Discovery
Discovery searches for a bypass strategy for the sites named in its field, or for the pages of one set. The result becomes a new set or goes into an existing one. Configured sets keep working during a run.
Unblocking a site walks through a first run, and Keeping sets working shows how the watchdog runs Discovery on its own. How Discovery searches explains a run in detail.

Input
Site or URL takes domains or full URLs, separated by commas or new lines. A domain is fetched as https://<domain>/, a URL as given. Local and private addresses are refused.
Find a strategy for a set runs Discovery for one set instead and fills the field with the set's Discovery addresses, or with suggestions. A run for a set tests at most five addresses. A set that routes its traffic to a proxy or an interface cannot be picked. When another set handles an address in the field, a warning names that set. Find a strategy in a set card's menu, or on the set editor's Discovery tab, opens this page with the set picked.
Note takes an optional line of up to 200 characters on why the search runs, such as the network or the address it is about. The note is shown during the run, on the results and in the history.
Options
The defaults are the usual choice.
| Option | What it does | Default |
|---|---|---|
| Check DNS for tampering | Compares each site's address with a trusted resolver's answer, adds a DNS fix for a forged one and looks for alternative addresses. Turning it off shortens the run when DNS is known to be clean. | On |
| Trusted DNS server | The resolver the DNS check trusts, for this run only. Empty uses the one in Settings. | Empty |
| Pinned addresses | Hosts-file lines, the address and then its names. Those names are tested through these addresses, and the set keeps them as pinned addresses. | Empty |
| Stop at the first strategy that works for every address | Shown while a set is picked. Turned off, the search tests every strategy, which takes longer and can find a faster one. | On |
| Try community sets first | Shown while the Community Hub is on. Tests up to three published sets per site before the built-in strategies. | On |
| Try strategies that worked before | Winners of earlier runs go first. Forget them clears that list. | On |
| Confirm each strategy | Fetches in a row a strategy has to pass during the search, 1 to 5. | 1 |
| TLS version for the probes | Limits the probes, and the created set, to TLS 1.2 or 1.3. | Auto |
| IP version for the probes | The same for IPv4 and IPv6. Shown while the capture engine handles both. IPv6 is unavailable under the TUN engine. | Auto |
| Generated or uploaded ClientHello as the fake payload | TLS payloads from Settings > Payloads that the fake packets carry. Discovery tries them before its built-in payloads. The list offers the payloads whose file is present. | None |
A run
Start begins a run through six steps: DNS, Without bypass, Strategies, Tuning, Combining and Confirming. A run that finds a strategy goes on to a seventh step, Page hosts, see Page hosts. A row per site shows what has been found so far, with the quickest strategy yet as Fastest so far, and the probed addresses. The card at the end can offer another strategy, see Results. Show log opens the log. A run takes from one to ten minutes, depending on the network.

Stop and confirm ends the search and confirms what was found, without tuning it. Stop ends a run that has found nothing. Stop now, during the confirmation, ends the run at once and keeps the results unconfirmed. Skip host check ends only the Page hosts step.
A run stops when b4 rewrites its DSCP rules during it, for example after a set's new DSCP value is saved. Its results cannot be applied, and the history marks the row stopped: firewall rules changed during the search, run it again.
One run is in progress at a time, the searches of the watchdog included. A watchdog search shows on this page but cannot be stopped here.
Under the TUN engine
While b4 runs the TUN engine, Discovery sends its probes through it, and the page shows a note about it. A run under TUN has three limits:
- The probes use IPv4 only. IPv6 under IP version for the probes is unavailable, and a run for a set whose IP Version Filter is IPv6 does not start.
- Strategies that act on the server's replies, Incoming Mode and RST Injection Protection, are not tested. Remembered winners, community sets and built-in strategies that use them are skipped, and the set's current strategy is tested without them.
- A run for a set with its own DSCP value does not start.
The engine that counts is the one b4 is running. A new choice under Settings > Core > Packet Engine counts after the restart that applies it. While the TUN engine is not running or has no uplink address yet, a run does not start.
