feat(launcher): measure host reachability behind a developer flag by Furroxide · Pull Request #68 · Furroxide/TopiaForge

Furroxide · 2026-08-22T18:45:53Z

Behavior

Adds a manually triggered developer reachability diagnostic and TFMP014 validation for declared client-to-server bandwidth exceeding the per-connection budget. This PR targets current dev independently; the superseded local-world stack and historical merge were removed during rebase.

The probe defaults off, requires developer mode, saved opt-in, and an explicit Run probe action, and has no built-in server list. It classifies STUN observations without retaining observed addresses in domain results or logs. Configured STUN servers and their advertised alternate endpoints see the request source IP address and port; the pane discloses exposure before opt-in. No aggregate result is uploaded, reportingApproved remains false, and P0-PRIV-01 remains a release blocker. No server destination or privacy approval is introduced.

The UDP transport keeps one subscription for the socket lifetime. Successful requests, timeouts, and late responses cannot close or complete another transaction; close drains pending work and overlapping requests are refused. This repairs the observed failure where only the first request reached a real loopback responder.

Only replies whose actual address and port match the requested source changes count as evidence. Missing or unusable OTHER-ADDRESS support stops discovery after the initial binding; filtering remains unknown. A completed supported probe may still report restrictive filtering when no reply arrives.

Current launch previews, profile persistence, request-correlated session outcomes, and subscription cleanup are preserved through the helper extraction. Existing review fixes for address-family binding, settings feedback, and TFMP014 wording are retained.

Verification

Five socket regressions failed before the repair and pass afterward, using an owned loopback responder without public STUN traffic.
Consent-disclosure regression failed before the UI change and passes afterward; opt-in and disabled-reporting gates remain covered.
Rebased source: 891 domain tests and 525 data tests pass; four existing Windows symlink skips. The earlier Windows multiplayer-test divergence did not recur.
Launcher Flutter: 84 tests pass; shared UI: three tests pass. Both analyzers are clean. Dart formatting is clean and all 490 non-generated tracked Dart files are at most 500 lines.
Fresh Windows Release solution build, rebuilt eleven-SDK/seven-harness verification, and Flutter Windows debug build pass. README, residue, trademark, asset-license and line-count audits pass. Website source checks pass: 33 tests, 26 guides, 128 Markdown files and 521 JSON/YAML files.
Protocol follow-up: 13 response-origin matrix cases and nine unsupported-discovery cases failed before their fixes. All 56 focused tests pass. The loopback responder handles observed nonblocking write backpressure without weakening assertions; a final full data run passes with SDK outputs stable. Earlier SDK rewrite and fixture-backpressure failures are preserved in local logs.
Exact final head 9e0c7ef3f0f956cef232f7ef16d3d3a0e9c1584b passed CI 34177635608 and CodeQL 34177633789. All 31 reported checks and all five required gates passed, including Windows/Linux data tests, all seven templates, and complete Linux documentation publication. GitHub reports the branch clean against dev.

No public server measurement, game execution, transport feasibility approval, or release acceptance is claimed.

Copilot

🟡 Changes recommended

There are correctness issues in the new reachability probe implementation and wiring (notably STUN attribute decoding control flow and IPv6/socket behavior) plus a misleading analyzer release note entry.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Adds two pieces of “pre-gate” multiplayer work across the codebase: (1) a new generator warning that checks declared client→server bandwidth against a recorded design budget, and (2) an opt-in, developer-mode-only launcher NAT reachability probe with structurally address-free results and reporting held closed.

Changes:

Add TFMP014 warning in the multiplayer contract generator for declared inbound bandwidth budget overflow, with generator tests and docs.
Add reachability probe domain/data implementation (STUN binding + RFC 5780 behavior discovery), persistence + policy gates, and a Dev-only Flutter pane to run it locally.
Expand documentation around privacy/capabilities and the multiplayer feasibility gate with the new partial-evidence records (including internal decision records).

