Ethereum's public test networks are launched, used heavily and then deprecated on a recurring basis. Each retirement follows from problems that accumulate in any long-running test environment.
Testnets exist to rehearse mainnet
Developers need somewhere to deploy contracts, and protocol teams need somewhere to run upgrades before they affect real value.
A test network runs the same client software and the same consensus rules, so behavior observed there is expected to transfer.
The essential difference is that its native asset has no market value, which is what makes experimentation and failure affordable.
Free assets create a distribution problem
Test assets must be obtainable at no cost, which means faucets, and faucets are trivially drained by anyone requesting repeatedly from many addresses.
Operators respond with rate limits, social account verification or requirements to hold a balance on mainnet, each of which adds friction for legitimate developers.
Secondary markets sometimes appear for test assets despite them being valueless by design, which undermines the premise and complicates faucet operation further.
State growth makes participation expensive
Every deployment and transaction adds to the network's state, and testnets accumulate a large volume of abandoned contracts and spam.
Syncing then requires substantial disk space and time, which discourages the independent node operators the network needs to stay decentralized.
Validator participation on testnets is voluntary and unrewarded, so as costs rise, operators drop off and the network becomes less reliable.
Drift from mainnet reduces usefulness
Testnets receive upgrades before mainnet, so for periods they run rules that mainnet does not, which is the point.
Over time, differences in configuration, validator composition and traffic patterns accumulate, and a network that no longer resembles mainnet tests less than it appears to.
Launching a replacement resets state, applies current configuration and re-establishes a validator set drawn from active participants.
What deprecation means in practice
Deprecation is announced ahead of time, after which faucets close, infrastructure providers withdraw support and block explorers eventually stop indexing.
Contracts deployed there are not portable, since addresses and state do not transfer, so teams must redeploy and update every reference in their tooling and documentation.
Local development networks avoid the whole cycle for most work, and are generally preferred until a project genuinely needs a shared public environment with other participants.
Forked local networks copy mainnet state into a private instance, which gives contracts real balances and real counterparties to interact with while remaining disposable and free to reset.