The TG Gifts Index (TGGI): what the first 16 days of the market showed

Market context: TGGI 108.87 ▲0.01% · 2026-11-03 · the index

TG Gifts Team · 2026-11-02 · all figures from the public files index.csv, floors-daily.csv and track-record.json

The Telegram gift market went, in two years, from stickers worth a few dollars to collections with floors in the thousands of GRAM. But it had no single number that answers “what happened to the market today”: collections trade on Fragment, Getgems, Portals and Tonnel with different floors, and arguments about whether the market is rising or falling were settled with anecdotes.

The TG Gifts Index (TGGI) is an attempt to reduce that to one number with a published method. It started at 100 points on 2026-10-18. On 2026-11-02, after 16 official trading days, it closed at 108.86 — 8.86% above the base. Over the last seven days: +2.23%; over the last day: +0.18%.

Below: what the index actually counts, what it showed in sixteen days and — more important — what it does not show.

How the index works

The base is public Fragment pages. The radar re-reads the cheapest buy-now listing of every collection roughly every 15 minutes; auctions are excluded. The first reading after 00:00 UTC becomes the daily snapshot and never changes afterwards. The value for a day is published from 01:00 UTC and frozen at 01:05 UTC: a published number is never recomputed, and a found error becomes a dated note in the corrections section (that list is empty today).

A collection enters a day if it has a floor on both the previous day and the current day, and at least 15 recent sales on at least one of them. The weight is proportional to the number of recent sales, capped at 60 (Fragment shows at most 60) and at 10% for a single collection; any excess is spread over the rest.

Honesty is needed here: almost every eligible collection hits the 60-sale cap, so in practice the weights are close to equal — about 1% per collection. The phrase “liquidity-weighted” describes the rule, not a large difference between members.

The formula is chain-linked: TGGI(t) = TGGI(t−1) × Σ w(i) × (Floor(i,t) ÷ Floor(i,t−1)). Collections can enter and leave without causing jumps.

What the sixteen days showed

The 8.86% rise over sixteen days was smooth: 14 of 15 daily changes were positive. The average daily change was +0.57%, the standard deviation 0.53%. The best day was 2026-10-25, +1.59%; the worst and only negative one was 2026-10-29, −0.61%. The market did not fly: it walked up in small steps, and one day of pullback did not change the path.

The second process matters more than the first. The number of collections in the index rose from 86 on the base day to 98 in the last four closes. The index rises not only because floors rise but because the market gets wider: more collections have both a floor and liquidity. Those are different things, and one number mixes them — another reason to look at the constituents, not only the headline.

The third fact is structural. Of the 98 collections in the last close, 96 have 60 or more recent sales, i.e. they hit Fragment's cap. The median floor is 8 GRAM, the minimum 5 GRAM, the maximum 1,444 GRAM. 53 of 98 collections cost less than 10 GRAM. More than half the market is the mass segment under 10 GRAM, and the index reflects that: it is sensitive to cheap collections, not only to blue chips.

One day up close: 2026-11-02

In the last close, 6 collections rose, 6 fell, 86 did not move. A quiet day: more than eighty collections did not shift by a single GRAM.

The moves were isolated:

Source: index.json, constituents array. Before and after are the frozen floors of two consecutive closes.

Note the asymmetry: four rises were steps of 1–2 GRAM on cheap collections, while the Durov's Glasses drop was minus 49 GRAM on a single collection. In percentage terms cheap collections move more on a smaller absolute change; that is why an index where each collection weighs about 1% smooths such moves instead of amplifying them.

What the index does not show

This is the most important part, because this is where wrong conclusions start.

It is not in dollars. Floors are in GRAM (formerly TON). If GRAM moves against the dollar the index does not see it: it measures prices in GRAM, not purchasing power. For a reader who counts in dollars, the index and their portfolio can move differently.

It is about Fragment, not the whole market. On 2026-11-02, 102 collections were listed on both Fragment and Portals. The median floor difference was −14.20% (Portals cheaper), and 47 collections differed by 15% or more. The same gift can cost visibly different amounts on different venues; the index describes one of them — the official one, but not the only one.

It is about asking prices, not trades. A floor is the cheapest price at which something is for sale, not the price someone paid. A collection can sit with an unchanged floor for a week without a single sale. The only liquidity hint in the file is the number of recent sales, and it is capped at 60.

It has a short history. Sixteen days is one market regime, with no seasonality and no full cycle. Any statement about a “trend” here is a statement about two weeks of October and early November 2026. In thirty days the index will have thirty points; in a year, a year. Until then we do not backfill: what did not exist does not exist.

It is almost equal-weighted. The 10% cap does not work in practice: the weights are already about 1%. For an index dominated by large collections, it is not suitable.

How accurate the valuations behind the index are

The index counts floors — observable quantities. Next to it lives a second thing: the estimated value of a specific gift or username. There the numbers are worse, and we publish them.

On the out-of-sample backtest (sales the model never saw; parameters frozen before the holdout) the median error of a gift valuation is 16.7% with n = 999. Within ±20% the estimate lands in 58.0% of cases. For gifts under 10 GRAM the median error is 12.5% (n = 662), for 10–100 GRAM 34.4% (n = 306) with a bias of -25%: the model systematically under-estimates. For 100+ GRAM it is 73% (n = 31, preliminary). The reason is clear: the premium for backdrop, symbol and rarity is not priced by a floor-based model.

For usernames the honest number is 28.2% (n = 170, 95% CI 21.8–42.8%), not the 5.2% that stood in our own public changelog: 5.2% was measured on the window the model was tuned on. Those are different things, and the second number will not be used as a headline again.

The public hunt record is a separate story. As of 2026-11-02 it holds 35 hunts older than 24 hours: 17 sold, 9 still listed, 9 unknown (no public sale price: a transfer or a sale off Fragment). Of the six sold gifts all six went at the asking price or just below, and our estimate was a median 25.3% above the realised price. That is not a confirmation of the model's accuracy but its limit: the buyer paid what was asked.

Data you can check

The index and every underlying series are open under CC BY 4.0, with no key and no sign-up:

• /data/index.csv — 16 official days: date, value, change, number of collections;
• /data/index.json — the same plus the last day's constituents with floors and weights, the method version and corrections;
• /data/floors-daily.csv — 1,613 rows: one floor per collection per day, 102 collections, 2026-10-18 – 2026-11-02;
• /track-record.json — the hunt log with outcomes;
• the repository with daily snapshots: github.com/w3radar/telegram-gift-floors;
• the citable dataset version with a DOI: 10.5281/zenodo.23897620 (Zenodo, CC BY 4.0, version 2026.11.03).

Gaps are not filled. In floors-daily.csv 19 collection-day slots out of 1,632 theoretical ones are missing (98.8% complete) — three collections that had no floor early in the period. We do not distinguish “no listing” from “no scan” and we do not substitute anything: if a day is absent, it is absent.

The only condition of use is attribution — “Data: TG Gifts (tggifts.com)” with a link. The data changes once a day, so a few requests per hour are enough.

What comes next

The index is published every day and the method stays open, corrections included. The first sixteen days prove nothing about the next sixteen: they describe a period in which the market rose in small steps while its coverage widened. That is the only conclusion the data supports — and it is enough to measure a market instead of retelling individual trades.

Index and method: /index
Data and field dictionary: /data
How we value, and how often we are wrong: /how-we-value · /price-truth
Hunt log: /track-record
Cite the dataset: doi.org/10.5281/zenodo.23897620

This material is not investment advice and contains no price forecast. The estimates are statistics on public Fragment data, not a promise of any result.