ScriptMasterLabs settlement and coherence-gate observations

First-party observations from ScriptMasterLabs systems. The schema has **16 fields: 9 required and 7 optional**. Missing optional values are omitted, not replaced by zeros, empty strings or invented observations. `source` is mandatory on every row; a batch-level sourceDeclaration does not replace it. `jev_act` is the decision model's literal text answer, such as `act`, not a numeric score.

## Method and interpretation
Settlements were reported as checked against chain receipt status and block when originally recorded. A chain receipt supports transaction execution; it does not independently establish application replay behavior or the gate decision. Coherence is the deterministic value |sum(w*d)| / sum(w) over independent paths; the reported gate passes at >= 0.6 with non-flat consensus. A `jev_act` result is model output, not a measured guarantee of safety or success. Probes are individual unauthenticated requests timed by the prober. `latency_s` is seconds for one measurement, not a mean, percentile or uptime statistic. `amount_usdc` is decimal USDC, not atomic units; `coherence` is unitless in [0,1]; `block` is an integer chain block.

## Vantage and self-tests
These are first-party, one-vantage observations. Results do not establish availability from other regions, networks or clients, and a single failed probe does not establish an outage. The original SML vantage location/host was not supplied in the retrieved samples; do not borrow WITAN's South Korea vantage. Record the actual vantage in `source` or `detail` when available without adding a seventeenth field.

Both available sample events are self-tests between ScriptMasterLabs-controlled wallets (`self_test: true`). They demonstrate the reported tested behavior, not independent customer demand, external revenue, broad reliability or third-party adoption. Preserve this disclosure on every applicable row. Do not default unknown self-test status to false.

## Timestamp policy
Use a full ISO 8601 UTC timestamp ending in Z when the original event/observation time and timezone are available, converting known offsets to UTC. Do not substitute upload time, email time or midnight for an unknown observation time. The two retrieved samples supply only 2026-10-25 and 2026-10-12; retain those date-only strings and explicitly disclose their precision until the original logs are recovered. The email provenance time is 2026-10-27T06:52:37Z and is not either event's observed_at. Jongwon's review allows strings and recommends full UTC times where available. No full observation timestamp was recoverable here.

## Derived counts
Rows with `kind: aggregate_count` are derived counts over an identified source set, not single observations or additional measured events. Their `source` and `detail` must identify the contributing records, selection/window, counting rule and count/unit; disclose overlaps and self-tests. Never add aggregate counts to their constituent rows as if they were new events. This 16-field contract has no dedicated numeric count column; preserve the count and derivation in `detail`, not in amount_usdc, coherence or latency_s. No aggregate row was reconstructed from the category totals alone.

## Scope and provenance
The proposal reports a 30-record set: 7 x402_settlement, 8 coherence_gate_check, 3 escrow_lifecycle, 9 probe, 1 jury_verdict and 2 aggregate_count. This package contains only the two actual sample rows retrieved from that proposal, not the complete 30-record bank. The other 28 rows and original logs were unavailable. These samples are transcriptions of the first-party email report, not independently reverified chain/probe measurements. Their source values say so explicitly. Source references identify evidence; they do not imply the private email is publicly accessible.

The Arc row describes the partial capture step only: 0.25 authorized, 0.01 captured, 0.24 remaining at that step. Later void/claim and zero-final-escrow reports are separate lifecycle evidence and must not overwrite that historical state. The Base replay claim applies only to the tested identical replays reported in the source.