# azdw: From a Delivery Question to Reviewed Action

## About This Video

This 60-second animated example follows one concrete workflow: investigating release risks across Azure DevOps, TFS, and GitHub, turning connected work-item data into findings, and approving two targeted source updates.

It is **one possible use of azdw, not a mandatory process or an exhaustive product tour**. Simple queries, non-AI reports, scheduled automation, AI-agent workflows, and more advanced custom applications are equally relevant. Future videos will explore these different approaches, both simpler and more advanced.

Watch with music: https://youtu.be/nRgN9x1Mr24

The local website preview is the same story without an audio stream. There is no voice-over. This is a conceptual animation, not a screen recording or a promise that all operations complete in 60 seconds. Organizations, people, projects, IDs, URLs, and findings are fictional.

## Files and Hosting

| File                     | Purpose                                                                                   |
| ------------------------ | ----------------------------------------------------------------------------------------- |
| `workflow-720p-loop.mp4` | Website preview: 1280x720, H.264/yuv420p, 30 fps, 1,800 frames, no audio stream.          |
| `workflow-1080p.mp4`     | Local archival/upload copy: 1920x1080, same timing, stereo AAC music. Not deployed to the web server. |
| `workflow.vtt`           | Fifteen timed English caption cues shared by both versions.                               |
| `workflow-poster.webp`   | Local graph-reveal poster displayed before playback.                                      |
| `workflow-guide.md`      | This timeline, interpretation, use-case, and business-value guide.                        |

Both versions last exactly 60 seconds. The story ends at 56 seconds, fades during 56-58 seconds, then holds black through 60 seconds. The music version also fades to silence. Both deployment preparation scripts remove only `workflow-1080p.mp4` from the generated `.dist` site. The source-tree archive remains alongside the other videos. Deploy the prepared `.dist` output, not the source tree, to enforce this exclusion. A source-tree development server can still serve the archival file, but the page never requests or links to it.

## Scenario and Visual Language

Maya, a release lead, is preparing **Lumen Release 3.2** for the **Cedar customer pilot**. Delivery work spans three workforce identity contexts, six systems/organizations, and fourteen projects or repositories.

- **Lumen Workforce (Entra ID):** Lumen Product (Azure DevOps), with Core Platform, Customer Portal, and Release Readiness; Lumen Delivery (Azure DevOps), with Customer Delivery; Lumen Legacy (TFS), with Device Gateway and Operations Console.
- **Vector Workforce (Entra ID):** Vector Engineering (Azure DevOps), with Identity Services and Event Services; adjacent Vector Open Tools (GitHub), with Identity SDK, Event SDK, Developer Docs, and Test Harness.
- **Bridgeworks Workforce (Entra ID):** adjacent Bridgeworks Integrations (GitHub), with Customer Connector and Deployment Templates.

Blue labels identify identity contexts, amber identifies systems, and pale teal identifies projects/repositories. Labels follow `Name (Type)`. GitHub is adjacent to an identity context: the picture does not mean GitHub repositories are hosted within Entra ID. TFS is the legacy, on-premises system in this example.

The azdw core represents a common access and relationship layer, not a mandatory new database or a migration of source ownership. The brain represents a user-configured AI model. The human, agent, report display, JSON document, and application represent different consumers of prepared information.

Connection lines mean **configured, authorized access**, not an automatic bypass of tenant boundaries or permissions. Amber motion highlights activity and relationships. Red marks blockers. Green after approval means an update was applied, not that the release risk disappeared.

## Detailed Timeline

### 00:00-00:04 | Start With a Business Question

**Shown:** “Which dependencies put our next release at risk?” appears against black.

**Meaning:** The starting point is a decision, not a report format or the mechanics of joining exports. The question defines the purpose of collection and analysis.

**Feature and value:** azdw supports question-led exploration alongside explicit commands. For release leads, portfolio owners, and consultants, the useful output is evidence for a delivery decision. This does not mean every question can be answered from incomplete source records.

### 00:04-00:08 | Reach the Connected Core

**Shown:** Maya (Release Lead), Web UI, the CLI alternative, azdw, and the source landscape emerge.

**Meaning:** Web UI is Maya's entry point here. CLI is an alternative, not another compulsory hop. Work stays in its Azure DevOps, TFS, and GitHub sources.