File summaries

File	Description
tests/TopiaForge.Mods.Multiplayer.Generators.Tests/Program.cs	Adds tests for the new TFMP014 inbound budget diagnostic and ensures it doesn’t affect lock identity.
src/TopiaForge.Mods.Multiplayer.Generators/MultiplayerContractGenerator.cs	Implements per-connection inbound design budget constant + TFMP014 warning logic.
src/TopiaForge.Mods.Multiplayer.Generators/AnalyzerReleases.Unshipped.md	Documents the newly added analyzer rule.
packages/launcher_domain/test/reachability_probe_test.dart	Unit tests for reachability classification, policy gates, and report shape.
packages/launcher_domain/lib/src/reachability_probe.dart	Domain types for NAT observation/classification (address-free by construction).
packages/launcher_domain/lib/src/reachability_probe_policy.dart	Opt-in settings, refusal reasons, reporting gate constant, and gateway API.
packages/launcher_domain/lib/launcher_domain.dart	Exports the new reachability probe domain APIs.
packages/launcher_data/test/reachability_probe_test.dart	Data-layer tests for STUN codec, server parsing, runner sequencing, and service persistence/gating.
packages/launcher_data/lib/src/reachability/stun_transport.dart	UDP STUN transport abstraction + implementation for the probe.
packages/launcher_data/lib/src/reachability/stun_message.dart	Minimal STUN codec (binding req/resp) used by the probe.
packages/launcher_data/lib/src/reachability/reachability_probe_runner.dart	Runs the STUN transaction sequence and reduces it to boolean evidence.
packages/launcher_data/lib/src/reachability_probe_service.dart	Persists opt-in settings, enforces policy, and runs one local probe.
packages/launcher_data/lib/launcher_data.dart	Exports reachability probe data-layer APIs/services.
docs/PrivacyAndCapabilities.md	Records the decision around IP exposure and capability disclosure for future transport work.
docs/MultiplayerHostingFeasibility.md	Updates gate notes with where partial evidence lives (including TFMP014 and probe).
docs/Multiplayer.md	Adds documented bandwidth budget + explains what TFMP014 does/doesn’t cover.
docs/internal/MultiplayerTransportOptions.md	Internal decision record covering vendor analysis, budgets, topology, and spend controls.
docs/internal/LauncherReachabilityProbe.md	Internal design/approval dependency record for the launcher probe.
apps/topiaforge_launcher_flutter/test/widget_test.dart	Wires new reachability widget test cases into the suite.
apps/topiaforge_launcher_flutter/test/widget_reachability_test_cases.dart	Widget tests for Dev-only reachability pane behavior and opt-in gates.
apps/topiaforge_launcher_flutter/lib/src/screens/developer_screen.dart	Adds the reachability pane to the Dev screen layout.
apps/topiaforge_launcher_flutter/lib/src/screens/developer_reachability_pane.dart	New Dev-only UI pane to enable/run the probe and show results locally.
apps/topiaforge_launcher_flutter/lib/src/screens.dart	Registers the new Dev reachability pane part.
apps/topiaforge_launcher_flutter/lib/src/launcher_state.dart	Adds reachability probe settings and last-run result to state.
apps/topiaforge_launcher_flutter/lib/src/launcher_reachability_actions.dart	Adds bloc handlers for probe settings + running the probe.
apps/topiaforge_launcher_flutter/lib/src/launcher_event.dart	Adds reachability-related launcher events.
apps/topiaforge_launcher_flutter/lib/src/launcher_event_dispatch.dart	Wires reachability events into the central dispatch switch.
apps/topiaforge_launcher_flutter/lib/src/launcher_bloc.dart	Adds probe gateway dependency + loads settings during initial load.
apps/topiaforge_launcher_flutter/lib/src/launcher_bloc_helpers.dart	Extracts shared `_guard` + snapshot projection helpers from `launcher_bloc.dart`.
apps/topiaforge_launcher_flutter/lib/src/launcher_app.dart	Plumbs `ReachabilityProbeGateway?` into the bloc from the app root.
apps/topiaforge_launcher_flutter/lib/main.dart	Instantiates `ReachabilityProbeService` and reads `TOPIAFORGE_REACHABILITY_SERVERS`.

