Skip to content

RELEASE ENGINEERING / OWNED TESTNET

What earns a release
the right to move forward?

A healthy deployment is the beginning. Follow one candidate through useful traffic, controlled failures, recovery, and the evidence behind its next step.

A documented lab, with bounded claims. No live status feed.

THE RELEASE QUESTION

  1. 01
    Know the candidateSource, digest, rendered revision.
  2. 02
    Observe useful workPaid traffic before injection.
  3. 03
    Challenge the boundaryPostgreSQL. Redis. Signer.
  4. 04
    Account for the resultScore, recovery, cleanup, eligibility.

The application creates short URLs, optionally through an x402 payment. The release platform asks whether this exact candidate can do useful work, tolerate the tested failures, recover, and leave its experiment clean. Source, image, rendered revision and run identity connect those decisions.

Two layers connect URL request responsibilities with publication, Kargo, Argo CD, gate runner and measurement responsibilities.
One application, two connected planes: useful requests and release decisions.

The client uses FastAPI for durable creation and redirects. PostgreSQL owns mappings; Redis accelerates established redirects with database fallback. The paid client obtains signer authorizations, while the application calls the facilitator for verification/settlement before persistence. CI publishes and signs selected image bytes. Kargo selects Freight/chart inputs and renders environment output; Argo CD reconciles it. The staging runner performs paid preflight, serial faults and cleanup, and the scorer consumes returned Prometheus evidence. Staging success grants normal eligibility; prod-like promotion remains explicit.

Source-derived explanation; not a live observation.
Read the diagram step by step
  1. FastAPI creates and resolves URLs; PostgreSQL owns durable mappings and Redis accelerates lookup.
  2. The paid client uses the signer; the application separately verifies and settles through the facilitator.
  3. CI publishes selected image bytes and records digest identity.
  4. Kargo selects and renders; Argo CD reconciles the environment output.
  5. The gate exercises the candidate and joins measurements with workflow/load outcome and cleanup.
  6. Staging success grants normal eligibility; an operator still requests prod-like promotion.
Public implementation sources

app/main.py · app/payment.py · kubernetes/jobs/scripts/loadgen.js · .github/workflows/build-push.yaml · kubernetes/kargo/warehouse.yaml · kubernetes/kargo/stage-staging.yaml · kubernetes/apps/applicationset.yaml · kubernetes/chaos-experiments/orchestrate.sh · kubernetes/chaos-experiments/score_experiment.py · kubernetes/kargo/project.yaml

Engineering work: decisions with consequences

Section titled “Engineering work: decisions with consequences”

Follow the choices that make the result trustworthy: rejecting missing measurements, separating cleanup from scoring, and surviving cache loss without inventing a stronger dependency guarantee.

Explore all five engineering cases, including controller ownership and the payment–database consistency boundary. Each case identifies source, tests, historical observations and what remains unproven.

Satyam Agnihotri

DevOps & Cloud Engineer · devSatym

I focus on Kubernetes, cloud infrastructure, and secure delivery workflows. This handbook explains how a release system works together, why its decisions were made, and what its bounded evidence demonstrates.

The full engineer profile and contribution boundaries distinguish source-backed implementation and test contributions from claims about sole original design or who personally ran a past campaign.

The retained records include a successful release and a deliberately blocked candidate. These are different campaigns with different identities. A later screenshot capture date never moves an earlier observation into the fresh release window.

observedpass

October 3 release verification

Observation: 2026-10-03 · staging 14:37:29–14:46:28 UTC

The public verification report records paid preflight, three dependency faults, recovery and cleanup. Fresh raw bundles are privately retained; the screenshot is a reviewed supporting view.

Read the supporting record
historicalfail

A degraded candidate stayed ineligible

Observation: 2026-10-02 · preflight 05:08:44–05:10:11 UTC

The historical negative candidate could not prove paid traffic within 75 seconds. Verification failed before injection. This result is independent of the October 3 release.

Read the supporting record

REVIEWED EVIDENCE / CAPTURE 11

Did this staging candidate complete verification?

The reviewed Kargo capture shows the fresh October 3 staging verification as Successful. It does not expose the privately retained raw metrics bundle.

Observed window: 2026-10-03 14:37:29 UTC–14:46:28 UTC

Exact AnalysisRun title, green Analysis passed summary, and green chaos-verdict plus readiness; selected Freight history card shows hoping-warthog. Kargo's table value 853b0a1 is the verification-history ID, live-mapped to this Freight

Original available · recorded screenshot · numerical verdict not assessed from pixels

Campaign identity and provenance
Captured
3 Oct 2026, 15:08:49 UTC
Source identity
v1.0.0 / 3b70835335462f1b6f9b1dd17ab20d1108724f89; chart source 0a98c08ccab5fa7b05efcecf877c714e5c89f025
AnalysisRun
staging.01m4135pfqh94e76vzzn7fpbs4.d58be07 — Successful; verification ID 853b0a11-2953-4801-835f-5caba72e314c
Image / OCI identity
sha256:ab88d89c20ccf1a79b9f47f90aa417ac0f54d7852c39b444b7fb788975fdd6a1
Rendered revision
64d2ddedb1493e0c59aef9ecad0ad0a2ff4196cf
Original SHA-256
ace8e91b625aa69414f756762f7f5cf3c005bcfe0e432e760c324caebe6eff83

Inspect the negative-case capture and its identity, or browse all 34 original reviewed screenshots. The gallery keeps availability, observation role and verdict interpretation separate.

Three failures, three different hypotheses

Section titled “Three failures, three different hypotheses”

Readiness establishes a starting condition. Injection establishes that a fault occurred. Measurements establish bounded behavior. Cleanup establishes a separate operational outcome. Normal staging verification grants downstream eligibility; an operator still requests prod-like promotion.

Learn why a green score can still be insufficient, or trace the payment settlement before database persistence.

Resilience Gate documents one owned Radius testnet lab. Environments share a cluster. prod means production-like testnet, with no claim of public-production availability, mainnet settlement, custody, disaster recovery or broad chaos coverage. The October 3 release, October 4 lab retirement, and later documentation/license publication are separate events.

Meet the engineer and read the contribution boundaries. Related work includes the GCP Security Architecture Handbook and Azure AKS GitOps Observability source.

Inspect the limitations, integrated system design, and website implementation report for the exact boundaries and verification state.

Open original image

Maintained by Satyam Agnihotri · DevOps & Cloud Engineer