**Feature and value:** One underlying access layer can serve an interactive browser experience or an expert's command line. Teams can choose an appropriate interface without rebuilding provider access for each experience.

### 00:08-00:12 | Optional AI Interprets the Question

**Shown:** Configured AI helps shape the question into scope and an explicit query plan.

**Meaning:** AI interprets intent; it does not grant access, invent relationships, or replace deterministic collection. Configuring an AI provider and model is a prerequisite for this AI-assisted route.

**Feature and value:** Natural-language assistance can reduce the effort of expressing an investigation. Experts can still run explicit commands without AI. Separating interpretation from collection keeps the workflow understandable and reproducible.

### 00:12-00:18 | Query Across Source Boundaries

**Shown:** Queries fan out to fourteen tiles. Two representative work-item copies rise from each tile, twenty-eight in total. Local relationships appear.

**Meaning:** The rising objects are read-only representations, not records moved out of their source systems. These are illustrative counts, not performance benchmarks, throughput ceilings, or pricing limits.

**Feature and value:** Multi-source collection can reduce separate exports, spreadsheet joins, and provider-specific plumbing. Source identities allow results to be traced back to original work. Core collection uses explicit queries without AI tokens; AI interpretation and later reasoning may consume tokens.

### 00:18-00:24 | Refine Scope With Customer Commitments

**Shown:** Delivery Agent asks, “Include customer commitments,” through MCP. Some copies descend while relevant copies remain.

**Meaning:** MCP is the main agent route in this example, with CLI as an alternative. Fifteen copies remain selected and thirteen return visually to their source tiles. **Nothing is deleted at source.**

**Feature and value:** Reusable tooling can serve both people and agents. Customer context connects engineering dependencies to delivery obligations rather than only one team's backlog. Not every workflow requires both a person and an agent.

### 00:24-00:29 | Gather Context With Provenance

**Shown:** Selected copies converge near azdw in source-related clusters.

**Meaning:** IDs, source locations, and canonical references remain attached to the working context. This is neither a source migration nor a second set of authoritative records.

**Feature and value:** Reviewers can inspect findings and return to original items instead of relying on ambiguous titles or detached spreadsheet rows. Traceability helps identify the owning team, but is not by itself a formal audit or compliance guarantee.

### 00:29-00:35 | Follow Dependencies in Both Directions

**Shown:** Cross-source relationships resolve and brighten. Opposing pulses travel over links. The caption explicitly includes one-way hyperlinks in bidirectional traversal.

The primary chain is:

`CUST-204 Cedar pilot sign-in -> PORTAL-84 Enable federated sign-in -> identity-sdk#88 Handle signing-key rotation`

A second branch reaches:

`customer-connector#73 Map pilot identity claims`

**Meaning:** A canonical hyperlink recorded in one direction can be interpreted as a relationship explored from either endpoint. An item can be relevant because another item points to it, even without a reciprocal hyperlink. **No reverse source hyperlink is written merely to enable traversal.** The original reference and direction remain part of provenance.

**Feature and value:** Dependency closure and impact analysis can include inbound references that would be missed by following only outgoing links. “All dependencies” means following relationships available in the configured, accessible scope. It is not a promise to discover undocumented dependencies, inaccessible records, unsupported formats, or unconfigured systems. Relationships derive from supported mechanisms and canonical hyperlink interpretation, not AI invention.

### 00:35-00:40 | Reason Over Prepared Evidence

**Shown:** Graph context reaches Configured AI. Two issues turn red; the connector also shows an owner gap.

**Meaning:** The consistent finding is **2 blockers, 1 owner gap**. The blockers are `identity-sdk#88` and `customer-connector#73`. The connector lacks an owner. These are authored scenario facts, not evidence of an infallible AI risk detector.

**Feature and value:** Connected context supports cross-team reasoning by people or configured AI. Potential benefits include earlier dependency conversations, focused escalation, and clearer readiness discussions. Human review remains important. No quantified savings or risk reduction are claimed.

### 00:40-00:47 | One Finding, Different Consumers

**Shown:** A human response, Release Readiness report, JSON, and Release Desk application appear. REST/GraphQL API and the .NET library are integration options.