Review details

Files reviewed: 31/31 changed files
Comments generated: 4
Review effort level: Lite

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

`UdpStunTransport.bind` always bound `anyIPv4`, but `StunServerList` parses bracketed IPv6 literals and hands them straight to `send`. A configured IPv6 server therefore reached an unconnected IPv4 socket, where `send` throws rather than reporting a miss; the service's `SocketException` catch turned that into "Could not open a UDP socket", which is not what happened. Binding for a family exposed the deeper version of the same problem. A run uses one unconnected socket for every transaction and decides mapping by comparing reflexive endpoints across a server's address and port. Endpoints in different families are not comparable, so a mixed server list produces a mapping verdict that means nothing — the IPv4-only socket was accidentally hiding that by making every IPv6 entry fail. The family is therefore a property of the run. `StunTransportFactory` takes it, `bind` selects `anyIPv4` or `anyIPv6` and keeps only same-family interface addresses for `matchesLocalEndpoint`, and the service derives it from the configured servers and refuses a list that mixes families instead of quietly probing one of them. `anyIPv6` is not dual-stack everywhere — Windows defaults `IPV6_V6ONLY` on and Dart cannot clear it on a datagram socket — so choosing beats assuming. `request` also drops a wrong-family destination rather than letting `send` throw, since the transport's contract is that a transaction reports evidence and never raises. `reachability_probe_test.dart` reached 560 lines with the new cases, past the 500-line cap CI enforces, so the service group moves to a part file. launcher_data 375/375. Reported by Copilot on #68.

…llowed `_guard` emits its success message only while `state.isBusy` is still set, so a callback that clears `isBusy` itself has to supply its own message. Every other handler in the bloc does one or the other. The reachability settings-changed handler did neither: it emitted `isBusy: false` with no `statusMessage`, so saving the opt-in reported nothing at all and 'Reachability probe settings saved.' was dead text. It now leaves `isBusy` alone and lets `_guard` emit. The probe-run handler is unchanged; it already supplies its own message alongside `isBusy: false`, which is the other half of the same contract. The existing opt-in widget test now asserts the message reaches the status bar, and fails on the previous code. Flutter launcher 59/59. Reported by Copilot on #68.

The Notes column in this file describes each rule's invariant, not its emit condition — TFMP008 reads "Generated command codecs fit the declared payload limit" and fires when they do not. TFMP014's row followed that convention but read as a bare statement of fact, which is easy to take for the condition that produces the warning. "must fit" keeps the convention and removes the ambiguity, matching TFMP001's "must be non-generic top-level partial classes". Raised by Copilot on #68 as an inverted note; the table's convention is what made it read that way.

`UdpStunTransport.bind` always bound `anyIPv4`, but `StunServerList` parses bracketed IPv6 literals and hands them straight to `send`. A configured IPv6 server therefore reached an unconnected IPv4 socket, where `send` throws rather than reporting a miss; the service's `SocketException` catch turned that into "Could not open a UDP socket", which is not what happened. Binding for a family exposed the deeper version of the same problem. A run uses one unconnected socket for every transaction and decides mapping by comparing reflexive endpoints across a server's address and port. Endpoints in different families are not comparable, so a mixed server list produces a mapping verdict that means nothing — the IPv4-only socket was accidentally hiding that by making every IPv6 entry fail. The family is therefore a property of the run. `StunTransportFactory` takes it, `bind` selects `anyIPv4` or `anyIPv6` and keeps only same-family interface addresses for `matchesLocalEndpoint`, and the service derives it from the configured servers and refuses a list that mixes families instead of quietly probing one of them. `anyIPv6` is not dual-stack everywhere — Windows defaults `IPV6_V6ONLY` on and Dart cannot clear it on a datagram socket — so choosing beats assuming. `request` also drops a wrong-family destination rather than letting `send` throw, since the transport's contract is that a transaction reports evidence and never raises. `reachability_probe_test.dart` reached 560 lines with the new cases, past the 500-line cap CI enforces, so the service group moves to a part file. launcher_data 375/375. Reported by Copilot on #68.

