ObliviOS Developer portal Project site

Home / Decisions / ADR-245

Carry out the open items of the source-only stage that need no owner choice.

Accepted, recorded 17 November 2026.

Decision
ADR-245
Date
2026-11-17
Status
AcceptedAccepted (owner, 2026-11-17). Done and committed locally: oblivios-platform 21a1b43, oblivios-services c9139fe and afcedc5, the evidence and tools in this repository, oblivios-docs regenerated. Not done: the push of oblivios-manifest, oblivios-platform and oblivios-services and the change of visibility (each needs the owner's yes), a rewrite of the history of oblivios-platform for the OTA host, the owner's read of the final set, a run of the new exitd on a phone, a repeat of the mutation run of exitd, the sentence-level read of the SOURCE.md records (about 30 files, screened by pattern only) and a re-run of the counts in the datamap, keyboard and sensor READMEs.
Mentions
ADR-171, ADR-241, ADR-244
Mentioned by
ADR-246, ADR-247, ADR-248, ADR-265

1Decision

Carry out the open items of the source-only stage that need no owner choice (ADR-171, ADR-241, ADR-244), as the owner asked on 2026-11-17 ("fix it", "do what is best", "do everything you can"). (1) The host of the private OTA server leaves the published tree of oblivios-platform: the patch of release/ota-server-url carries the reserved placeholder https://ota.example.invalid/ and no index line, the README and SOURCE.md say so.

(2) exit-daemon/Cargo.lock moves tokio 1.53.1 to 1.53.3, rustix 1.1.4 to 1.1.6 and ctutils 0.4.2 to 0.4.4 (version and checksum lines only), and the prebuilt exitd is rebuilt from that source (SHA-256 d5dca624755f9a40f9cc3bcf19b2cf4fd485f1a1c4b5c21e8fd3f917d1119e63, 19 366 456 bytes, two builds identical, 344 host tests passed).

(3) The five advisories of ADR-244 are assessed in exit-daemon/RISKS.md: three fixed by the update, crypto-bigint is in the lock and in no dependency graph, libsqlite3-sys is linked and not fixable without a new Arti release (accepted); the OSV query of the same day adds chrono and humantime (unmaintained, no vulnerability).

(4) The aggregated licence text of the Rust packages exists: privacy/exit/daemon/THIRD-PARTY-LICENSES.txt (548 packages) and passwords/core/THIRD-PARTY-LICENSES.txt (72), made by exit-daemon/tools/third-party-licences.py. (5) The accepted legal risk per item is recorded in licensing-and-branding-review.md, with the trademark search of 2026-11-17.

(6) The frameworks/base patch that ADR-244 said modifies a file that is absent at the pin is the rename of ReferenceGlobalRootComponent.java (present at the pin with an Apache-2.0 header) to .kt; the evidence tool read the new name. fetch-upstream-evidence.py now reads the old name of a renamed file, and NOTICE counts 90 of 100 headers.

(7) The claim review reads the 15 package READMEs of oblivios-platform in full and screens the SOURCE.md records with a pattern list.

2Why and evidence

Each item was a task that needed no decision: a host name the docs scan already forbids, advisories whose fix is a patch-level bump the lock accepts, a missing licence text, and statements the READMEs could not support. A rebuilt exitd makes the published source and the published binary the same thing, which ADR-244 listed as not true (23 files differed); between 7beab16 and c9139fe the daemon changed by the rename and the three versions only.

3Consequences

The published patch no longer reproduces the landed commit 7a0d44d: a build that wants the private channel sets the real URL in the one line before git am. The history of oblivios-platform still holds the host in older commits until a rewrite, which would need a second force-push; the owner has not asked for it.

The READMEs of the decoy and dummy packages now say that they are dated narratives, and the counts of compartments-profiles were corrected (627 and 492 checks, run on 2026-11-17). The trademark search found no mark containing oblivios in the EU, WIPO or US data of TMview and several near marks in classes 9 and 42 (OBLIVION, OBLIVIOUS and others); it is not an official register search and not an opinion.

exitd has run on no phone since the rebuild; a device row (NET-*, MSK-*) on the first image that carries it is the check.

4Review when

A conflicting mark appears in a register or a letter; OSV lists a new advisory for a linked package; an Arti release uses rusqlite 0.41; the owner wants the OTA host out of the history; exitd fails a device row.

Read from docs/governance/decision-log.md, entry ADR-245.