# Kindynos evidence register — 2026-08-06

Every row traces to a genuine captured run or a source-verified path. **Status vocabulary is the six primary terms only** (Observed interface < Source verified < Test exercised < Executed < Integrated path < Deployed); qualifiers such as *prior run*, *research prototype*, *in-process* live in the Notes column. No item is Integrated-path or Deployed. Commits: EVA `4f030361`, EOS `v0.24 (6b9ae85)`; KAILA/CSUITE/STAR are the canonical OTHER-APPS/KINDYNOS-APPS working copies (no git SHA).

An anchored HTML view is at `evidence-register.html` (e.g. `evidence-register.html#KAILA-01`). Packaged artifacts are in `downloads/evidence/`.

| ID | Application | Capability | Status | Command / endpoint | Output file | Claim | Boundary | Notes |
|---|---|---|---|---|---|---|---|---|
| EVA-01 | EVA | Event impact & stakeholder consequences | **Executed** | `pytest engine10/tests/test_2_6_direct_impact_correctness.py engine10/tests/test_2_7_conservation_pair_ecosystem.py` | eva-six-families-run.txt | One event produces typed, signed per-stakeholder consequences that conserve to the transfer. | Synthetic golden ecosystems. Servicer -180,000,000.00 / regulator +100,000,000.00 / taxpayer welfare; flood owner +6,550,000.00 / insurer -6,550,000.00 sum-to-zero. | Captured in the consolidated six-families run. |
| EVA-02 | EVA | Investment Workbench | **Executed** | `pytest portfolio_intel/tests/{test_portfolio_intel_engine,test_impact_tracer,test_no_fabrication_unvalued,test_cross_module_evaluator}.py` | eva-six-families-run.txt | A portfolio is evaluated against events with per-position attribution and honest UNVALUED. | Strategy/execution leg source-verified, not separately executed. | Captured in the consolidated six-families run. |
| EVA-03 | EVA | Forensics | **Executed** | `pytest forensics/tests/{test_forensic_engine,test_counterfactual,test_forensic_query,test_loan_intake,test_regulatory_report}.py` | eva-six-families-run.txt | Forensic attribution runs backward from an outcome with counterfactuals and a case artifact. | 3 temporal/reverse-traversal tests error on the archived-engine9 tripwire (migration targets, not defects). | Tripwire-erroring files excluded from the consolidated run; live replacements pass. |
| EVA-04 | EVA | Decision / Persona | **Executed** | `pytest game_theory/tests/ decision/tests/{test_reaction_models,test_decision_objects}.py` | eva-six-families-run.txt | A proposed action is evaluated against reactive actors with a payoff/minimax verdict and a decision record. | No literal 'backfire verdict' class — verdict = reaction selection. test_decision_engine.py stale on engine9 tripwire. | — |
| EVA-05 | EVA | Cross-ecosystem / systemic | **Executed** | `pytest identity/tests/ grkg/tests/{test_grkg_propagation,test_grkg_engine}.py` | eva-six-families-run.txt | Identity links + shared stakeholders/events drive a named cross-ecosystem propagation path with output. | systemic/reverberations source-verified only; one (1)-snapshot identity test asserts a stale endpoint count. | — |
| EVA-06 | EVA | Model-input governance & refusal | **Executed** | `pytest engine10/tests/{test_f10_typed_input_defaults,test_phase1_resolver}.py portfolio_intel/tests/test_no_fabrication_unvalued.py` | eva-six-families-run.txt | Inputs resolve runtime->asset->cited default->MISSING; a missing required input refuses, never silently defaults. | Status is REAL / PARTIAL / UNVALUED; governed no-substitute behavior. | — |
| EVA-07 | EVA | 13-stage firing contract | **Source verified** | `read engine10/firing.py:156-181 (fire_event; on_stage_complete hook)` | A1_firing_stage_sequence.txt | EVA fires an event through a fixed 13-stage pipeline via one fire_event entrypoint. | Documented stage order from live source; a live 13-stage trace is captured separately (EVA-12). | — |
| EVA-08 | EVA | Live firing/values regression | **Executed** | `pytest engine10/tests/test_us_housing_v9_permanent_regression.py -v` | eva-firing-regression-full.txt | A live firing reproduces the borrower equity decomposition (equity-realization P&L + lien-release balance-sheet swap). | Regression-test record, NOT a full 13-stage trace. 5 passed. Asserted EXPECTED_EQUITY_REALIZATION=175,111.0244 (post-migration model-derived sale_proceeds 560,000.4411), tolerance +/-0.05; the docstring's 175,110.5833 is a stale derived intermediate from the old hardcoded 560,000 sale price and is historical only. | Full raw pytest capture replaces the earlier curated note. |
| EVA-09 | EVA | Hash-locked deterministic replay | **Executed** | `pytest engine10/tests/test_phase8_replay.py -v` | deterministic-replay-full.txt | A firing path replays byte-identically to a recorded trace-hash baseline (12 events locked). | 4 passed, 1 deprecation warning. Full raw capture incl. summary. Hashes are asserted in-test against the baseline fixture and do not print to stdout; baseline HOME_SALE_PAYOFF trace_event_v2_hash b15d1fad... lives in the fixture. | Replaces the earlier curated extract. |