A run for a set
A run for a set ends with one verdict for the whole set, above Results per address. The set's own strategy is tested first, as Current strategy of the set, and a strategy is confirmed on all addresses together.
For a set with its own DSCP value, the probes carry that value, as the set's traffic does, and take the set's route through the router in front of b4. The progress and the results show DSCP <value>, as the set's traffic. Under the TUN engine, such a run does not start. With the NFQUEUE engine, the probes of any other run carry the value of Set DSCP under Settings > Core > Firewall while it is on, or none, also to the addresses of such a set. Under TUN they carry the same, except that to the addresses of such a set they carry its value.
| Verdict | Meaning and what the card offers |
|---|---|
| one strategy covers every address | Apply to <set> writes the strategy into the set, see Writing into an existing set. |
| current strategy works | The set's strategy loads every address. For sites that still fail in a browser, the card names the addresses another set handles. It also notes that the browser may use QUIC (HTTP/3), which Discovery does not test, or another IP version than the run, such as IPv6. IP version for the probes sets the version a run tests. |
| covers some addresses | Apply to <set> anyway writes the best strategy. Remove them from this set's Discovery addresses drops the uncovered addresses. |
| no bypass needed | Every address loads without b4, so these pages cannot show whether the set works. |
| nothing worked | The cards below show what happened to each address. |
| stopped before a result | Nothing to apply. |
The set's Discovery tab shows the last run for it.
Results
Every site gets one card, under Results per address in a run for a set.

| Badge | Meaning and next step |
|---|---|
| confirmed 3 of 3, not confirmed | A strategy, a DNS fix, an alternative address or removing an ECH key made the site load. The card describes the set it creates and, for a confirmed strategy, lists the page hosts it can add. |
| no bypass needed | The site loads without b4. |
| address blocked | Connections to every known address fail, so no strategy can help. The site needs a VPN or a proxy. |
| intercepted by the gateway | The first hop in front of this host answers the connection itself. b4 has to run on that gateway, or this host has to be excluded from its redirect. When b4 is the router itself, only a proxy helps. |
| does not resolve | The name has no address, as with a misspelled domain. |
| link answers 400 | The link answers HTTP 400 even without b4. A link to a page or a file the site serves is needed. |
| nothing found | No strategy worked. A new run with TLS 1.2 or a generated ClientHello sometimes helps. |
Sites that one strategy loaded share one card and one set. The card offers the strategy that loaded the most sites. Speed does not decide, see Which strategy a card offers. N other strategies worked lists up to twelve other strategies that loaded every site of the card, fastest first, each with Use instead.
When the last run to end was started from this page, reloading the page or opening it again within 30 minutes of the end shows its results again. New search clears them in the browser where it is clicked. The history keeps them, also after b4 restarts.
Probed addresses
Under each site, the card lists the addresses the probes connected to. A check mark follows each address the site loaded from. A search with Check DNS for tampering off lists no addresses.
Next to an address stands the network that announces it, such as AS15169 GOOGLE, when a set's ASN targets already cover the address. Look up ASN asks RIPEstat for the networks of the other addresses.
Page hosts
The Page hosts step reads the page of each site with a confirmed strategy and tests the other hosts the page takes its stylesheets, scripts, images and other files from. A host that fails without b4 and loads with the strategy is listed on the card with a checkbox. The ticked hosts are added to the set by exact name when the result is applied.
| Host marked | Ticked by default |
|---|---|
| stylesheet, script, image, media, CSS or preload, with or without tracker that holds the page until it loads | Yes |
| in page code, frame, icon, preconnect, base address or redirect | No |
part of <service>: a set of its own fits better, whatever its type. A tracker that belongs to a service, such as googletagmanager.com or connect.facebook.net, gets this note too. | No |
| handled by set <set>, whatever its type. Another enabled set already handles the host. | No |
Only the page itself is read. Hosts that its stylesheets and scripts load in turn, or that a script builds while the page runs, are not found. How Discovery searches explains how the hosts are found and tested.
