A Release You Can Verify

Inspect a release manifest, acceptance checklist, and rollback plan. See how a winning change becomes a controlled handoff.

OpenTour illustrative sample

TYPE

Illustrative release packet

SCOPE

One invoice workflow

TIMELINE

After protected evaluation

SOLUTION

Acceptance & rollback

08

08

Handoff records

02

02

Distinct approval gates

SAMPLE 03 / RELEASE RECORD

Evidence first. Authorization next.

01 / EXACT CHANGE

Candidate package

Route + artifact version

Recipe + evidence lineage

Model, adapter, or repair

02 / ACCEPTANCE

Pending evidence

Protected quality + retention

Critical, cost, latency gates

Named acceptance owner

03 / PRODUCTION

Not authorized

Runtime + fallback rehearsal

Release owner + stop policy

Recorded rollback path

SAMPLE STATUS: HOLD / No customer evaluation, approval, or production release is implied.

OpenTour mark

An evaluation pass supports acceptance. Production traffic still requires the customer’s release authority and a tested fallback.

The OpenTour standard

Illustrative delivery package

01 // OVERVIEW

A decision your team can reproduce

The selected change may be a model, an adapter, or an application repair. The handoff ties that exact change to its data, evaluation results, operating limits, and release decision.

This sample is deliberately on hold. It shows the structure of a release packet without fabricating results, approvers, or deployment dates. Real values are recorded only when the corresponding work has been completed.

THE CHALLENGE

A completed training run is not a release decision. Missing evidence, incompatible fallback, or absent authorization keeps the change on hold.

02 // SAMPLE ARTIFACT

Inspect the release scorecard

Each gate links to a reproducible result or an accountable decision. Quality acceptance and permission to expose traffic are separate. The illustrative statuses below are pending, not passed.

GATE

REQUIRED EVIDENCE

SAMPLE STATUS

GATE

Quality & uncertainty

REQUIRED EVIDENCE

Paired results vs. the best baseline; useful effect and critical errors.

SAMPLE STATUS

Pending / HOLD

GATE

Retained behavior

REQUIRED EVIDENCE

Unchanged cases, coverage, and abstention within agreed limits.

SAMPLE STATUS

Pending / HOLD

GATE

Runtime & economics

REQUIRED EVIDENCE

Exact artifact reload, latency, actual costs, compatible fallback.

SAMPLE STATUS

Pending / HOLD

GATE

Release authority

REQUIRED EVIDENCE

Customer approval, observation owner, stop rules, rollback rehearsal.

SAMPLE STATUS

Not authorized / HOLD

03 // HANDOFF

The handoff, including the way back

The delivery index includes eight records: behavior contract; data and permission manifest; reproducible recipe; exact model or repair artifact; evaluation packet; deployment and fallback record; decision and economics; closeout and control inventory.

Before an authorized rollout, rehearse artifact loading and a fallback compatible with the current policy. Record stop conditions, observation owner, outcome window, and rollback steps. A hold never promotes.

The first pilot is 30 days from readiness with service fees waived and agreed compute capped. The next real change is a separately scoped paid continuation, measured against the same reliability bar.

MANIFEST / ILLUSTRATIVE FILENAMES

01 / CONTRACT & DATA

behavior-contract.json
evidence-manifest.json

Schema, rubric, source families, permissions, and split versions.

02 / RECIPE & CHANGE

recipe.yaml
artifact-manifest.json

Exact model or repair version, dependencies, hashes, and reload path.

03 / TEST & FALLBACK

evaluation-index.json
rollback-plan.md

Paired cases, scorecard, runtime checks, stop rules, and fallback.

04 / DECISION & CLOSEOUT

decision-record.json
closeout-inventory.md

Approval state, costs, owners, retained assets, and open limitations.

Rollback rehearsal: resolve the approved fallback → verify compatibility with the current policy → restore the prior route → check sentinel cases and log the decision. Exact commands and stop thresholds belong to the agreed customer runtime.

[READY TO IMPROVE YOUR AI WORKFLOW?]

IMPROVE WHAT YOU’RE

BUILDING

NEXT

Book a focused workflow review to assess recurring errors, your model route, and the records needed to verify a meaningful improvement.

↳ Prefer to send an enquiry?

Tell us about one live workflow, your open-weight model, and where errors create rework. We’ll assess the fit and define a practical starting point.

START WITH [ONE REPEATABLE WORKFLOW]

*Share a workflow overview here. Please do not include customer records or sensitive data.

[READY TO IMPROVE YOUR AI WORKFLOW?]

IMPROVE WHAT YOU’RE

BUILDING

NEXT

Book a focused workflow review to assess recurring errors, your model route, and the records needed to verify a meaningful improvement.

↳ Prefer to send an enquiry?

Tell us about one live workflow, your open-weight model, and where errors create rework. We’ll assess the fit and define a practical starting point.

START WITH [ONE REPEATABLE WORKFLOW]

*Share a workflow overview here. Please do not include customer records or sensitive data.

[READY TO IMPROVE YOUR AI WORKFLOW?]

IMPROVE WHAT YOU’RE

BUILDING

NEXT

Book a focused workflow review to assess recurring errors, your model route, and the records needed to verify a meaningful improvement.

↳ Prefer to send an enquiry?

Tell us about one live workflow, your open-weight model, and where errors create rework. We’ll assess the fit and define a practical starting point.

START WITH [ONE REPEATABLE WORKFLOW]

*Share a workflow overview here. Please do not include customer records or sensitive data.