Testing and validation

The repository keeps no CI configuration, no test report, no coverage file and no run log: there is no .github directory, and .gitignore excludes *.log and the .symbiote/ run directory.

Sourcerepository file tree (161 entries) · .gitignore @ 9243204

The build ledger reports “847 tests passing; 6/6 end-to-end experiments passing; ruff, format, mypy --strict clean” at its last gate.

Nothing retained in the repository shows that run. 847 is plausibly the collected case count after parametrization, but this portal did not collect or run the suite and cannot confirm it.

Sourcedocs/BUILD_STATUS.md (“Last verified gate”) @ 9243204

The deterministic CLI run keeps head hash 7eddddf83ff9a85a….

No test pins this value. test_cli.py::test_deterministic_runs_are_byte_identical asserts that two runs agree with each other, not that they produce this hash.

Sourcedocs/BUILD_STATUS.md (“Determinism check”) @ 9243204

scripts/gate.sh runs pytest, ruff check, ruff format --check, mypy --strict, a deterministic CLI run with replay, the six experiments re-verified from disk, and an inspect of the self-evolution log.

The script exists; no output from running it is kept.

Sourcescripts/gate.sh @ 9243204

395 test functions are present across 22 test files; 58 of them are parametrized, so pytest would collect more cases than there are functions.

A count of what exists. It says nothing about what passes.

Sourcetests/test_*.py (parsed with Python’s ast; def test_* at module level and in Test* classes) @ 9243204

Architecture tests enforce the layering, the purity of the decision path, the banned imports, zero runtime dependencies, and that no placeholder package exists for an unbuilt layer.

Sourcetests/test_architecture.py::test_layering, ::test_decision_path_is_pure, ::test_runtime_has_no_third_party_dependencies, ::test_only_implemented_packages_exist @ 9243204

Replay tests forge events — decisions, verdicts, bond transitions, routing, generations, inflated metrics — and assert that replay rejects each one, against an honest stream that replays cleanly.

Sourcetests/test_policy_engine.py::test_forged_decision_event_is_detected @ 9243204 · tests/test_verification.py::test_replay_rejects_forged_verdicts @ 9243204 · tests/test_bond_engine.py::test_replay_rejects_forged_transitions, ::test_an_honest_bond_stream_replays @ 9243204 · tests/test_models.py::test_replay_rejects_forged_routing @ 9243204 · tests/test_evolution.py::test_replay_rejects_forged_generations, ::test_replay_rejects_inflated_metrics @ 9243204

Determinism tests assert that identical inputs produce byte-identical audit logs, from the CLI run through every domain component to the six experiments.

Sourcetests/test_cli.py::test_deterministic_runs_are_byte_identical @ 9243204 · tests/test_world_model.py::test_identical_runs_produce_identical_audit_logs @ 9243204 · tests/test_symba.py::test_experiments_are_deterministic @ 9243204

The experiments’ lab is three in-memory services — svc-alpha (ok, 12 ms), svc-beta (ok, 20 ms) and svc-gamma (degraded, 900 ms) — each with one health observer. The capability the experiments bond to is reliance on a service’s health reading.

Stated so the experiments are read at their real size: they prove the machinery behaves, not that SYMBIOTE can discover anything in an environment it does not already describe.

Sourcesrc/symbiote/symba/experiments.py (synthetic_lab) @ 9243204

Each experiment records named checks (38 across the six) and then a full integrity check of its own log: hash chain, decision re-derivation, domain authorization, and live state equal to replayed state.

Sourcesrc/symbiote/symba/experiments.py (run_experiment) · src/symbiote/symba/telemetry.py @ 9243204 · tests/test_symba.py::test_integrity_flags_a_forged_event, ::test_integrity_flags_a_live_state_that_diverges_from_the_log @ 9243204

nominal-mission: An operator-issued mission runs the full loop and ends with two verified reliances, findings in memory, the swarm closed and the organism back in DORMANT.

6 named checks — mission completed; both reliances verified; returned to DORMANT; findings recorded to memory; stayed inside the mission boundary; swarm fully closed.

Sourcesrc/symbiote/symba/experiments.py (nominal_mission) @ 9243204 · tests/test_symba.py::test_every_experiment_passes_with_full_integrity @ 9243204

operator-takeover: The operator takes CONTROL mid-experiment; the kernel can neither spawn nor release it, and the mission completes only after the operator hands back.

5 named checks — CONTROL paused the mission; kernel cannot spawn under CONTROL; kernel cannot release CONTROL; an observer console cannot release CONTROL; mission completed after release.

Sourcesrc/symbiote/symba/experiments.py (operator_takeover) @ 9243204 · tests/test_symba.py::test_every_experiment_passes_with_full_integrity @ 9243204

containment-escape: A Scout reaching outside its footprint freezes the organism in ESCAPE before the breach reaches the lab; only the External Controller can clear it, and it terminates the run.

9 named checks — mission frozen; organism in ESCAPE; the breach never reached the lab; no spawning while frozen; the operator cannot clear containment; the kernel cannot clear containment; the operator cannot take CONTROL during containment; External Controller terminated the run; TERMINATED is absorbing.

Sourcesrc/symbiote/symba/experiments.py (containment_escape) @ 9243204 · tests/test_symba.py::test_every_experiment_passes_with_full_integrity @ 9243204

honest-failure: A degraded service fails its pre-registered criteria: the mission fails, nothing is verified, and the failed bonds are kept as evidence.

4 named checks — mission failed; no bond verified; failed bonds kept as evidence; failures counted.

Sourcesrc/symbiote/symba/experiments.py (honest_failure) @ 9243204 · tests/test_symba.py::test_every_experiment_passes_with_full_integrity @ 9243204

special-modes: OVERKILL raises only the swarm ceiling, DESPERADO is refused without documented failures, neither the kernel nor an observer can set a special mode, and GOOFY GUY changes no fact.

6 named checks — OVERKILL raises only the swarm ceiling; DESPERADO needs documented failures; a kernel cannot set special modes; an observer cannot set special modes; GOOFY GUY changes no fact but the flag itself; GOOFY GUY only frames the rendering.

Sourcesrc/symbiote/symba/experiments.py (special_modes) @ 9243204 · tests/test_symba.py::test_every_experiment_passes_with_full_integrity @ 9243204

self-evolution: One gated improvement cycle: a measured limitation, a refused proposal against the Permission Engine, an operator-accepted held-out improvement, a rejected rival and a rejected self-reported win, then a rollback by the External Controller.

8 named checks — a measured limitation was detected; evolution may not touch the Permission Engine; the organism cannot approve itself; the measured improvement was accepted; a rival with false positives was rejected; a self-reported win was rejected; the accepted generation is active; rollback restored genesis.

Sourcesrc/symbiote/symba/experiments.py (self_evolution) @ 9243204 · tests/test_symba.py::test_every_experiment_passes_with_full_integrity @ 9243204