How LowHunt works

1. What LowHunt does

LowHunt is a shopping-intelligence layer that puts today's product price into historical context. A price in isolation is weak evidence: a retailer can show a large promotional badge even when the item was cheaper for weeks, while an ordinary-looking price can actually be the best observed point in months. LowHunt therefore starts from observations, quality rules and reproducible comparisons rather than promotional copy. The practical question is simple: how interesting is this price compared with the evidence we actually have?

Monitoring is marketplace-specific. The same product can have different prices, taxes, availability, currency and history in Spain, Germany, Canada or the United States. Signals therefore belong to a product-marketplace pair, not to a global product in the abstract. The public experience selects the marketplace relevant to the request instead of mixing countries into one feed.

LowHunt never invents an offer end time and does not turn missing coverage into certainty. When history supports a conclusion, the interface states a verifiable window such as a 90-day, six-month or twelve-month observed low. When the evidence is weaker, wording is deliberately more conservative.

2. From observation to a publishable opportunity

The pipeline begins with price observations and product attributes. Every observation carries a timestamp and belongs to a marketplace. Those points update rolling price state: recent lows, previous price, current price, movement direction and historical references when they are available. The data layer is kept separate from the presentation layer so a card cannot manufacture a claim that the underlying state does not support.

A low number is not automatically a public signal. Products must pass commercial and sanity gates: valid price, active signal, enabled marketplace slot, sufficient reviews and rating when required, fresh observations, minimum LOWHUNT Score and LOWHUNT Rating, and other publication rules. Correctness is intentionally more important than artificial deal volume.

Affiliate economics are isolated from eligibility. An affiliate link cannot raise Score, improve ranking or make an ineligible product public. That separation is central to explainability: the opportunity should remain defensible even if the affiliate parameter were removed.

3. How price history is interpreted

Historical analysis is not just a search for the smallest number in a database. The strength of a statement depends on the observation window. A 30-day low is not the same claim as a twelve-month low. LowHunt therefore exposes the strongest supported window rather than presenting a short sample as an all-time fact.

Current price and reference price are also separate concepts. A reference is a valid historical point used to contextualize the current offer. When the current price is below it, LowHunt can calculate absolute and percentage savings. When it is not, the system does not force a positive discount. A product may still be at a recent low while showing little or no meaningful saving against another reference.

Charts represent available LowHunt observations. They are evidence views, not guarantees of continuous market coverage. Observed averages refer to the recorded points in the shown window; they are not manufacturer list prices or an official Amazon average.

4. Understanding the LOWHUNT Score

LOWHUNT Score is a proprietary 0–100 summary of multiple dimensions. The current model combines Price Value, Discount Quality, Product Quality and Offer Quality when evidence for that dimension exists. Missing dimensions are not fabricated; weights are renormalized, and caps or direction penalties prevent incomplete evidence from appearing stronger than it is.

The color scale is absolute. Low values start in green, move through blue and violet, then fuchsia, and the highest values reach red. A score bar reveals only the portion from zero to the product's actual score. It must not compress the complete rainbow into every card. This makes the endpoint color itself meaningful.

LOWHUNT Rating is a simpler companion scale. It does not replace the underlying facts. Score, class, price history and reasons are shown together so users can understand why an opportunity is being surfaced rather than being asked to trust an opaque recommendation.

5. Local market and language

LowHunt does not treat offers from different countries as interchangeable. A Canadian price and a Spanish price can differ in currency, tax, catalog and deliverability. The public catalog is therefore filtered to the marketplace relevant to the request. When infrastructure provides a trustworthy country header, LowHunt uses it; otherwise the locale in the route is the deterministic fallback.

Localization serves both readability and market coherence. Users can still choose another interface language, but the system avoids using language choice as a reason to mix foreign offers into the current catalog. Exact delivery eligibility for a street address remains an Amazon-side fact and is not claimed unless it can be verified.

This approach also keeps historical comparisons valid. A reference from one marketplace must never be used to justify the price of another marketplace.

6. How deal cards are built

Cards prioritize fast decision evidence: product identity, current price, meaningful discount when present, absolute saving, strongest supported historical claim and LOWHUNT Score. Marketplace, category, detection time and rating provide secondary context. The layout is designed to use space for information rather than decorative emptiness, while preserving mobile legibility.

Product images are resolved only from allow-listed Amazon media hosts. If safe media cannot be resolved, LowHunt does not substitute an invented product photo. Telegram has a deliberate no-image fallback so media failure does not stop QA, and the event is recorded for diagnosis.

Amazon links use the approved affiliate tag where configured. Affiliate disclosure is visible and the tag does not affect ranking. Raster pixels inside a Telegram image are not independently clickable; real navigation is provided by the message link and native controls, while web CTAs are explicit sponsored product links.

7. Freshness, lifecycle and idempotency

Opportunities have a lifecycle. Active signals feed the public catalog and publication candidates; ended signals are not presented as current deals. Publication freshness rules ensure that a stale observation cannot continue circulating simply because it once looked attractive.

External publishing adds idempotency. Every opportunity has a stable identity based on product-marketplace, score version and price. The production canary is limited to one external attempt. If a previous production publication is pending or failed and therefore uncertain, another one-shot live run is blocked until reconciliation.

The same evidence discipline applies to releases. A commit alone is not a production fact. CI, exact-SHA deployment and post-deploy health checks are required before runtime is considered updated.

