What this is

Signal Desk scans public financial news feeds, groups duplicate coverage of the same event into one story, ranks stories by a transparent scoring formula, and optionally adds an AI-generated summary. It is a content prioritization and information tool — not investment advice, not a trading signal, and not a substitute for reading the original reporting.

1. Where the data comes from

Every hour, an automated scan pulls the latest items from these public RSS feeds:

• Reserve Bank of India — official press releases
• SEBI — official press releases
• Economic Times — Markets, Economy, Foreign Trade, ETtech (startup & tech)
• Mint (livemint.com) — Markets, Companies, Industry
• BusinessLine (thehindubusinessline.com) — Markets, Economy, Money & Banking, Forex
• Inc42 — dedicated Indian startup/funding coverage
• Economic Times Wealth & Mint Money — personal finance, taxation, savings
• Economic Times Stocks — stock-specific corporate news
• Mint Economy — macro policy coverage

Two further sources are pulled directly rather than as news coverage:

• NSE corporate announcements — every listed company's filings, straight from the exchange's own feed. This is the primary source that board meetings, dividend record dates, director and auditor changes, results and fund-raising eventually get reported from, so it arrives before the coverage does. Mutual-fund NAV declarations and duplicate newspaper-publication copies are filtered out at ingest; everything else is kept and categorised.
• World Bank Open Data — annual macro indicators for India (GDP, inflation, trade, government debt, employment and more), published under CC BY 4.0, which is what makes it legitimate for us to re-publish and offer for download rather than only display.

Business Standard and Moneycontrol were dropped from the source list — both consistently failed to respond from our server's network (confirmed across multiple scans, not a one-off outage), so rather than leave dead feeds in permanently, we removed them and rely on the sources above for the same categories (IPO, commodities, currency, banking, general economy). The source-health counter on the live feed reflects only feeds actually in rotation, so it should normally read close to full.

2. How duplicate coverage is merged

The same event is often reported by multiple outlets with slightly different headlines. Each headline is compared with candidate clusters using significant actor, action and number tokens. A conservative similarity threshold is used, and conflicting reported numbers receive a strong penalty. This reduces the risk of merging separate earnings, rate decisions or market moves merely because they mention the same institution.

3. How the priority score (0–100) is calculated

Before enrichment, the scan uses a preliminary editorial score so new items can be ordered immediately. After the source-grounded brief is available, the score is recalculated from these transparent signals:

Final score = Freshness×0.28 + Evidence×0.30 + Urgency×0.16 + Impact×0.20 + Extraction confidence×0.06, rounded to a whole number. Enriched policy stories with no demonstrated market relevance are capped at 25.

This score ranks how much editorial attention a story deserves — it is not a market prediction, and it does not indicate whether a stock or index will go up or down.

4. Evidence status and coverage confidence

Primary source present means an official institution or exchange document is linked. Cross-reported means at least two distinct publisher domains cover the event. Single-source reporting means only one publisher is linked. These labels describe the evidence trail; they do not claim that every causal interpretation, quote or number has been independently verified.

5. AI enrichment (Gemini by default; paid OpenAI opt-in)

New or updated stories are sent to Google's Gemini model along with the headline and short excerpts pulled from the linked source articles. Paid OpenAI processing is disabled by default and runs only when an operator explicitly enables it. When enabled, both providers receive the same instructions and produce the same fields — which provider enriched a given story is stored, but doesn't change how it's treated. The model is explicitly instructed to:

• Base every field only on the supplied headline and excerpts — never invent a number, name, or outcome not present in the text
• Leave a field empty rather than guess when the source text doesn't say enough
• Return sentiment, urgency, and impact fields as descriptive labels (e.g. "Positive"/"Negative"/"Neutral"), not as buy/sell recommendations

It extracts: a rewritten, publication-style headline; a plain-English summary; a one-line "why it matters"; the companies/entities/sectors involved; a sentiment reading; and topic tags drawn from a fixed vocabulary (Banking, IPO, RBI, Inflation, Commodities, etc.) used to power the category filters.

Enrichment is rate-limited to stay within free usage tiers, so newly-discovered stories may take up to a few hours to receive it.

6. Plain-English explainers, and which surface shows what

Alongside the short summary, every story gets a longer plain-English rewrite aimed at a commerce, economics or finance student who is smart but hasn't met the topic before. The model is instructed to rewrite the story from scratch in everyday language, define any technical term the first time it appears ("repo rate — the rate at which the RBI lends to banks"), avoid unexplained abbreviations, and finish by stating plainly why it matters — while still adding no fact that isn't in the source text.

The three surfaces deliberately behave differently, because they are used differently:

• Live feed — the original scraped headline (lightly cleaned) with the AI summary underneath. Built for skimming what is happening now.
• News Portal — the AI-rewritten headline, the short digest, and the full plain-English explainer. This is the "understand it" surface, so the jargon-free version leads.
• Browser extension — the publisher's own headline, with the outlet named on every row, and every row linking straight to the original article. The extension's job is to get you to the reporting quickly while you are already looking at a chart; a rewritten headline there would make the destination feel like a different story. One link at the bottom takes you to the explained versions on the portal.