| EVA-10 | EVA | Access surfaces (direct / REST / MCP) | **Source verified** | `engine10/api.py:727 POST /api/v1/firing/fire-event ; eva_mcp/eva_mcp/server.py (9 tools)` | — | One canonical fire_event is invocable directly, over REST, and over MCP (connected in Claude Desktop). | Direct = Executed (EVA-01/08). REST = source-verified, server not booted. MCP = implemented + connected (claude_desktop_config.json), acceptance test not run (permission-denied); no transcript fabricated. | Live REST/MCP runs remain in the gap plan. |
| EVA-11 | EVA | Operator interface (breadth) | **Observed interface** | `captured screenshots` | assets/screenshots/eva-operator.webp | EVA is a substantial operator environment with many interface surfaces. | Interface existence/breadth only; captured 2026-08-04; does not by itself prove every backend path. | Also eva-event-impact.webp, eva-decision.webp. |
| EVA-12 | EVA | 13-stage firing trace (live, offline) | **Executed** | `read-only driver attaching on_stage_complete to fire_event (offline PackageLoader+PythonRunner)` | eva-13stage-trace.txt | All 13 firing stages are observed live for one event, with 27 stakeholder impacts and channel_1=4. | Different payload from the phase8 baseline, so this trace's event_v2_hash (a0400b24...) is NOT the phase8 replay hash. Offline, no server. | Driver lives in scratchpad, not in the EVA repo; capture is raw stdout. |
| KAILA-01 | KAILA | Canonical signal + version negotiation (Connect) | **Executed** | `pytest sdk/tests/test_data_contracts.py sdk/tests/test_version_negotiation.py -v` | kaila-signal-contract-run.txt | A canonical RiskSignal validates with a deterministic content hash, and incompatible engine versions are rejected. | 47 passed (test_canonical_hash, test_incompatible_raises_version_conflict). In-process only. | — |
| KAILA-02 | KAILA | Artifact registry + hash-chained audit (Connect) | **Executed** | `pytest registry/tests/test_registry.py -v` | kaila-registry-run.txt | Artifacts register with content hashes and a verifiable SHA-256 hash-chained audit ledger. | 81 passed. In-process registry, not a networked service. | — |
| KAILA-03 | KAILA | Calibration engine (Connect) | **Executed** | `pytest calibration/tests/test_calibration.py -v` | kaila-calibration-run.txt | Named quantitative calibration runs and produces real values (Hawkes stability, copula PSD projection, MLE). | 127 passed (test_stability_constraint, test_psd_enforced, test_psd_projection_2x2). Pure-Python. | — |
| KAILA-04 | KAILA | Global freeze governance (Connect) | **Source verified** | `freeze/global_freeze.py (in-process over temp SQLite, prior reader demo)` | — | A cross-app freeze snapshot enforces and raises on missing declarations. | Demonstrated in a prior in-process reader demo (FreezeViolationError); not separately packaged this pass. | qualifier: prior run, in-process. |
| KAILA-05 | KAILA | Mediation / route resolution (Connect) | **Source verified** | `tests/test_phase2_mediation.py (source-verified; prior reader run 83 passed)` | — | A signal's route resolves and hops are hash-chained to a finalized chain. | Recorded routing INTENT + in-process hash-chaining, NOT network transport delivery. KAILA does no transport in this codebase. | qualifier: in-process; delivery status is caller-supplied, not executed transport. |
| KAILA-06 | KAILA | Mediation spine (Connect) | **Test exercised** | `pytest tests/test_phase2_mediation.py (prior reader: 83 passed)` | — | Route resolution + hop hashing + chain finalization pass as a suite. | Not re-captured this pass; see KAILA-05 boundary on transport. | qualifier: prior run. |
| CSUITE-01 | CSUITE | Deterministic rehearsal spine (Rehearse) | **Executed** | `pytest execution/tests -v` | csuite-execution-suite-run.txt | A rehearsal freezes state, forks branches, applies interventions, and verifies an append-only hash-chain replay. | 49 passed (test_fork, test_verify_replay, test_get_final_hash, test_immutability_fork_replay, test_persist_intervention). Agent reactions are deterministic KAILA sandbox stubs, not a live LLM. | — |
| CSUITE-02 | CSUITE | Scored rehearsal session (Rehearse) | **Source verified** | `read csuite_session.db (ro)` | — | A completed rehearsal session carries a component-weighted score that reconciles. | SSN_49443fc2 MARKET_SPREAD_SHOCK overall_score 0.725 = 0.4x0.8 + 0.3x0.7 + 0.3x0.65. Prior run, not re-executed; top-level session route persisted 0 branches. | qualifier: prior run (database read). |
