AdTAO turns "did that ad change actually work?" into a sealed, on-chain prediction market where are paid only for forecasts that hold up against measured reality.
//What is AdTAO
AdTAO ( 21) is a prediction market for advertising outcomes. When a marketer changes a Google Ads campaign, raises a budget, switches a bidding strategy, or pauses a campaign, the real question is what that change actually caused. AdTAO's miners predict the result before it is knowable, and validators later score those predictions against what really happened.
The simple version: It is like a sealed-bid weather forecast for ad campaigns. Everyone writes down their forecast and locks it in a tamper-proof envelope before the outcome exists, then the envelopes are opened and the most accurate forecasters get paid.
Centralized equivalent: The closest comparisons are advertising incrementality and lift-measurement tools such as Google's Conversion Lift, Meta Lift studies, and independent measurement firms like Haus or Measured. Those answer "what did this ad change cause?" inside one company's walls. AdTAO tries to do it as an open, verifiable market.
How it works:
Miners receive a structured of a Google Ads account (a 60-day history plus the specific change being applied) and submit a probabilistic forecast: not a single guess, but a P10/P50/P90 range over how the campaign will move across 7-day and 14-day horizons.
Validators wait until the real outcome is measured, then score each miner's forecast for accuracy and calibration, and write those scores back to the chain.
Why This Matters
//
The problem it solves: Advertising measurement is famously gameable. The platform that sells you ads also grades its own homework on whether they worked. AdTAO separates the forecaster from the scorer and anchors every step on chain, so no one, including the operator, can quietly rewrite the record after the outcome is known.
The opportunity: Calibrated, counterfactual prediction ("what would have happened without this change?") is the hard part of ad measurement. A decentralized market that rewards only accuracy is a genuinely different approach to a problem the big platforms have a conflict of interest in solving honestly.
The Bittensor advantage: The value here is verifiability, and that is exactly what a public chain is good at. Predictions are sealed with timelock encryption before the outcome exists, scores are reproducible from open-source code, and anyone can re-run the verifier against any past to confirm or contradict a validator's claim.
Traction signals: Development is the clearest signal. The repository ships a 412-test suite including a dedicated set of adversarial attack scenarios, and the commit history shows the team publishing weekly prediction episodes and revealed outcomes on a regular cadence, with the most recent commits dated to June 2026. A live validator endpoint and an open testnet environment (netuid 466) are documented for onboarding.
//Full Analysis
Category: Other (advertising impact prediction) | Centralized Competitor: Google Conversion Lift, Meta Lift, Haus, Measured
Advertising spend runs into the hundreds of billions a year, and most of it is graded by the same platforms that sell it. Incrementality, the question of what an ad change actually caused versus what would have happened anyway, is where measurement is weakest and most disputed. AdTAO's bet is that a market which pays for accuracy and proves its scoring on chain can produce more trustworthy answers than a vendor scoring itself.
Mechanism:
According to the subnet's repository, the protocol runs as a sealed prediction cycle. An outcome signer publishes a weekly stream of "episodes," each a structured snapshot of a real Google Ads account: account state, a 60-day pre-window time series, and the specific action being applied (a budget change, a bid-strategy switch, a target adjustment, or a campaign pause). The digest of each release is committed on chain before anyone forecasts.
Miners read an episode and submit P10/P50/P90 distributions per campaign and horizon, rewarding calibrated uncertainty rather than a single point guess. Each submission is encrypted, its decryption key is timelock-encrypted to a future randomness round, and the ciphertext hash, key, and archive location are committed on chain. That ordering is the core guarantee: a forecast is bound to the miner that produced it and locked in before the outcome is knowable, so late or rewritten predictions are detectable from chain state alone.
After the timelock reveals, validators fetch the predictions, run an eight-check scoreability rule, and score each forecast against the measured outcome. The score combines four components per the repo: quantile accuracy (the largest weight, at half the score), interval calibration, directional correctness on the campaign's primary goal, and goal-accuracy via a Brier score. A null penalty cuts near-zero "lazy" predictions, and a skill-score gate means a miner that cannot beat a conditional-prior baseline earns zero. Validators commit their scoring artifacts as on-chain Merkle roots, a separate shadow validator runs the same open-source code on an independent key, and a public verifier script lets anyone reproduce or dispute any epoch's scoring from chain data.
On economics, the repository is explicit that this is a launch-stage configuration. The default validator runs simple score-normalization with the majority of burned to UID 0, and the more elaborate tier mechanics (participation gates, tier placement, pool shares) are implemented and unit-tested but opt-in rather than the runner default until a later review milestone. That matches the on-chain reading, where roughly 77 percent of the subnet's miner emissions are currently burned. The subnet's own codebase is the primary source for these mechanics, and it diverges from some third-party summaries that have not caught up: where they differ, the repo is the authority here.
On market state, AdTAO's alpha trades around 0.00387 TAO with a pool depth near 7,820 TAO. Its smoothed emission share currently reads zero. Under Bittensor's flow-based emission model (live since November 2025, where a subnet's emissions track its net staking inflows minus outflows), a subnet with non-positive receives no emissions, and AdTAO's recent net flow has been negative. Price has drifted down over the past month. These are snapshots at publish time, not a verdict on the technology, which is actively developed.
//Risk Factors
These factors move fast; captured at publishing date
Deregistration: AdTAO's smoothed emission share is currently zero and its recent net staking flow is negative. Its on-chain slot has been registered since mid-2024, so it is well past the four-month immunity window. Bittensor automatically deregisters the non-immune subnet with the lowest EMA price when a new subnet registers, so a sustained zero-emission position is the most direct risk to the slot, independent of how good the technology is.
Adoption and data dependency: The model depends on a steady supply of real Google Ads episodes and on miners and validators showing up to a weekly cycle. The protocol design is unusually rigorous, but the value of a prediction market is bounded by how many capable participants it attracts and how much real campaign data flows through it.
Market: The alpha price has trended down over the past 30 days. Thin and a low emission share can reinforce each other, since emissions under the current model follow net staking flows.
Execution stage: By the repository's own account, the richer reward tiers and parts of the validator programme are still ahead on the roadmap. The launch configuration burns most emissions, so the incentive design that is meant to reward miners at scale is not yet the live default.
Into the next one.
Keep exploring
Other research from the same neighborhood of the network.
