The Chain of Thought

Chaining works here, but only for chains made of research questions - each answer is sourced, so a chain becomes a cited trail you can audit backwards, which none of the others give you. Spaces are what make it practical: keep a Space for the project and every link inherits the context, so Link 3 does not need Link 1 re-pasted. For a chain whose later links are creative or technical, the honest shape is different - run the research links here, then carry the findings to another tool for the making.

• Chain only the research links; make elsewhere.
• Keep the project in a Space so links share context.
• Paste the prior answer and say "take this as given".
• Academic focus for evidence links, Web for market ones.
• Check late links against early ones - it re-searches each time.

**This is the one chain in the chapter that genuinely belongs here** - a business plan is mostly research, and a plan whose every figure carries a source is worth more than one that reads well. Make a Space called "Gachibowli tiffin service" first so the links share context. Links 1 and 2 here; Link 3 needs a judgement call, so see the note at the end.

---

**LINK 1 - THE FOUNDATION**

I am evaluating a business idea: a home-delivery service for authentic, home-cooked Hyderabadi meals, aimed at busy professionals in Gachibowli.

Before any SWOT, I need the ground facts. Cite everything with dates.

1. Which tiffin services, cloud kitchens and home-chef platforms currently operate in or deliver to Gachibowli, Madhapur and the HITEC City area? Name them. Include the small local operators, not only the funded national ones.
2. What is reported about the economics of this category in India - typical order values, delivery commission rates charged by the aggregators, reported margins, and how many of these ventures have shut down? I want the failure rate as much as the opportunity.
3. What does recent reporting say about demand for home-style and regional food among IT professionals in Hyderabad?
4. What licences does a home-kitchen food business actually need in Telangana - FSSAI registration versus licence, GST thresholds, local municipal requirements? Link the official pages, not a blog summary.
5. Which comparable ventures elsewhere in India have scaled, and what is credited for it?

For each: the source, its date, and whether it is a primary source, a news report or a consultancy figure everyone is quoting secondhand. Where you find nothing, say not found rather than estimating.

---

**LINK 2 - THE FIRST FLOOR**

Take the findings above as given. Now the channels question, still as research.

1. What is documented about how food delivery and tiffin brands in Indian cities actually acquire their first few hundred customers? I want reported tactics with sources, not a list of marketing ideas.
2. For a business targeting office parks and tech campuses specifically: what is reported about corporate tie-ups, campus access rules and bulk lunch arrangements in Hyderabad's IT corridor? Cite it.
3. What do current sources say about realistic costs in India - Instagram and Google ad rates for a local food business, the cost of a WhatsApp Business setup, what the aggregators charge for listing and promotion? Figures with dates; this is the part that goes stale fastest.
4. What is reported about customer acquisition cost and retention in Indian food delivery, and how much of what is published is vendor marketing rather than independent?

Use Web focus. Where only listicles and aggregator blogs exist, say so rather than dressing them up as research.

---

**LINK 3 - THE SECOND FLOOR**

**Do this one elsewhere.** Building the six-month budget is modelling, not research: it needs arithmetic, assumptions you choose, and ideally a spreadsheet you can change. Take the sourced figures from Links 1 and 2 into Gemini or ChatGPT and build it there.

Then come back with the finished budget and ask one question here: which of my assumptions are contradicted by current sources? That closes the chain, and it is the check the modelling tool cannot perform on itself.

**Wrong tool for the chain, right tool for one link.** Perplexity will not invent your detective and will not write your chapter - ask it to and you get a sourced survey of detective fiction conventions. Run Links 1 and 3 in Claude, ChatGPT or Gemini.

What it does uniquely well is the research that makes the invented parts credible, and for this particular novel that research is unusually rich, because the Nizam's jewels are a real collection with a real documented history. Use that as a research link inserted before Link 2.

**RESEARCH LINK - the factual backdrop**

I am writing a detective novel set in Hyderabad involving a missing piece of the Nizam's jewellery. I need the documented record, cited with dates.

1. What is actually known about the Nizam's jewellery collection - what it contains, how it came into government hands, where it is held and displayed now? Primary and scholarly sources where they exist.
2. What has been reported about disputes, legal cases or alleged thefts involving the collection or the Nizam's estate? Cite the cases. Separate what was proven from what was alleged - that distinction is both good practice and good plot material.
3. What is documented about antiquities smuggling networks operating out of Hyderabad or South India? Cite journalism and any court records you can find. I want reported mechanics, not atmosphere.
4. For the police procedural layer: how is the Telangana police actually structured, what would a dismissal or suspension of an officer formally involve, and what would an ex-officer legally need to work as a private investigator in India? Link the official sources.
5. What is documented about the Old City's historic jewellery trade around Laad Bazaar - the trades, the families, how it works?

Cite everything. Where the record is thin or contested, say so clearly - that is where I am free to invent, and knowing the boundary is exactly what I came for.

Use Academic focus for the history and Web focus for the legal and police material.

**Then go elsewhere for the chain itself:** character profile, plot outline, opening chapter. Bring this research along as source material. One question worth bringing back here afterwards: does my plot depend on anything that is factually impossible? A novel can bend the record, but it should bend it on purpose.

**Wrong tool for the chain.** Do not write Python here. It will produce something that looks like code and cite a blog post, and the pseudocode-to-script-to-error-handling sequence - which is the actual lesson of this example - needs a tool you can iterate with. Run all three links in ChatGPT, Claude or Gemini.

What this tool is genuinely good for in a technical chain is the question that precedes it and the one that follows it, both of which are research questions about a fast-moving world.

**BEFORE Link 1 - is this the right approach at all?**

I want to check the status of a list of website URLs from a Python script - connect to each, record which return HTTP 200 and which fail.

1. What is the current documented standard practice for this in Python? Is `requests` still the ordinary choice for a simple synchronous check, and what are the documented alternatives - httpx, aiohttp - and when do the docs say to prefer each? Cite the official documentation, not tutorials.
2. For checking many URLs, what does the documentation say about the cost of doing it one at a time versus concurrently? Link the relevant pages.
3. What are the documented pitfalls of checking status this way - how redirects are followed by default, what counts as a connection error versus an HTTP error status, why a HEAD request is sometimes recommended instead of GET? Cite sources.
4. Is there an existing maintained tool that already does this, so I do not write it at all? Cite the project and when it was last updated.

Cite documentation over tutorials throughout, and give me the date of each page. A tutorial from several years ago describing a deprecated argument is how most of these scripts go wrong, and that is the specific error a search can catch and a model's recollection cannot.

**AFTER Link 3 - checking what you built**

Here is the finished script: [paste it]. Not a review of the logic - do that where you wrote it. Two sourced questions only:

1. Does this use any `requests` method or argument that current documentation marks as deprecated or changed? Cite the doc page.
2. Are there documented security or etiquette considerations for a script that connects to many sites in sequence - rate limiting, user-agent conventions, what robots.txt does and does not cover for status checks? Cite the sources.

That is the honest split: the chain happens elsewhere, and the two ends - what is current, and what is deprecated - happen here, because they are the only parts of the task where having a source beats having an answer.