| STAR-01 | STAR | v4 governed media pipeline (Communicate) | **Executed** | `pytest v4/tests/ -v` | star-v4-suite-run.txt | A governed media pipeline deterministically plans scenes, assigns job identity, assembles metadata, records a 7-stage replay ledger. | 386 passed. Render is PLANNING-ONLY — no external provider invoked; render outputs are placeholder hashes. | — |
| STAR-02 | STAR | Determinism certification gate (Communicate) | **Executed** | `pytest v4/tests/test_certification.py v4/tests/test_determinism.py -v` | star-determinism-gate.txt | Running the institutional pipeline twice yields a byte-identical cumulative_hash (asserted in-test). | 75 passed (test_cert_identical_input_identical_hashes, test_cert_cumulative_hash_stable, test_cert_chain_integrity_N_runs, test_cross_bundle_determinism). | — |
| STAR-03 | STAR | Pre-rendered exports (Communicate) | **Observed interface** | `ls STAR MEDIA/exports/vega_export_*.mp4` | — | Prior rendered outputs exist on disk. | 11 vega_export_*.mp4 from a separate non-v4 render path; existence only — NOT outputs of the v4 deterministic pipeline; no live render performed. | qualifier: prior run, existence only. |
| EOS-01 | EOS/EPL | Named refusal (backend unavailable) | **Executed** | `eos run sequential_crash` | named-refusal-run.txt | A backend-unavailable path returns UNVALUED, publishes no effects, commits no state. | Two events UNVALUED (BACKEND_UNAVAILABLE), state v0->v0, exit 3. Bounded EOS path. | — |
| EOS-02 | EOS/EPL | State carry-forward | **Executed** | `eos run mortgage_rate_path` | mortgage-path-run.txt | State written by one event is read by the next. | 3 sequential events; balance/cum_prepay carry forward; example model. | — |
| EOS-03 | EOS/EPL | Domain-blind kernel | **Executed** | `python3 -m conformance.tests.test_blindness` | eos_domain_blindness.txt | The execution core contains none of the domains it runs. | 6 sub-proofs, exit 0; small authored worlds. | — |
| EOS-04 | EOS/EPL | Portfolio reconciliation | **Executed** | `eos run book_50` | portfolio-50-run.txt | Per-position results sum exactly to the book total. | BOOK delta +433,004.662908, 50 VALUED; synthetic book. | — |
| EOS-05 | EOS/EPL | EPL source example | **Source verified** | `not_applicable` | sequential_crash.epl | Declarative EPL source for a sequential-crash example world. | Source example; not an execution record. | — |
| EOS-06 | EOS/EPL | EOS full suite | **Test exercised** | `python3 -m pytest` | B2_eos_full_suite.txt | The EOS test suite passes. | 29 passed in 10.84s (blindness proof separate). | — |
| EOS-07 | EOS/EPL | Run provenance | **Source verified** | `not_applicable` | PROVENANCE.txt | The EOS output files are unmodified terminal output, re-verified. | Provenance doc; re-verified 3 Aug 2026 EOS v0.24. | — |
| SPEC-01 | Specialist · CRIS | Cyber risk | **Test exercised** | `pytest tests/test_cris_signal_contract.py -v ; verbatim PoissonCyberModel fit` | cris-contract-run.txt | The canonical cyber RiskSignal contract passes (deterministic hashing + severity mapping); the named Poisson model computes real frequency estimates. | 17 passed (signal envelope, NOT a calibrated cyber model). Poisson fit lambda 3.8333/mo, E[3mo] 11.5, P(=5|3mo) 0.0169793638. | model function exercised (Poisson run under a scipy interpreter — CRIS's own venv lacks scipy and the module also needs pandas), NOT integrated application path. |
| SPEC-02 | Specialist · DEFIANT | DeFi risk | **Executed** | `run DefiMonteCarloEngine.run_portfolio_simulation (real backend modules, unmodified)` | defiant-montecarlo-run.txt | A DeFi portfolio VaR engine runs with Cholesky-correlated shocks and beta-distribution losses, producing VaR percentiles. | expected_loss 359,695.52; VaR95 508,080.82; VaR99 555,375.34; cholesky_decomposition + beta_distribution; reproducible on same seed. | research prototype executed: 0 tests, config defaults, synthetic uncalibrated inputs; NOT integrated application path. |
| SPEC-03 | Specialist · USRISK | US economic risk | **Executed** | `run simulate_copula_hawkes (real hawkes.py + factors.py, unmodified)` | usrisk-hawkes-run.txt | A multivariate self-exciting Hawkes process driven by Student-t copula shocks computes aggregate tail risk, reproducibly. | branching_ratio 0.16 (subcritical); agg_VaR99 35.0; agg_TVaR99 38.8; bitwise-identical on same-seed re-run. | research prototype executed: 0 tests, model function only (the Flask app path is non-functional — app.py imports symbols this model version does not provide); synthetic params. |

**Total: 33 evidence items** (EVA 12, KAILA 6, CSUITE 2, STAR 3, EOS/EPL 7, Specialist 3). Specialist apps: CRIS/DEFIANT/USRISK; the full specialist inventory is in `SPECIALIST_DILIGENCE_ANNEX_2026-08-06.md`. Anything not yet captured is in `EVIDENCE_GAPS_AND_GENERATION_PLAN_2026-08-06.md`.