**Meaning:** These are alternative consumers and interfaces, not a required serial pipeline. Every output agrees:

```text
release=Lumen 3.2
customer=Cedar pilot
blockers=2
ownerGaps=1
reviewStatus=Needs review
```

**Feature and value:** A shared layer can feed reports, scripts, an internal portal, or agents. Decision-makers need not run the technical tooling themselves; experts can expose results in the right experience. The report, JSON, and form are illustrative representations, not screenshots of prescribed product templates.

### 00:47-00:48 | Request a Specific Action

**Shown:** Maya asks to flag the blockers for review.

**Meaning:** The requested change is narrow: mark two existing issues. It does not authorize unrelated modifications, resolve risks, or assign the missing owner.

**Feature and value:** Turning a finding into a precise follow-up action can reduce manual handoffs while keeping scope clear.

### 00:48-00:51 | Preview the Proposed Writes

**Shown:** An approval preview identifies two GitHub issues and the `release-review` label.

**Meaning:** Analysis and mutation are separate. The person inspects proposed targets and changes before authorizing writeback. This approval interaction belongs to this workflow; not every library/API call necessarily presents a human approval dialog.

**Feature and value:** Previewing changes supports informed control and avoids confusing a suggestion with an already-applied update. Hosts and automations must implement approval and authorization policies suitable for their use case.

### 00:51-00:52 | Explicit Approval

**Shown:** Maya approves the two label updates.

**Meaning:** Approval precedes mutation. No writes occur before this beat. A rejection would not authorize these writes, although that path is not animated.

**Feature and value:** A person can retain control while tooling performs repetitive execution. Approval does not override source permissions, license restrictions, or operational safeguards.

### 00:52-00:54 | Write to the Original Repositories

**Shown:** Two update signals travel through azdw to GitHub.

**Meaning:** `release-review` is added to `identity-sdk#88` and `customer-connector#73`. Preserved source identity targets the changes. This is not a detached-export update or a blanket graph modification.

**Feature and value:** Follow-up stays where owning teams work. This can reduce copying between analysis tools and backlogs, subject to caller permissions and applicable paid capabilities.

### 00:54-00:56 | State the Outcome Accurately

**Shown:** “Flagged for review” accompanies the unchanged counts: two blockers, one owner gap.

**Meaning:** The workflow exposed risk and initiated follow-up. It has **not** solved dependencies, assigned the missing owner, or declared the release ready.

**Feature and value:** A successful tool action is not automatically a successful business outcome. Teams must still review, prioritize, own, and resolve the underlying work.

### 00:56-01:00 | Fade and Loop Boundary

**Shown:** “Connected work. Clearer decisions.” accompanies the fade. Picture and music fade over 56-58 seconds; 58-60 seconds remains black and silent.

**Meaning:** The pause creates a deliberate boundary before the loop restarts. It contains no additional result, hidden action, or narration.

## Other Uses and Potential Business Value

| Use                       | Approach                      | Potential value                                         |
| ------------------------- | ----------------------------- | ------------------------------------------------------- |
| Everyday lookup           | Explicit CLI query against one source without AI. | A relevant, repeatable view without a complex workflow. |
| Recurring reporting       | Schedule collection and generate structured output or reports. | Less repeated export/merge work and more consistent inputs; freshness depends on run time and data configuration. |
| Dependency impact reviews | Explore hierarchy, dependencies, and configured hyperlinks. | Reveal cross-team implications and inbound references missed in a single-project view. |
| Customer delivery reviews | Combine commitments, engineering work, and ownership information. | Prioritize technical risks in the context of delivery obligations. |
| Consulting engagements    | Reuse commands and integration patterns with separately configured authorized client access. | Less repeated plumbing; no implication of mixing client data or permissions. |
| Internal applications     | Use REST, GraphQL, JSON, or the .NET library behind a purpose-built UX. | Deliver decision-ready information without requiring every user to operate the CLI. |
| AI-assisted exploration   | Configure a model to interpret questions and reason over prepared context. | Conversational investigation over an explicit collection layer; costs and behavior depend on model configuration. |
| Agent automation          | Expose authorized tools through MCP or CLI with appropriate approval policies. | Repeatable investigation and controlled follow-up within a broader workflow. |
| Governed updates          | Preview and approve targeted changes where supported. | Keep review actions at source and reduce handoffs without implying autonomous resolution. |

AI is optional for querying, relationship resolution, and non-AI reporting. A local model is a possible deployment choice, but the brain icon does not establish one. Data residency depends on every configured model, agent, host, and source endpoint. The film does not guarantee local-only processing.