…llowed `_guard` emits its success message only while `state.isBusy` is still set, so a callback that clears `isBusy` itself has to supply its own message. Every other handler in the bloc does one or the other. The reachability settings-changed handler did neither: it emitted `isBusy: false` with no `statusMessage`, so saving the opt-in reported nothing at all and 'Reachability probe settings saved.' was dead text. It now leaves `isBusy` alone and lets `_guard` emit. The probe-run handler is unchanged; it already supplies its own message alongside `isBusy: false`, which is the other half of the same contract. The existing opt-in widget test now asserts the message reaches the status bar, and fails on the previous code. Flutter launcher 59/59. Reported by Copilot on #68.

The Notes column in this file describes each rule's invariant, not its emit condition — TFMP008 reads "Generated command codecs fit the declared payload limit" and fires when they do not. TFMP014's row followed that convention but read as a bare statement of fact, which is easy to take for the condition that produces the warning. "must fit" keeps the convention and removes the ambiguity, matching TFMP001's "must be non-generic top-level partial classes". Raised by Copilot on #68 as an inverted note; the table's convention is what made it read that way.

`UdpStunTransport.bind` always bound `anyIPv4`, but `StunServerList` parses bracketed IPv6 literals and hands them straight to `send`. A configured IPv6 server therefore reached an unconnected IPv4 socket, where `send` throws rather than reporting a miss; the service's `SocketException` catch turned that into "Could not open a UDP socket", which is not what happened. Binding for a family exposed the deeper version of the same problem. A run uses one unconnected socket for every transaction and decides mapping by comparing reflexive endpoints across a server's address and port. Endpoints in different families are not comparable, so a mixed server list produces a mapping verdict that means nothing — the IPv4-only socket was accidentally hiding that by making every IPv6 entry fail. The family is therefore a property of the run. `StunTransportFactory` takes it, `bind` selects `anyIPv4` or `anyIPv6` and keeps only same-family interface addresses for `matchesLocalEndpoint`, and the service derives it from the configured servers and refuses a list that mixes families instead of quietly probing one of them. `anyIPv6` is not dual-stack everywhere — Windows defaults `IPV6_V6ONLY` on and Dart cannot clear it on a datagram socket — so choosing beats assuming. `request` also drops a wrong-family destination rather than letting `send` throw, since the transport's contract is that a transaction reports evidence and never raises. `reachability_probe_test.dart` reached 560 lines with the new cases, past the 500-line cap CI enforces, so the service group moves to a part file. launcher_data 375/375. Reported by Copilot on #68.

…llowed `_guard` emits its success message only while `state.isBusy` is still set, so a callback that clears `isBusy` itself has to supply its own message. Every other handler in the bloc does one or the other. The reachability settings-changed handler did neither: it emitted `isBusy: false` with no `statusMessage`, so saving the opt-in reported nothing at all and 'Reachability probe settings saved.' was dead text. It now leaves `isBusy` alone and lets `_guard` emit. The probe-run handler is unchanged; it already supplies its own message alongside `isBusy: false`, which is the other half of the same contract. The existing opt-in widget test now asserts the message reaches the status bar, and fails on the previous code. Flutter launcher 59/59. Reported by Copilot on #68.

The Notes column in this file describes each rule's invariant, not its emit condition — TFMP008 reads "Generated command codecs fit the declared payload limit" and fires when they do not. TFMP014's row followed that convention but read as a bare statement of fact, which is easy to take for the condition that produces the warning. "must fit" keeps the convention and removes the ambiguity, matching TFMP001's "must be non-generic top-level partial classes". Raised by Copilot on #68 as an inverted note; the table's convention is what made it read that way.

