Final evidence matrix
This matrix records what was actually observed. A status is not a capability claim beyond the stated hypothesis and retained evidence.
| Experiment | Hypothesis | Method / actual execution | Observed evidence | Result | Limitations | Evidence path |
|---|---|---|---|---|---|---|
| 01 Multiple AI | Independent participants can respond to one work item. | Retained prior run. | One work event and three independent proposals. | PASS | Text record only in this checkout. | experiments/01-multiple-ai/results/run-001.md |
| 02 Human review | A human can participate through the same boundary. | Retained prior run. | A human ProposalApproved event with causal fields. |
PASS | Text record only in this checkout. | experiments/02-human-review/results/run-001.md |
| 03 Event chain | Independent participants can form a causal chain. | OCI Redpanda run 8edca9097cf843499738b693d8ef33b4. |
Four events; each derived event points to its recorded parent. | PASS | Deterministic, bounded run only. | experiments/03-event-chain/evidence/run-002/ |
| 04 Replay | Retained events suffice to reconstruct a projection. | OCI Redpanda run 956d0816c8e34246bf2ebaf9769381cb, then fresh-log replay. |
replay.json replays all four IDs and matches the recorded state. |
PASS | One schema/version and short retention window. | experiments/04-event-replay/evidence/run-003/ |
| 05 Failure recovery | A restarted participant catches up from retained input. | Prior crash/restart run. | Two work events, one proposal. | INCONCLUSIVE | No raw assignment/offset telemetry. | experiments/05-failure-recovery/RESULT.md |
| 06 Live LLM | A live participant can emit a causally derived proposal. | OCI Redpanda run 5189d571b4954c688fd9f9f79f484aa8; one Azure Sweden/Luna call. |
WorkCreated → Azure ProposalCreated; provider/model/latency metadata retained. |
PASS | Immutable payload participant_id conflicts with authoritative source/actor_id; corrected going forward. |
experiments/06-live-llm/evidence/run-005/ |
| 07 Participant substitution | Deterministic and live participants can use one contract. | OCI Redpanda run 7234b30169c84ecd8c353e7fb662eebc; two deterministic and one Luna participant. |
One input caused three proposals, including Azure/Luna. | PASS | Concurrent consumption, not ordered hand-off or multi-provider substitution. | experiments/07-provider-hot-swap/evidence/run-002/ |
| 08 Competing consumers | Same-group consumers share work with measurable behavior. | Prior three-worker batch run. | Empty worker logs; no measurement JSON. | INCONCLUSIVE | No readiness, assignment, or performance evidence. | experiments/08-parallel-scaling/RESULT.md |
| 09 Audit trail | Causal records can generate an audit projection. | OCI Redpanda run 30a8737c586645ecad705c9959492a4b. |
All event IDs and causal links appear in audit.mmd. |
PASS | One stream; no policy/governance evaluation. | experiments/09-audit-trail/evidence/run-002/ |
| 10 End-to-end SDLC | Event participants can reach a terminal SDLC event. | OCI Redpanda run 44dd6f87389f453e9d4d4ed3fd0fd9c9. |
Ten events with causal parents; terminal DocumentationCreated. |
PASS | Deterministic participants and simulated approval. | experiments/10-end-to-end-sdlc/evidence/run-003/ |