Fragment Radar: Monitoring Telegram Collectibles

Why you need a separate Radar

Nana Fragment At the same time, new usernames and numbers appear, prices, rates and auction terms change. The marketplace interface is well suited for a separate purchase, but does not answer research questions: what patterns are more common, how the price changes, what appeared per day and what auctions will soon end.

Therefore, a separate analytical layer appeared in GRAM. He regularly reads open pages, normalizes the found lots and stores their changes over time. This is a continuation of the already implemented blockchain-circuit: wallets and payment transactions GRAM are described in this article. Project P2P platformand the release of NFT in Case TON NFT.

Main Frontier: Reading Only

Radar does not log in to Fragment, does not connect a Telegram account, and does not access the wallet. The provider does not have methods for betting, buying or transferring an asset. It performs only public requests that are needed to search and read the lot card.

This separation is important both in terms of security and data quality: analytics should not become a trading bot.

How public data is collected

Fragment uses short-lived hash and cookie sessions. Therefore, the provider first opens the public page, extracts the current parameters, and then refers to the same JSON endpoint search, which uses the web interface. The detailed card is downloaded separately and disassembled by the HTML parser.

1. Discovery

   Search queries find usernames and numbers. Each result gets a stable type, value, slug and public URL.

   public search · normalize

2. Classification

   Individual classifiers calculate features: length, numbers, repetitions, sequences, symmetry and dictionary matches.

   features · tags · categories

3. Refresh

   Background jobs update the price, next bet step, number of bets, status and time of completion of the auction.

   queue · staggered jobs

4. History

   The new snapshot is only created when the state changes. Comparison generates one-time domain events.

   snapshots · events

5. Research

   Adminka combines filters, sorting, asset card, history and decryption of score.

   filters · score breakdown

Primary set: 2 630 assets

At the time of fixing the case, the primary scan collected 2 630 unique assets: 1 500 usernames and 1 130 numbers. Among them, lots are recorded in the states of auction, resale, sale and sold. This set already allows you to check the classification and interface, but is not yet sufficient history for market conclusions.

Continuous mode is controlled by a separate feature flag. While it's off, the scheduler doesn't address Fragment. This allows you to first apply migrations, check the queue and limits, and then turn on regular surveillance without changing the code.

Two Classifications for Different Asset Types

Username and the number cannot be valued by a single formula. For a short word, length, character composition and dictionary coincidence are important. For the number - repeating numbers, sequences, mirror, palindrome and round ending.

Results are stored as separate features, categories and tags. Therefore, the formula can be changed, and the initial observations remain available for recalculation.

Radar Score: Rating that can be explained

Each asset receives score from 0 to 100. For numbers, the rarity of numbers, repetitions, sequences, symmetry and length are weighed. For usernames, length, dictionary match, character composition, number of digits, and noticeable pattern.

The database stores not only the result, but also breakdown: the value of each factor, its weight and contribution. The formula has a version, so after adjusting the weights, you can recalculate the entire set by the team. fragment:score:recalculate and maintain a clear methodology.

Important: Radar Score is not a future price estimate or a financial recommendation. It is a heuristic indicator of structural expressiveness, designed for sorting and primary selection.

Snapshots and events instead of overwriting a line

The current state is needed by the interface, but transitions are more important for analysis. Therefore, the state of the lot is compared with the last observation. If nothing has changed, no duplicate snapshot will be created. If it has changed, a new slice and the corresponding event are saved.

Disposable events are protected from re-creation. This makes future notifications and reports predictable: one state transition corresponds to one event.

Filters and asset card in adminka

The list can be filtered by type, status, price range, minimum score, pattern, first detection time, and near completion. Available sorting by price, score, number of bets, deadline, activity and time of addition.

The asset card shows the reference to Fragment, the price in TON, the approximate equivalent in USD, the next step, the history of snapshots and events, the characteristics of the classifier and the full breakdown Radar Score. The TON/USD recalculation is rarely performed and is used only for indicative display.

Lines, Limits, and Recovery from Errors

Discovery and refresh operate via Redis Queue. Updates are distributed over time so that multiple tasks do not create a spike in requests. A minimum interval is kept between calls, and 429 is turned on when cooldown is answered.

Each request of the provider falls into the technical log: HTTP status, delay, number of elements, rate limit sign or lock and error text. This makes it possible to distinguish the Fragment markup change from network failure and from constraint triggering.

The scheduler supports separate intervals: discovery - usually once in 15 minutes, refresh auctions - every 5 minutes, updating the indicative course TON/USD - every 3 hours. Tasks are protected from overlap and are executed by a single scheduler node.

From data accumulation to analytics

Team. fragment:dataset:report generates a reproducible report: number of unique assets, new and completed lots, sales, range of observed prices, price change, frequency of patterns, average auction duration and distribution of score.

For meaningful comparisons, a time series is needed. The minimum window is seven days, preferably fourteen or more. Prior to the accumulation of this period, the figures in the panel confirm the performance of pipeline, but should not be interpreted as a statistical market model.