Furroxide · 2026-08-28T20:15:49Z

Not ready to merge, but worth keeping — this needs retargeting rather than closing. The feature itself looks sound; the branch it sits on does not.

`CLEAN` here is not evidence

This PR targets feat/roboworld-local-import, not dev. That branch's own PR, #67, is CLOSED with mergedAt: null — its work landed separately as #84. So mergeStateStatus: CLEAN describes a merge into an orphaned branch and says nothing about dev.

Against dev it does not merge: git merge-tree exits 1 with 5 conflicts, all from the two superseded commits.

There is effectively no CI evidence

	check runs
this PR's head `7067cb8`	1 (`Required / Registry validation`)
`dev` head	19

The workflows never fired because the base branch does not match their trigger. In particular CodeQL has never run on these 3,201 lines, which include a hand-rolled STUN binary parser and a raw UDP socket. The codeql-high-critical ruleset has no bypass actors, so if that analysis surfaces a high alert, the PR becomes unmergeable by anyone — better to learn that now than after a rebase.

Three commits are already superseded

a5ab363 and b935fff are #67's work, now on dev via #84. abbcf21's docs/CustomWorlds.md link fix is on dev verbatim. These should be dropped, not merged — they are the entire source of the 5 conflicts.

The four commits that are actually yours: a2bd6fa, d9d9f90, e50cc91, 7067cb8.

Suggested path

git rebase --onto origin/dev abbcf21 feat/launcher-reachability-probe
gh pr edit 68 --base dev

That drops the three superseded commits and keeps the four probe commits, after which the full matrix runs for the first time.

One thing to check when it does

P0-PRIV-01 is blocking and was explicitly not softened for 0.x. A developer flag gates activation, not classification — please confirm the capability declarations cover outbound STUN traffic, and that nothing contacts a host before the flag is enabled. Worth settling before the register has to answer it.

github-actions · 2026-09-01T13:07:02Z

Unity PR artifacts are ready.

Artifact	Link
VPM package listing	Download artifact

Workflow run: Unity Source Validation #281
Commit: 9e0c7ef

Artifacts are temporary and expire according to the workflow retention policy.

The multiplayer transport cost model turns on P(host unreachable): relay demand is P x (lobby size - 1), drawn once per lobby because reachability is a property of the host. Across plausible P the monthly relay volume spans about a factor of five, which dwarfs any per-byte optimisation available to us, and nobody has measured P. Published NAT-type distributions come from populations that do not resemble ours, so guessing at P is guessing at the answer rather than at a rounding error. The launcher can measure it without anything the hosting feasibility gate blocks: it never touches Robotopia, loads the game, runs a session, or needs a headless server or a transport. It probes the launcher's own host, which is the host a session would run on anyway, and it already makes network calls for update checks and the package registry, so the capability surface is not new. Deliberately not an ICE agent: no candidate gathering, no pairing, no connectivity checks, no nomination. STUN binding requests only. Reporting stays blocked and the pane sits behind the developer flag, so nothing leaves a machine yet. Two internal decision records land with it; they stay under docs/internal/ because they carry monetary figures and that directory is excluded from release zips. The multiplayer contract generator gains a matching analyzer rule. Signed-off-by: Furroxide <221987073+Furroxide@users.noreply.github.com>