## Prerequisites and Licensing

Configure connections, credentials, permissions, relationship interpretation, and any AI provider before running an equivalent workflow. Results are bounded by those permissions and settings. Interfaces expose host-appropriate capabilities; this illustration is not a complete endpoint-by-endpoint contract.

The multi-connection and writeback portions are **paid capabilities**, not an example of everything included in the Free tier. Simpler usage may omit AI, multiple connections, agents, custom applications, or writes. Consult current documentation and pricing for exact limits and availability.

## Website Playback and Privacy

The How it works preview uses only the local silent MP4, poster, and VTT. It loads when sufficiently visible, plays muted, pauses out of view or in a hidden tab, and respects manual pause. Reduced-motion or data-saving preferences keep the poster visible until explicit playback. Native controls become available when media loads; a labelled play/pause button is also provided.

The **Watch with sound on YouTube (new tab)** link is separate from playback controls: clicking the video does not unexpectedly leave the site. The adjacent disclosure explains that Google/YouTube may use cookies and track activity on the external site. The link uses `noopener noreferrer` and `no-referrer`, protecting the opener and suppressing referrer information; this does not make YouTube anonymous or prevent its own tracking.

The preview adds no embedded YouTube player, remote YouTube thumbnail, or pre-click YouTube request. The separate existing product tour and site cookie choices retain their own behavior. Accepting site cookies does not turn this preview into an embed.

## Verification and Production Notes

The production files were verified for duration, frame counts, H.264/yuv420p delivery, fast-start metadata, nonblank scenes, and sixty final black frames. The loop has no audio. The music version uses 48 kHz stereo AAC and measures approximately -18.02 LUFS integrated and -2.48 dBTP. Source PCM is silent from 58 seconds; the decoded AAC transform tail is below -80 dBFS.

The approved story and timing are preserved. Remotion/Three.js geometry and motion are deterministic; independent full-resolution ninety-frame samples match. Reproduction instructions live in the azdw development repository under `docs/marketing/assets/workflow/`.

Run `node --test scripts/workflow-video.test.mjs` from AppsWebPage for visibility, manual pause, preferences, disclosure, and packaging checks. Browser validation additionally covers actual desktop/mobile playback, local requests, captions, and the external link. This 16:9 film has fine-grained source labels that need fullscreen or a larger screen; it is not a separately edited portrait mobile video.

## One Logical Source of Truth, Not Another System of Record

Organizations want their work-management systems to provide a single source of truth. When all relevant work lives in one authoritative system, that system can fulfill this role. When work is scattered across multiple systems, each may be authoritative for its own records, but the organization effectively has multiple sources of truth. If they are poorly connected, no individual system provides the complete picture: dependencies, customer commitments, ownership, and delivery risks can fall between the boundaries.

**azdw does not introduce another source of truth. It connects the existing sources into one logical source of truth and provides flexible, advanced data analysis capabilities on top.** The original systems remain authoritative for their records. The logical view brings accessible work and its relationships together while preserving source identity and provenance, so people, agents, reports, and applications can analyze the connected picture without relocating ownership into a competing system of record.

In the film, the collection, refinement, and relationship-resolution sequence at 00:12-00:35 illustrates this distinction. The gathered copies are a working view of source records, not new authoritative records. The shared findings at 00:40-00:47 show how that connected view supports consistent analysis across different consumers. The approved updates at 00:52-00:54 go back to the original repositories, reinforcing where authority remains.

The potential business value is a more complete basis for decisions across teams and tools, less manual reconciliation of separate exports, and reusable analysis through queries, dependency exploration, reports, integrations, or optional AI. Teams can retain the systems that fit their work while gaining a connected analytical view across them.

Here, "one logical source of truth" means a unified view within the configured, authorized scope, not a guarantee that all underlying information is complete, current, or consistent. Missing links, conflicting records, and data-quality problems still require attention in the source systems; azdw does not automatically decide which conflicting claim is correct.