solstone/core/fixtures/ at main · solpbc.org/solstone · Tangled

solpbc.org / solstone

0

personal memory agent

0

solstone / core / fixtures

1 folder 6 files

at main

Jer Miller feat(native-sol): port sol link join 11hrs ago

feat(native-sol): port sol link join Consume spl-rust at tag = "v0.1.0" (resolved 6ccda09777fd0da0489d77b3098892492c90c819), declared once at the workspace level. Add the host-only solstone-core-sol-link crate for all spl-transport use and exclude it from check-rust-ios. sol link is a desktop/linked-system surface; phones pair through spl-swift, so this is a permanent product boundary, not deferred iOS debt. Keep sol link routed to Python. TOP_LEVEL_COMPAT_COMMANDS, sol_compat_inventory.py, sol_compat_cli.py, and the sentinel remain untouched; the cutover lode flips routing. Document seven declared expected-differs in docs/design/native-sol-client/08-link-join-design.md. Leave ring's three-target native-dependency release proof to the cutover lode that wires this crate into the shipping closure.

11 hours ago

callosum_registry.json

feat(core): generate callosum and cogitate contract fixtures Generate the core contract fixtures from the live Python sources through public APIs so downstream ports have deterministic data instead of hand-copied literals. The cogitate fixture keeps the runtime preamble digest-only for low-noise verification and carries future tiers separately from the capability matrix because they intentionally have no capability entry.

1 week ago

cogitate_contract.json

fix(cogitate): one approved journal-command vocabulary Define {identity, health, talent} once in cogitate_contract.py and render it into the policy, the prompt lint, the inventory scan, the runtime preamble, both model-visible tool descriptions, and the sol-tool hint. The preamble no longer bans bare `journal ...` while policy allows three families; it now states the positive rule — normal access is `sol` / `sol call`, the three approved host families run directly as `journal <family> ...` and must not be prefixed with `sol call`. Refuse the two forms seen in the stuck production run before execution, each with the working repair: `sol call journal <family> ...` for any of the three families, and the bare `journal search` / `journal facet` retries.

23 hours ago

edge_schema.json

feat(core): create derived-edge schema in native indexer The native solstone-core indexer now creates the edge_files and edges tables, the three edge indexes (edges_path, idx_edges_src, idx_edges_dst), and the edge schema-version sentinel row (edge_files path='edges:__schema__', mtime=1) as part of the index DB, matching Python's _ensure_edges_schema exactly. Tables are created empty (sentinel only) - edge extraction/population remains a later lode. This makes a native-built index a safe drop-in for readers that query edges/edge_files. All schema DDL (files, chunks, edge_files, edges, indexes) plus the sentinel now commit in a single transaction so an interrupted setup never leaves tables without the sentinel. A committed golden fixture (core/fixtures/edge_schema.json), generated from Python's real _ensure_edges_schema via scripts/build_core_fixtures.py, backstops schema drift; a Rust PRAGMA-introspection test asserts native/Python parity. No edge extraction, no --rebuild-edges subcommand, and no port of Python's in-place schema-version migration branch (native targets fresh/--reset DBs only). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

1 week ago

markdown_chunks.json

fix(indexer): count markdown bounds in characters, not bytes The native markdown formatter used str::len() for the 2048-char line bound and the 4096-char chunk bound. In Rust that counted UTF-8 bytes, while solstone/think/markdown.py counts Unicode characters. Non-ASCII markdown could therefore be dropped or header-stubbed earlier than Python, and the stub could label a byte count as chars. Use .chars().count() for both bounds and pass the same character count into the oversized header stub as the reported original size. Add three format_markdown-generated oracle cases: non_ascii_line_under_char_bound_over_byte_bound, non_ascii_line_over_char_bound, and non_ascii_chunk_under_char_bound_over_byte_bound. These record cardinality and warnings only; token lists are omitted with an explicit marker because non-ASCII falls outside the fixture's ASCII tokenizer-equivalence guarantee. The markdown producer test now consumes core/fixtures/markdown_chunks.json through produce_chunks(Family::Markdown, ...) and asserts exact ordered token lists, replacing the earlier prefix-only assertions. Both issues were found by self-audit of 971f4f840. Co-Authored-By: OpenAI Codex <noreply@openai.com>

5 days ago

speaker_filterbank.json

test: embed speaker filterbank reference waveform Add waveform.samples_f32_le_base64 to the speaker filterbank fixture so the reference waveform round-trips as exact little-endian f32 bytes. The fixture previously stored only NumPy PCG64 generation parameters, which a Rust port cannot reproduce byte-for-byte. Bump identity.fixture_version from 1 to 2 for the fixture schema change. Apple hardware was not available; this fixture-only change does not exercise CoreML session construction or macOS loader behavior.

21 hours ago

speaker_stage_boundaries.json

fix(core-fixtures): harden speaker fixture drift checks Make the ambiguous-band speaker-evidence case prove it lands in the else branch by asserting its preconditions against imported calibration constants, so a future recalibration that moves the case out of that band fails loudly instead of silently dropping coverage. Reject non-finite tolerance-managed values. NaN and infinities no longer evade comparison; abs(nan - expected) > tolerance is false, so a committed nan previously passed as agreement. Unparseable row tokens now produce a located diagnostic instead of a bare ValueError. Remove the unused ArtifactDescriptor.path field and drop bare production numerals from tolerance rationale comments so the text cannot go stale against recalibration.

2 days ago