`UdpStunTransport.bind` always bound `anyIPv4`, but `StunServerList` parses bracketed IPv6 literals and hands them straight to `send`. A configured IPv6 server therefore reached an unconnected IPv4 socket, where `send` throws rather than reporting a miss; the service's `SocketException` catch turned that into "Could not open a UDP socket", which is not what happened. Binding for a family exposed the deeper version of the same problem. A run uses one unconnected socket for every transaction and decides mapping by comparing reflexive endpoints across a server's address and port. Endpoints in different families are not comparable, so a mixed server list produces a mapping verdict that means nothing — the IPv4-only socket was accidentally hiding that by making every IPv6 entry fail. The family is therefore a property of the run. `StunTransportFactory` takes it, `bind` selects `anyIPv4` or `anyIPv6` and keeps only same-family interface addresses for `matchesLocalEndpoint`, and the service derives it from the configured servers and refuses a list that mixes families instead of quietly probing one of them. `anyIPv6` is not dual-stack everywhere — Windows defaults `IPV6_V6ONLY` on and Dart cannot clear it on a datagram socket — so choosing beats assuming. `request` also drops a wrong-family destination rather than letting `send` throw, since the transport's contract is that a transaction reports evidence and never raises. `reachability_probe_test.dart` reached 560 lines with the new cases, past the 500-line cap CI enforces, so the service group moves to a part file. launcher_data 375/375. Reported by Copilot on #68. Signed-off-by: Furroxide <221987073+Furroxide@users.noreply.github.com>

…llowed `_guard` emits its success message only while `state.isBusy` is still set, so a callback that clears `isBusy` itself has to supply its own message. Every other handler in the bloc does one or the other. The reachability settings-changed handler did neither: it emitted `isBusy: false` with no `statusMessage`, so saving the opt-in reported nothing at all and 'Reachability probe settings saved.' was dead text. It now leaves `isBusy` alone and lets `_guard` emit. The probe-run handler is unchanged; it already supplies its own message alongside `isBusy: false`, which is the other half of the same contract. The existing opt-in widget test now asserts the message reaches the status bar, and fails on the previous code. Flutter launcher 59/59. Reported by Copilot on #68. Signed-off-by: Furroxide <221987073+Furroxide@users.noreply.github.com>

The Notes column in this file describes each rule's invariant, not its emit condition — TFMP008 reads "Generated command codecs fit the declared payload limit" and fires when they do not. TFMP014's row followed that convention but read as a bare statement of fact, which is easy to take for the condition that produces the warning. "must fit" keeps the convention and removes the ambiguity, matching TFMP001's "must be non-generic top-level partial classes". Raised by Copilot on #68 as an inverted note; the table's convention is what made it read that way. Signed-off-by: Furroxide <221987073+Furroxide@users.noreply.github.com>

Keep the UDP listener alive for sequential transactions and drain pending work on close. Add loopback failure regressions. Disclose source IP and port exposure before opt-in. Preserve empty server defaults, activation gates, and disabled aggregate reporting. Signed-off-by: Furroxide <221987073+Furroxide@users.noreply.github.com>

Signed-off-by: Furroxide <221987073+Furroxide@users.noreply.github.com>

Match actual UDP response origins to CHANGE-REQUEST flags before accepting filtering evidence. Preserve the transaction when an unrelated source responds. Keep unsupported discovery unknown: require a usable alternate and explicit completed filtering probes. Add loopback origin and unsupported-server regressions before fixing each defect. Signed-off-by: Furroxide <221987073+Furroxide@users.noreply.github.com>

Handle nonblocking responder backpressure without increasing deadlines or weakening delivery assertions. Preserve one listener per owned socket until teardown. Signed-off-by: Furroxide <221987073+Furroxide@users.noreply.github.com>

Furroxide · 2026-09-08T01:45:50Z

Addressed the routing, CI, and privacy concerns in the rebased branch, now at 9e0c7ef3f0f956cef232f7ef16d3d3a0e9c1584b. It is based on current dev (da47dc7); the superseded #67 commits and historical merge were removed, with the original source history preserved. Current launch previews, profile/session invalidation, and subscription cleanup were retained and checked against dev.

The pane now discloses that configured STUN servers see the source IP address and port before opt-in. Developer mode, saved opt-in, explicit Run, empty server defaults, and disabled aggregate reporting remain intact. P0-PRIV-01 still requires real approval; local address-free results do not grant a privacy exemption.

