FIGURE · operant-attrs
operant test attributes
4,959 (static)
- How it was taken
- Static test-attribute count over crates/: lines matching #[test], #[tokio::test], #[test_case], or #[rstest], at checkout 5a116305. Reproduce with: git grep -E "#\[(tokio::)?test\]|#\[test_case|#\[rstest" 5a116305 -- crates/ | wc -l
- Measured
- 2026-10-26
- Reproducible
- Yes: clone and re-derive.
What the number is made of
The count command, byte for byte
git grep -E "#\[(tokio::)?test\]|#\[test_case|#\[rstest" 5a116305 -- crates/ | wc -l returns 4,959 on the public tree at the pinned commit.
Open the reference →The executed figure above it
The gate ran 7,168 tests with zero failures at commit ca33d4c. The numbers policy prefers the executed figure; the static count stays as the labelled upper bound.
The register entry →The reconciliation, R20
The register previously printed 4,979 for this commit and the command returned 4,959. The correction is recorded, not smoothed over.
The register entry →
How to check it
Run the command in the method line at the pinned commit: it returns 4,959, and the register prints what the command returns rather than a figure remembered from a pass that did not record one (see reconciliation R20). Or run the suite and read the runner output: the executed figure is operant's gate result in this register, which the numbers policy prefers.
Open the artifactThis figure's history
- R3
9,249 test attributes, 19 crates → 9,811 test attributes, 18 crates
Both prior figures came from the outreach profile, which is an outreach asset rather than a measurement. The crate count was also wrong in the other direction from what the profile implied.
- R10
9,811 → 4,979 (static)
Re-measured on the consolidated workspace with the pattern recorded in the method line. The September tree and this tree are not the same codebase at the attribute level; the halving is the restructure, not a claimed quality change, and the register does not dress it up as one.
- R20
4,979 at checkout 5a116305 → 4,959 at the same checkout, with the command recorded byte for byte
Re-running the documented pattern at the pinned public commit returns 4,959, not 4,979. The prior figure did not reproduce because the pass that produced it did not record its exact command. The row now carries the command, and the register prints what the command returns. The rule, learned from the other side of the source ledger's first lesson: a figure is a claim about what a command returns, and the command is part of the figure.
Where this site cites it
The repository's own documents
- AD-006 · Bodies are fenced from the supervisor
- AD-007 · An unfenced run is refused
- AD-023 · Plugins cannot replace a built-in
- AD-035 · One libclang resolver, and no dangling link
- AD-040 · The dashboard token only goes to its holder
- The operant mechanism library
- Organization — the Holon runtime, seats, and work graph
- Kernel — the persistent Python kernel + continual harness
- The agentic core — the ReAct loop
- The tool system
- Context management — eviction, decay, and the compaction ladder
- Gateway — the channel surface
- MCP — both sides of the protocol
- Plugins — WASM extensions
- Sub-agents — delegation, swarm, async
- Browser — the sourcehound engine
- Web tools — sourcehound intelligence
- TUI — the terminal interface
- Vision — the auxiliary-LLM contract
- Voice — TTS, STT, and the voice-mode state machine
- Verification — the harness that proves work
- Autonomous mode — the self-directed loop
- Background review — the post-turn learner
The documents the figures were measured from, published in full rather than summarised.