On the website, clicking a story opens it here — a reader page carrying our headline, the plain-English explanation, why it matters, what it touches, and the price reaction — with a "See here as source" button at the top and every contributing outlet listed at the bottom.

We do not republish the publisher's article text, and the reader page says so plainly. Source article text is fetched only to ground the AI while the summary is being written, and is discarded immediately — it is never stored in our database or served to anyone. Everything on the reader page is our own writing about the story, which is why the link to the original is placed at the top rather than buried at the bottom.

7. Designing your own landing page

A signed-in user can shape the desk four ways, all stored against the account so they apply on every device:

• Tell your subjects — pick any number of Finance, Economics, Accounting, Commerce, Management, Investing, Banking & Insurance. Stories tagged with those topics rise to the top.
• Share your portfolio — the tickers you actually hold, with an optional note. News, filings and results touching them are ranked above everything else.
• Share your strategy — a short description of how you invest or what you are researching, in your own words.
• Page design — choose a layout (Standard, Compact, Reader) and drag every section into the order you want, hiding any you do not use.

Ranking weight runs holdings > watchlist > subject match, on the reasoning that what you own affects you more directly than what you merely follow.

Personalisation reorders, it never filters. Anything matching your setup is lifted to the top; everything else stays exactly where it was, further down. A feed that silently hid the rest of the market would be actively misleading about what is happening, so it does not.

On the portfolio: we deliberately collect ticker symbols and a short note only. We do not ask for — and the stored shape cannot carry — quantities, prices paid, or account values. Ranking your news needs to know what you follow, not how much of it you own, and holding the latter would make this a record of your finances for no additional benefit to you. Your strategy text is used only to order your own feed; it is never shown to other users, and Signal Desk does not give investment recommendations.

8. Downloadable datasets and exchange filings

The Data & Filings page carries India-wide macro indicators — GDP and GDP growth, consumer-price inflation, unemployment, exports and imports, current account, FDI, reserves, government debt and tax revenue, lending rates, and sector shares of the economy. Each series can be downloaded as CSV on its own, or the whole catalogue as one wide file with every indicator side by side per year. Downloads carry the source, license and unit in the file header. The same page lists NSE corporate filings, filterable by event type and ticker, each linking to the filing document itself.

The macro series come from the World Bank's open Indicators API and are annual figures that revise a few times a year, so they are refreshed at most once a day rather than hourly. They are not a substitute for the primary release — where the exact vintage matters, cite the World Bank or the original national statistics release directly.

9. What counts as "news about" a company

The company profile page and the browser extension both answer one question: does this story actually name the company, or does it just mention it in passing? A story only counts if the company's name (or a known alias — e.g. "Reliance" for Reliance Industries, "SBI" for State Bank of India) appears in the headline itself, or the story names three or fewer companies overall (a genuine multi-company story, not a roundup where any one name is incidental). This deliberately excludes things like a general Nifty-technicals story that happens to list a bank among several examples, or a rival's news that name-drops a company only as "other banks may follow suit." The goal is that the sentiment shown for a company is built only from stories that are actually about it.

10. Market data

NIFTY 50, NIFTY BANK, SENSEX, USD/INR, Gold, Crude Oil, S&P 500 and the other index/commodity tickers shown around the site are rendered by TradingView's free, official, display-only embedded widgets. Those widgets are drawn directly by your own browser from TradingView's servers — nothing about the price is fetched, cached, or stored by Signal Desk itself. We used to fetch some of these server-side (NSE's own live feed for the indices, Yahoo Finance's unofficial endpoint for everything else), but dropped both: neither NSE nor Yahoo permits automated collection of their data without a licence, so continuing to scrape either one server-side wasn't something we could do at zero legal risk. Policy rates, CRR/SLR, and government securities/T-Bill/call-rate data shown in the sidebar are still scraped directly from the Reserve Bank of India's own site — RBI publishes these specifically for reuse, unlike an exchange's live quote feed.

11. Track record: does the score mean anything?

For every story that resolves to a specific NSE-listed company, we used to capture that stock's price at the moment the story was first scored ("T0"), then again roughly a day later, and compute the % change between the two. The Track Record page groups stories by score band and shows the average realized move for each — an honest test of whether a higher score actually correlates with anything happening in the stock, rather than just asserting it does.

This capture is currently paused. It depended on Yahoo Finance's unofficial endpoint, which we removed along with the rest of this app's Yahoo usage (see section 10) — and there is no free API that grants the same automated-collection rights for NSE equity prices to replace it with. Whatever was captured before the pause stays on the page and is real, not fabricated; the sample simply won't grow further until a licensed market-data feed is in place.

"A day later" meant the next actual trading day, not literally +24 wall-clock hours — a story discovered Friday evening was compared against Monday's price, not Saturday's (when the market is closed and the price hasn't moved at all by definition).

12. Update cadence

News scan and AI enrichment: every hour, automatically. Price-outcome tracking is currently paused (see section 11). Market prices and rate sidebars: every 30 seconds while the page is open (paused when the browser tab isn't visible).

Disclaimer

Signal Desk is an independent, informational editorial tool. Scores, sentiment, and AI-generated summaries reflect an automated reading of public news coverage — they are not verified financial analysis, not investment advice, and not a recommendation to buy, sell, or hold any security. Always confirm material facts against the original source before acting on anything you read here.