Review also fixed the socket being closed after its first transaction, acceptance of replies from incorrect sources, and unsupported discovery being reported as restrictive filtering. Each defect has regressions established before its fix, including an owned loopback source matrix. No public STUN server was contacted.

Final local verification: 891 domain tests, 525 data tests (four existing Windows symlink skips), 84 launcher tests, 56 focused probe/socket tests, clean analyzers/formatting/line limits, fresh Windows launcher build, all eleven SDK packages/seven C# harnesses, and repository/documentation audits. The old Windows data-test divergence did not recur. A transient SDK-output rewrite race and responder-backpressure failure were corrected/isolated without waiving checks; the final full data run is green.

Exact final-head verification is complete: CI 34177635608 and CodeQL 34177633789 passed on 9e0c7ef. All 31 checks and five required gates passed, including complete Linux documentation publication. The branch is clean against dev. Release/privacy approval remains separate.

Copilot AI lite review requested due to automatic review settings August 22, 2026 18:45

Copilot started reviewing on behalf of Furroxide August 22, 2026 18:46 View session

Copilot AI reviewed Aug 22, 2026

View reviewed changes

Furroxide force-pushed the feat/roboworld-local-import branch from d52fe85 to 96777e3 Compare August 22, 2026 20:21

Furroxide force-pushed the feat/launcher-reachability-probe branch from b790a2d to 024e812 Compare August 22, 2026 20:21

Furroxide force-pushed the feat/roboworld-local-import branch from 96777e3 to 39b4dc4 Compare August 22, 2026 22:04

Furroxide force-pushed the feat/launcher-reachability-probe branch from 024e812 to 251a178 Compare August 22, 2026 22:04

Furroxide force-pushed the feat/roboworld-local-import branch from 39b4dc4 to abbcf21 Compare August 26, 2026 14:28

Furroxide force-pushed the feat/launcher-reachability-probe branch from 251a178 to 7067cb8 Compare August 26, 2026 14:28

Furroxide force-pushed the feat/roboworld-local-import branch from abbcf21 to b9b07cd Compare August 26, 2026 14:39

Furroxide changed the base branch from feat/roboworld-local-import to dev September 1, 2026 12:48

Furroxide added 6 commits September 8, 2026 03:08

docs(launcher): refresh probe contract references

8ada9f0

Signed-off-by: Furroxide <221987073+Furroxide@users.noreply.github.com>

Furroxide force-pushed the feat/launcher-reachability-probe branch from 057fecd to 8ada9f0 Compare September 8, 2026 01:19

test(launcher): await loopback UDP writability

9e0c7ef

Handle nonblocking responder backpressure without increasing deadlines or weakening delivery assertions. Preserve one listener per owned socket until teardown. Signed-off-by: Furroxide <221987073+Furroxide@users.noreply.github.com>

Navigation Menu

feat(launcher): measure host reachability behind a developer flag - #68

feat(launcher): measure host reachability behind a developer flag#68
Furroxide wants to merge 8 commits into
devfrom
feat/launcher-reachability-probe

Furroxide commented Aug 22, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Furroxide commented Aug 28, 2026

Uh oh!

github-actions Bot commented Sep 1, 2026 •

edited

Loading

Uh oh!

Furroxide commented Sep 8, 2026 •

edited

Loading

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

Furroxide commented Aug 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Behavior

Verification

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Furroxide commented Aug 28, 2026

CLEAN here is not evidence

There is effectively no CI evidence

Three commits are already superseded

Suggested path

One thing to check when it does

Uh oh!

github-actions Bot commented Sep 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Furroxide commented Sep 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Furroxide commented Aug 22, 2026 •

edited

Loading

`CLEAN` here is not evidence

github-actions Bot commented Sep 1, 2026 •

edited

Loading

Furroxide commented Sep 8, 2026 •

edited

Loading