The hosts are tested with the card's strategy only, and Use instead opens the dialog without them. In a run for a set, the hosts are on the cards under Results per address, and Apply to <set> on the verdict writes only the strategy. Watchdog searches skip this step.
Applying
Apply as a set, Apply in the history and Use instead open one dialog. It previews the set and offers:
- Create a new set, then Create set: see Creating a set.
- Write the strategy into an existing set, then Replace strategy: see below. This is the default when the run was for a set or a set lists the site exactly.
- Add to <set>, which already uses the same strategy, then Add to set: adds the sites to that set. When the set's strategy changed after the dialog opened, nothing is added.
Copy JSON copies the set without creating it, in the Import/Export format.
Page hosts in the dialog repeats the card's page hosts with their ticks. When a host marked handled by set <set> is ticked, applying moves it out of that set if the set lists it by its exact name, as with a site. Each of the three choices adds the ticked hosts, even when Add <sites> to this set is off, and Copy JSON includes them.
A result is applied only to a set whose traffic takes the network path the run tested: the DSCP value of the run's probes, written on the same interfaces. A change to the set's DSCP value, to Set DSCP or to its interface list after the run can change that path. The error then names the difference, and a new run for the set tests the new path. For a few seconds after such a change is saved, while b4 applies its DSCP rules, applying and starting a run are refused.
Writing into an existing set
Which set to change lists the run's set and the enabled sets without routing that already match the site. Another set picks any other set without routing. Only the bypass strategy is written: TCP, fragmentation, fake packets and, when the result uses them, a DNS redirect, pinned addresses and removing ECH keys. The port filter, RST protection, UDP, routing and escalation stay as they are. Use these addresses as the set's discovery URLs also replaces the set's Discovery addresses.
When the set does not match the sites, Add <sites> to this set adds them and moves them out of other enabled sets. The dialog warns when another set actually handles a site, since the new strategy has no effect there.
Creating a set
The suggested Set name is the site without its prefix and top-level domain. Match can shorten a single site to a parent domain, so the set also covers its subdomains. A site listed in another enabled set moves into the new one.
The new set goes to the top of the list, enabled. It gets a DNS redirect or pinned addresses only for a site that needed them. The tested addresses become its Discovery addresses, so the set can be searched again or watched.
Alternative addresses
Some blocks hand out an address that refuses every connection, while clients in other regions get addresses that work. With Check DNS for tampering on, Discovery looks for such addresses when every known address of a site fails. The set pins the site's name to up to three addresses that complete a TLS handshake, so devices get those in DNS answers.
The pins are fixed at the time of the run. When the block moves, a new run with the DNS check on finds new ones. Watchdog searches skip the DNS check and do not refresh pins.
Sites with an ECH key
A site that publishes an ECH key is probed the way a browser with the key connects, so the ClientHello names the key's public name, such as cloudflare-ech.com. The set targets that name as well.
When the network drops ClientHellos with ECH and the site loads without it, the card is marked ECH blocked. The result is ECH keys removed: a set that turns on Remove ECH keys from DNS answers and changes no packets. A browser with its own secure DNS still sends ECH, so it needs one of the other strategies listed under the result.
History
The history keeps the latest result of each site, newest first. Under the site, a row shows the run's note and its probed addresses. A later search of the site from this page replaces the note, while a search by the watchdog or over MCP keeps it.

- Apply creates the set the run built, without a new run. +N page hosts under the strategy counts the page hosts the run offered, and Apply offers them again with the default ticks.
- Run again repeats the run for the site, or for the set the row came from, with the same note.
- Remove from history and Clear history remove one row or all of them.
- A row expands to the other strategies its run's card listed, up to twelve, fastest first, each with Use instead.
- Tried marks a strategy that has been applied as a set, also after the set is deleted.
- via watchdog and via MCP mark runs that the watchdog or an AI assistant over MCP started.
Last run log opens the log of the last run, also after a restart, and can download it as a text file.