# Raven System Activity Update — public methodology

Release: 2026-09-04.public.v2  
Data cut: 4 September 2026, 12:14 UTC  
Canonical publication: https://ravensystems.group/system-activity/2026-09-04/

## Scope and evidence

A scoped public record of what Raven measured and processed in its own environment — not a customer-performance claim.

The audit window is 1 February 2026 through the data cut. Retained source records span 204 UTC dates. The earliest producer-bound measurement in this scope is 14 February 2026, 13:24:31 UTC; that is not a claim about the first-ever Raven measurement. Dates with records do not establish continuous operation.

Measurements were recovered across a multi-host Raven-owned environment. Two hosts were directly audited and retained archives from a third host were assessed separately.

The selected population contains 351 files and 229 byte-distinct contents, representing 4.86 GB of physical retained data with copies included. Distinct file contents may overlap. A conservative lower bound of 1,949,874 distinct payloads applies to one dataset, not to globally unique events.

At least 11 confirmed collector families and 10 measured domains are represented. These counts do not certify a complete deployment census or equal domain coverage. The domain inventory provides only the public domain names.

## Bounded LTE/fiber comparison

Two anonymized same-destination ICMP comparison groups are published as separate populations. The accompanying JSON retains existing audited values for N, median, P95, maximum and the median paired LTE-minus-fiber difference. The website rounds RTT display to three decimal places; the JSON preserves audited precision. No measurement was recalculated for this public revision.

Both comparison groups belong to the same 501 acquisition cycles recorded on 8–10 June 2026. They are not additive. N counts valid capture-level mean RTT observations, not individual packets or independent incidents. Historical clock uncertainty limits simultaneity claims. These bounded observations do not establish continuous uptime or general network performance.

The paired difference is computed in the frozen audit over valid pairs; it is not the difference of the two displayed medians. A missing valid RTT observation is excluded from the RTT denominator, not converted to zero. This publication does not claim a general fiber-versus-LTE performance advantage or incident lead.

## Scoped Runtime activity

One bounded historical population contains 624 validated source references associated with 208 generated envelopes. A separate retained processing population contains 2,000 inputs and 2,000 normalized outputs. Two separate health-evaluation populations contain 24,482 and 121,522 retained snapshots. These populations are not summed, do not form an end-to-end funnel and are not counts of unique incidents or successful decisions.

These observations support scoped processing activity, not production readiness or verified decision outcomes.

## Temporal evidence boundary

Retrospective threshold chronology was calculated using a frozen, past-only method. Exact detector parameters remain in the private evidence package.

Reconstructed threshold chronology was observed, but no qualifying evidence-ready, recommendation or approved-decision lead was established.

Prior lead-time figures were re-audited and excluded where they were synthetic, non-reproducible, cross-target, cross-path or methodologically invalid.

## What this proves

- Real retained measurements.
- Multiple hosts in a Raven-owned environment.
- Multiple measured domains.
- A bounded LTE/fiber comparison.
- Scoped Runtime processing.
- Auditable separation of signal and decision claims.

## What this does not prove

- Customer validation.
- ORLEN or car-wash performance.
- Production readiness.
- Recommendation lead.
- Approved-decision lead.
- Downtime reduction.
- ROI.

## Public artifact integrity

The public package contains five content artifacts and a SHA-256 manifest. The manifest covers those five artifacts; it cannot hash itself. The JSON includes hashes of the four independent companion content artifacts, excluding itself and the manifest to avoid circular dependencies. All six published files are checked against the release copy during deployment verification. File hashes verify bytes, not the truth of a claim or completeness of the private evidence.