8. What a LowHunt signal does not mean

A signal is not a guarantee of stock, delivery, offer duration or personal suitability. Prices and seller conditions can change after observation. LowHunt provides factual context for a shopping decision; the final merchant page remains the place to confirm seller, variant, taxes, shipping and final price.

Not every percentage deserves visual emphasis. Tiny rounded discounts are suppressed to avoid misleading 0% badges or exaggerated savings. A recent-low claim can be useful even when the saving against one reference is small, so historical context and discount are shown as separate facts.

A high Score also does not mean a product is right for everyone. Product Quality uses evidence such as rating and review volume, but it cannot know an individual's needs. The Score describes the observed opportunity, not personal utility.

9. Reading a deal in under thirty seconds

Start with current price and the historical claim. Read the window literally: a 90-day observed low is a claim about the supported 90-day window, not the entire lifetime of the product. Then check reference and saving. They explain how the current price compares with a tracked historical point and are not an official MSRP.

Next read LOWHUNT Score, class, marketplace and detection time. On the detail page, use the chart to see whether the price move is persistent, cyclical or unusual within the available period. Finally open Amazon and verify seller, delivery, variant and final checkout price.

This order is designed to reduce impulsive decisions based on a single promotional badge. A useful LowHunt deal should be explainable with visible evidence.

10. Data quality and continuous improvement

Coverage varies by product and market. Some items have longer histories and more frequent observations than others. LowHunt preserves that uncertainty instead of hiding it. Missing evidence can reduce a dimension, trigger a score cap or weaken the wording of a historical claim.

Regression checks protect against inconsistent current prices, broken images, malformed affiliate links, zero-percent badges and cross-market mixing. Material releases are validated with automated tests, canonical previews and runtime checks. Visual changes must not reduce factual accuracy.

As real usage evidence accumulates, Score calibration, prioritization and cadence can improve. Those changes should be driven by observed outcomes rather than intuition. The product principle remains stable: surface useful, trustworthy buying opportunities through explainable Deal Intelligence.

11. Why LowHunt does not mix countries or currencies

A price only makes sense inside a purchasing context. Amazon Spain, Amazon Germany, Amazon France, Amazon Italy, Amazon US and Amazon Canada can have different sellers, taxes, catalog availability, Prime eligibility, currency and delivery rules for what appears to be the same product. Combining those records into one feed would create false comparisons and could lead a shopper to click an offer that is irrelevant to the country from which they are browsing. LowHunt therefore treats product plus marketplace as the commercial identity used for monitoring, scoring and publication.

At request time, the site uses country context supplied by trusted infrastructure when it is available. If the infrastructure does not expose a country, the localized route becomes the deterministic fallback. This keeps the experience predictable and prevents a generic global feed from silently mixing Canadian dollars, US dollars and euros. A user can still choose an interface language, but language preference is kept separate from the factual price history that belongs to a specific marketplace.

Marketplace selection is not a promise of address-level deliverability. Amazon can restrict delivery by seller, postcode, island, territory, product class or stock location. LowHunt cannot truthfully infer those final checkout conditions from a country alone. The site therefore narrows discovery to the relevant Amazon market and clearly leaves exact delivery confirmation to the merchant page. This is intentionally more conservative than showing every international listing that might sometimes ship abroad.

The same rule protects the historical analysis. A French historical low must not be used as the reference for a German current price, even when both pages use euros. Taxes, sellers, promotions and availability can differ. Keeping market histories isolated makes every low, saving and Score explanation reproducible within the context that generated it.

12. How LowHunt grows deal coverage without lowering standards

A trustworthy deal engine can look smaller than a coupon directory because it refuses to turn every merchant promotion into a signal. LowHunt grows coverage by monitoring more products and discovering more high-quality candidates, not by weakening the definition of an opportunity. The autonomous data engine maintains a product universe per supported marketplace, refreshes monitored items according to priority tiers and uses available provider budget to bootstrap additional candidates over time.

Discovery is intentionally upstream from publication. A newly discovered product does not appear on the website simply because it was returned by a provider. It first needs trustworthy price history, current commercial evidence and enough product-quality context. The signal engine then evaluates whether the current price represents a new historical opportunity under the configured windows. Only after that does the publication policy apply freshness, Score, Rating, review and sanity gates.

This separation lets LowHunt increase visible deal volume safely. More discovery expands the number of products that can potentially produce signals; more frequent refresh improves the chance of observing a genuine low while it is still current; neither action changes the factual threshold used to call something a low. When the catalog is sparse in one market, the correct response is therefore to increase monitored coverage and refresh capacity before considering any change to the definition of a signal.

Coverage is also market-specific. A country with fewer monitored products may initially expose fewer public opportunities than another country. LowHunt should communicate that honestly rather than filling the page with foreign deals. Over time the monitoring universe, category balance and refresh cadence can be adjusted using measured provider cost, observed signal yield, click usefulness and commercial quality. Those changes remain auditable because discovery, signal detection, scoring and publication are separate stages.

For shoppers, the practical result is a feed designed to become broader without becoming noisier. The goal is not the largest possible count on any given day; it is a sufficiently broad local catalog in which every visible opportunity can still answer the same questions: what is the current price, what historical evidence supports the claim, how fresh is the observation, what marketplace does it belong to, and why did it pass LowHunt's publication rules?

Frequently asked questions