overview planning

The backlog tasks finished in November 2026, moved here from Backlog when closed.

Sources: Backlog, Status.

idtaskclosedcommitverified
B-001Fill the backlog: one row per open item of Milestones (M2 onwards), the “waiting” items of Status, open SP-nnn problems and unanswered SR-nn requirements2026-11-17not committed yet (branch dev-danielb)rows B-002 to B-059 written from Milestones (M3b to M7, the order of work, P0 demonstration 3), Semantic Boundary Experiments (E-0 to E-10), the open problems of the Problem Log, the requirements with status open, confirm or planned, and the open items of Review 2026-11-10 Fable (external); every wikilink in Backlog checked to resolve. Found on the way: SR-89 said demonstration 2’s Moonkale side was open, while Milestones records it passing (Moonkale 46a06b65); SR-89 corrected
B-002Annotations move with the truth: give_truth(…, exported_annotations) refuses with AnnotationsNotExported; take_truth(…, annotations) stores them in the same transaction; annotate_many skips an annotation already stored2026-11-17a8e2c45 (branch dev-danielb)annotations_move_with_the_truth in crates/sophia-store/tests/folder.rs on SQLite, redb, LMDB and DuckDB covers the three rules, a re-sync logging nothing (before and after give_truth), a refused switch storing nothing, and a changed status being stored and logged; cargo test --workspace --all-features: 342 passed, 0 failed; clippy -D warnings and fmt --check clean. Settled while building: “already stored” means stored exactly as given, so a status update is still written. Reviewed by Mistral (Review 2026-11-17 Mistral B-002 (external)): no defect, one missing test added. Moonkale told through HANDOFF-from-sophia.md
B-003Interface::propose_text(namespace, text, token) for sophialang files, returning a TextProposal (a Proposal and the parser’s diagnostics)2026-11-18809b7a9 (branch dev-danielb)SPEC-010 acceptance 3 passes on SQLite, redb, LMDB and DuckDB: an_exported_file_proposes_back in crates/sophia-interface/tests/acceptance.rs (the round trip changes nothing and writes nothing; the proposed ids equal what write_text and create_text store; stale token, parse error), plus a_proposal_refuses_names_outside_the_file; 14 spec tests from Mistral’s probes in tests/propose_text.rs. cargo test --workspace --all-features: 364 passed, 0 failed; clippy -D warnings and fmt --check clean. Found while building: the export did not round-trip, because a named relation signature printed again as a new relation. Such names are now export aliases. Reviewed by Mistral (Review 2026-11-17 Mistral B-003 (external)): one documented limitation (provisional ids of new relations), one race not reproduced; no accept yet (B-060). Moonkale told through HANDOFF-from-sophia.md
B-004Specify sharing a store-truth folder: the text export committed beside the generated files and the rebuild from it (option b); the pointer-file idea as a later option2026-11-18e461b94 (branch dev-danielb)SPEC-011 Sharing a Store-Truth Folder drafted (a proposal; five questions for Daniel). It covers what is exported and why, the layout and envelope, git merge behaviour, Store::export_text and import_text with their checks, Moonkale’s triggers, what a rebuild loses, and the pointer file as §9. SR-149’s status, SPEC-010 and Specifications point to it. The draft’s claims about the store were checked against the code by Mistral (Review 2026-11-18 Mistral SPEC-011 (external)): 12 findings, 11 right and 1 partly, all folded in. Nothing built, nothing measured (acceptance 5). Moonkale asked to review it through HANDOFF-from-sophia.md
B-011Check Fable’s S-9 (deleting a namespace leaves sub-namespace markers; refused while a name equals its path) and S-11 (a removed fact reaches subscribers with no keys); fix what is right2026-11-18ab5214a (branch dev-danielb)both right, both fixed, each with a test that failed before its fix: deleting_a_namespace_removes_sub_namespaces and a_removed_fact_reaches_the_feed_with_its_key in crates/sophia-interface/tests/acceptance.rs on SQLite, redb, LMDB and DuckDB. Codex reviewed the diff: deleting the root is now refused (the_root_namespace_is_not_deleted, failed before); the concurrency gap is SP-008’s (B-010). cargo test --workspace --all-features: 373 passed, 0 failed; clippy -D warnings clean; triage in Review 2026-11-10 Fable (external)
B-012wasm tests in CI (today run by hand under Node)2026-11-186908ee1 and the closing commit (branch dev-danielb)a wasm job in .github/workflows/ci.yml runs tests/wasm_read.rs and tests/wasm_deep.rs of sophia-interface on wasm32-unknown-unknown under Node, with wasm-bindgen-test-runner installed at the version Cargo.lock gives wasm-bindgen (0.2.129). Locally: 4 of 4 and 3 of 3. On GitHub, run 44208690030 (dispatched on dev-danielb): both jobs succeeded, wasm with test result: ok. 3 passed and ok. 4 passed (gh run view 44208690030 --log). Browser runs stay by hand. Notes: sophia_interface.md, Review 2026-11-10 Fable (external)
B-014The set-oriented native executor: a prefix access path, anti-joins for negation over scans, borrowed field reads2026-11-1897c8721 (branch dev-danielb)B-17 re-run at 10⁴ (four backends) and 10⁵ (SQLite, redb, LMDB), before and after on the same machine, in Benchmark Results: names by prefix 30–43 → 0.08–0.16 ms; unreferenced definitions 320 → 142 ms on SQLite at 10⁵, 2 011 → 73 ms on DuckDB at 10⁴; count by kind 30–48 % faster; recursive workloads unchanged. The differential tests (execution.rs, properties.rs) pass unchanged; new tests/set_oriented.rs compares three tunings on four backends and the SQL translation. Found on the way: the estimated byte count dropped base tuples read in the middle of a join (fixed, regression test). cargo test --workspace --all-features: 385 passed, 0 failed; clippy -D warnings and fmt --check clean. Reviewed by Mistral (Review 2026-11-18 Mistral B-014 (external)): no defect, two tests added. Remaining gap to SQL on SQLite: a full-row scan of t_definition (B-061, proposed). No Moonkale handoff: the query API is unchanged apart from additions (Tuning, run_tuned, two Stats fields)
B-017A hard memory bound for untrusted queries, from outside the executor2026-11-180f761b9 (branch dev-danielb)Daniel chose a worker process (2026-11-18). sophia_query::bounded::run_bounded runs a query in sophia-query-worker under RLIMIT_DATA (Linux); over the bound it returns Error::MemoryBound and the caller survives. Each backend gained open_read_only. tests/bounded.rs on SQLite, redb, LMDB and DuckDB with every logical limit and the byte estimate off: 700² derived pairs fail with MemoryBound under 64 MiB, a closure gives the in-process rows under the same bound; measured once: 1 500² pairs answer under 6 GiB, fail under 128 MiB. Codex’s review found three bugs, all reproduced and fixed (Review 2026-11-18 Codex B-017 (external)). cargo test --workspace --all-features: 391 passed, 0 failed; clippy -D warnings and fmt --check clean. Limits: Linux only; redb and DuckDB only while no other process holds them for writing. No Moonkale handoff (additions only)