Did the Evidence Reach the Agent? Reconstructable Evidence Lifecycles for Reliable Experiments

Can an Evidence Lifecycle Observation Contract reconstruct which immutable source version was admitted, what ordered material crossed each model invocation boundary, and how claims and the terminal output were bound to those events, better than final records and generic traces?

Proposes an Evidence Lifecycle Observation Contract (ELOC) and a reconstructability ladder for experiment records. E19-A is executed deterministic diagnostics: 44 named checks across 15 constructed fixtures, plus a static collector audit classifying 4 of 18 requirements as present, 3 partial, and 11 absent at the required granularity. These are schema and implementation-audit results, not model performance. E19-B and H19-1/H19-2/H19-3 are proposed and untested. No new model benchmark, model call, human annotation, production inference change, or deployment.

• Constructed collision: terminal output plus aggregate usage cannot determine whether the corrective source was delivered. A narrow indistinguishability result over constructed records, not model performance.
• Static audit: 4 of 18 collector-contract requirements present, 3 partial, 11 absent at the required granularity.

• E19-A Deterministic evidence-lifecycle validator and static collector audit Executed (diagnostic)
• E19-B Reconstruction under frozen trajectories Proposed

• H19-1 B0 exposure ledgers increase exact evidence-exposure reconstruction over A1.
• H19-2 B1 claim lineage improves exact first-divergence localization over B0 without increasing false causal attribution.
• H19-3 An external-content mode using version, digest, length, locator and access receipt reduces routinely captured sensitive bytes while preserving identity reconstruction.

Proposed measurement contract only. No Noetis behavior, inference, routing, billing, risk value, or collection-gate change.

• Do exposure ledgers improve exact exposure reconstruction over final records and generic traces?
• Does claim lineage improve first-divergence localization without false attribution?
• Can hash, version and locator binding reduce captured sensitive content while preserving identity reconstruction?

Builds on Round 18, Round 17, Round 16, Round 13. Continued in Round 21, Round 20.