason, source_url. Three real rows:

company,month,reason,source_url
Alle,January,wound up operations after struggling to build a sustainable business,https://india.entrepreneur.com/news-and-trends/elevation-capital-backed-ai-fashion-startup-alle-shuts-down/501875
Satvacart,August,shut down after 12 years,https://inc42.com/buzz/online-grocery-startup-satvacart-shuts-down-after-12-year-run/
Zupee,November,on the verge of shutting down operations,https://inc42.com/buzz/why-zupees-ai-pivot-was-a-smokescreen-for-its-shutdown/

The reason column keeps each source's own status words. Entrepreneur India called Alle "the first startup to publicly announce a shutdown in 2026"; it had raised a $3 million seed round in December 2023 from Elevation Capital, Bharat Founders Fund and The Singhal Children. Satvacart, a Gurugram online grocery founded in 2014, shut down after 12 years with 28 August as its final day, having raised over $2 million. Zupee's row says "on the verge" on purpose: Inc42 reported on 12 November that Zupee had laid off its entire workforce and was on the verge of shutting down operations, weeks after announcing an AI pivot. That is not a completed closure, and the column should not say otherwise.

If your reason text contains commas, wrap the field in double quotes — csv.DictReader handles quoted fields for you.

AI-generated illustration.

The script

#!/usr/bin/env python3
"""Check that every row of a shutdown tracker has a live, on-topic source."""
import csv
import sys
import urllib.error
import urllib.request

TIMEOUT = 15  # seconds
HEADERS = {"User-Agent": "tracker-check/1.0"}


def check_source(company: str, url: str) -> str:
    """Return 'OK' or a 'FAIL: ...' reason for one tracker row."""
    request = urllib.request.Request(url, headers=HEADERS)
    try:
        with urllib.request.urlopen(request, timeout=TIMEOUT) as response:
            body = response.read().decode("utf-8", errors="replace")
    except urllib.error.HTTPError as exc:
        # The server answered, but not with 2xx: 403, 404, 500 ...
        return f"FAIL: HTTP {exc.code}"
    except (urllib.error.URLError, OSError) as exc:
        # DNS failure, connection refused, timeout
        reason = getattr(exc, "reason", exc)
        return f"FAIL: {reason}"

    if company.lower() not in body.lower():
        return "FAIL: company name not found in page"
    return "OK"


def main() -> None:
    in_path = sys.argv[1] if len(sys.argv) > 1 else "tracker.csv"
    out_path = sys.argv[2] if len(sys.argv) > 2 else "report.csv"

    with open(in_path, newline="", encoding="utf-8") as f_in, open(
        out_path, "w", newline="", encoding="utf-8"
    ) as f_out:
        reader = csv.DictReader(f_in)
        writer = csv.writer(f_out)
        writer.writerow(["company", "month", "source_url", "result"])

        for row in reader:
            result = check_source(row["company"], row["source_url"])
            writer.writerow(
                [row["company"], row["month"], row["source_url"], result]
            )
            print(f"{row['company']}: {result}")


if __name__ == "__main__":
    main()

What each piece does

The User-Agent. urllib's default is Python-urllib plus the Python version, which some sites block outright. Sending tracker-check/1.0 names the bot honestly and makes server logs debuggable.
The timeout. Fifteen seconds per URL means one hung server can't stall the whole run.
Two exception paths. With urllib, a non-200 response raises HTTPError (the YourStory-style 403 lands here), while DNS failures and timeouts surface as URLError or OSError. Both are recorded per row, so one dead link doesn't stop the batch.
The name check. The body is decoded as UTF-8 with errors="replace" and searched case-insensitively for the company name. A page can return 200 and still be the wrong page — a redirect to a homepage, a deleted article, a tag listing. This catches those.
The report. Every row lands in report.csv with a result column, so failures can be reviewed without rerunning.

Run it

python3 check_tracker.py tracker.csv report.csv

If all three rows pass, you'll see:

Alle: OK
Satvacart: OK
Zupee: OK

Read failures the right way

A FAIL means "go look by hand", not "the row is wrong".

Name variants. The match is a literal substring. A page about The Ayurveda Co may mostly say "T.A.C", and NeuroPixel.AI carries punctuation. Normalize both sides if this bites you.
Bot blocks. A 403 from Cloudflare says nothing about whether the story exists — the YourStory Alle page blocked curl while the shutdown was real and confirmed elsewhere.
Status nuance. The script can't tell "on the verge of shutting down" from "shut down". That's exactly why the reason column stores the source's wording instead of a paraphrase.

Small upgrades

Add a http_status column to the report, check for a distinctive phrase from the reason rather than only the company name, and run the script before every republish of the tracker.

LazyFounder covers startup, funding and AI news for builders, including Indian startup shutdowns, at lazyfounder.in.

Posted from the Mantis team account under the LazyFounder organization. LazyFounder is founded by Tarun Kumar.

DEV Community