# Into: Tag101 The subnet that used to sit on slot 101 with no name and no docs now has both: Tag101 turns X posts into structured semantic tags, scored by network consensus. // Reading the timeline, one tag at a time --- > New to Bittensor? Start here. Experienced users can skip to the full analysis. ### What is Tag101? Tag101 is Bittensor subnet 101. It runs a decentralized tagging network: miners read posts from X (formerly Twitter) and return a short set of tags that capture the key entities, topics, events, and meaning in each post. Validators then score those tags, and the tags that best match the network's shared interpretation earn the most. Slot 101 previously carried only an on-chain symbol (the Georgian letter ე, read "eni") with no published website, repository, or description. The same owner has since claimed a full identity for it. The subnet now ships a documented codebase and a stated purpose, which is what this refresh covers. **The simple version:** It's like a crowd of annotators labeling a firehose of social posts, where the labels everyone independently agrees on are treated as the correct ones. **Centralized equivalent:** Think of a cloud text-tagging API like AWS Comprehend or Google Cloud Natural Language, but instead of one company's model deciding the tags, a competitive network of miners proposes them and consensus picks the winners. **How it works:** - **Miners** read a given X post and submit a small set of tags identifying its entities, topics, events, and context. The reference miner in the repo calls an OpenAI model to generate them. - **Validators** pull posts from a shared database, hand each one out as an independent tagging task, then score every miner's tags on three axes: agreement with the network, relevance to the post, and internal variety. --- ### Why This Matters - **The problem it solves:** Raw social posts are messy and unstructured. Turning them into clean, consistent tags makes a real-time stream searchable and usable as context for downstream AI. Tag101 is trying to produce that structured layer without trusting a single tagging model. - **The opportunity:** Semantic tagging sits upstream of search, trend detection, and retrieval. If the network produces tags that are consistently good across many topics and writing styles, that is a reusable signal, not just a one-off dataset. - **The Bittensor advantage:** A single model has one view of what a post "means." Scoring tags against a decentralized consensus, and rewarding the ones closest to the shared interpretation, is meant to stay more adaptive as topics and phrasing shift. - **Traction signals:** 243 miners are active on the subnet at snapshot time, a clear change from the empty, undocumented slot this used to be. The alpha token was up roughly 37% over 30 days and briefly led all subnets as top gainer on June 20, per a Bittensor daily recap on X. Capital has cooled since: net flow over the past 7 days is about 185 TAO outward. --- ## Full Analysis **Category:** Search and Information Retrieval | **Centralized Competitor:** AWS Comprehend, Google Cloud Natural Language Tag101 registered on 2 May 2025 and spent its early life as an unnamed slot with no public material. The owner has now published a website (tag101.ai), a Discord, and a public repository, and the on-chain identity resolves to that same repo. The pitch is narrow and concrete: convert social posts into structured semantic tags that systems can index, use to track emerging topics, and feed as cleaner context into AI workflows. **Mechanism:** Per the project's repository, validators draw tasks from a centralized database of X posts collected from a publicly disclosed whitelist of accounts. The initial whitelist is focused on AI-related accounts, which the docs call a "topic season," giving the subnet one consistent domain while its scoring matures. The whitelist can be rotated to new seasons over time. Each post is handed to miners as an independent task, and each miner returns a set of tags. Scoring is where the design lives. Each tag gets a TagScore of 0.6 times a consensus score plus 0.4 times validity times diversity. Consensus embeds and clusters every miner's tags for a post, then rewards tags that sit close to the center of the dominant cluster, so a tag is worth more when many miners independently land near the same interpretation. Validity checks that a tag is actually relevant to the post and correctly formatted, mapped into discrete tiers (0, 0.3, 0.6, or 1.0) to keep it stable. Diversity rewards a miner for tags that are distinct from its own other tags, so padding a submission with near-duplicates earns nothing. A miner's task score averages its tag scores, and a duplicate penalty scales that down when many miners submit the exact same tag set, discouraging copied answers. Validators aggregate task scores over time into the weights that drive emissions. On the economics, Tag101's current share of network emissions reads as 0%. Under Bittensor's price-based emission model (the one active since June 2026), a subnet's share scales with root_prop times its EMA price times (1 minus miner burn), then normalized across all emitting subnets. Tag101's miner burn is 0%, so nothing is being withheld there; the 0% share means its weighted price share currently rounds to about zero against larger subnets. Its EMA price sits near 0.00584 TAO, with a spot price around 0.00570 TAO. The AMM pool holds roughly 4,442 TAO, which is modest depth: meaningful positions will move the price and take slippage in both directions. Market cap is around 16,700 TAO. On development, the direct GitHub API was not reachable during this run, so activity is sourced from IntoTAO's own GitHub sync dated 8 July 2026, which records the most recent commit on 2 July 2026 against a single tracked contributor. The repository itself is readable and current: it pins version 0.2.0, builds against the bittensor 10.2.1 SDK, and ships Docker and PM2 deployment paths plus reference miner and validator runtimes. That is a working, recently touched codebase rather than a placeholder, though it is early and thinly staffed. --- ### Risk Factors - **Deregistration:** Tag101 is past its 4-month immunity window and currently draws 0% of network emissions. A subnet with no immunity and the lowest EMA price is the one deregistered when a new subnet registers. On-chain ranking puts Tag101's deregistration risk at low for now (mid-pack among all slots), but a 0% emission share leaves no margin, so this is the risk to watch. - **Centralized data dependency:** Tasks come from a centralized database of whitelisted X accounts, and the reference miner calls the OpenAI API. Both are pragmatic for bootstrapping quality, but they are single points the subnet controls and depends on today, not yet decentralized. - **Execution:** The identity and documentation are recent and the tracked contributor count is one. The mechanism is clearly specified, but there is limited public track record so far, and a thin team is a real dependency. - **Liquidity:** At roughly 4,442 TAO of pool depth, the alpha token is thin. Recent 7-day net flow is negative (about 185 TAO out), so exits face real slippage. - **Concentration:** A gini around 0.78 on the top positions (via IntoTAO's mirror, since the third-party screener was unavailable this run) points to concentrated ownership or stake distribution, where a few large positions can swing pool dynamics. --- Another subnet, unpacked.