sdr-visualizer

Static-output visual catalog generator for Adobe Customer Journey Analytics (CJA) and Adobe Analytics (AA) implementations. Consumes JSON snapshots from cja_auto_sdr and aa_auto_sdr and produces a single self-contained HTML file with:

• A searchable, filterable component catalog (the primary view)
• An interactive force-directed reference graph
• Per-segment anatomy diagrams that make deeply-nested segments legible
• Per-calculated-metric formula trees with click-through to referenced metrics
• Snapshot-to-snapshot Changes and multi-snapshot Trend views

Live examples: CJA report · AA report

The output is one HTML file: no server, no consumer-side build step, and no CDN dependencies. Its JSON, CSS, JavaScript, and D3 runtime are embedded, so it opens in a modern browser without an internet connection and makes no network requests.

Offline does not mean cleared for distribution. A report can contain implementation and component names, descriptions, segment and calculated-metric logic, owners, identifiers, timestamps, and source paths. Treat it as derived from the source snapshot. Move, email, post, or attach it only in authorized locations and in accordance with your organization's confidentiality and data-handling policy.

Install

Install from PyPI with uv:

uv tool install sdr-visualizer

Or with pip:

pip install sdr-visualizer

For development, run from a clone:

git clone https://github.com/brian-a-au/sdr-visualizer
cd sdr-visualizer
uv sync
uv run sdr-visualizer --help

Quickstart with a saved snapshot

Saved snapshots are the simplest and most reproducible input. You do not need either upstream generator installed to visualize a JSON file you already have.

# From a snapshot file
sdr-visualizer path/to/snapshot.json

# From a directory of snapshots (uses the most recent)
sdr-visualizer path/to/snapshots/

# Compare against an earlier snapshot: adds a Changes view to the report
sdr-visualizer snapshot_new.json --compare-to snapshot_old.json

# Chart evolution across a directory of snapshots: adds a Trend view
sdr-visualizer ./snapshots/ --trend

The output lands at ./visualize-{instance_id}-{timestamp}.html by default. Open it in a browser — that's the whole experience.

Standard input is also supported when another process already emits a complete compatible snapshot:

some-snapshot-command | sdr-visualizer -

Live CJA

Live CJA mode requires the separate cja_auto_sdr executable. Follow that project's current installation, configuration, and authentication instructions, then confirm cja_auto_sdr is on your PATH. Its generator may require a newer Python version than sdr-visualizer's Python 3.11 minimum.

sdr-visualizer --dataview dv_prod_web

sdr-visualizer runs the generator with JSON output, a temporary output directory, and --include-all-inventory, then selects the generated snapshot. The temporary source data is removed after the report is built. This is not a stdin pipeline: CJA's complete inventory is directory-backed.

Live AA

Live AA mode likewise requires the separate aa_auto_sdr executable. Complete its current setup and authentication steps, and confirm aa_auto_sdr is on your PATH. The generator may require a newer Python version than sdr-visualizer.

sdr-visualizer --rsid prod_us

AA live mode reads the generator JSON snapshot from stdout and embeds the supported normalized component catalog in the report. Retain the original generator snapshot when you need platform-specific details that the visualizer does not embed. For both live modes, the upstream repository is authoritative for credentials, permissions, and generator compatibility.

Useful flags

What's in the output

Every report has two base top-level views:

1. Catalog — a searchable, filterable, sortable table of every component. Click a row to slide out a detail panel with description, properties, references, and anatomy.
2. Reference graph — a force-directed view of every component and the edges between them; small implementations (under 20 components) use a static radial layout instead. Hover dims unrelated nodes; click opens the same detail panel; drag pins; pan/zoom.

Segment anatomy and calculated-metric anatomy are contextual detail content, not separate navigation destinations. They open from the Catalog detail panel. Segment anatomy renders nested containers and references; calculated-metric anatomy renders operations, operands, and metric references.

At most one conditional top-level view is added. With --compare-to, a Changes view appears, listing components added, removed, and modified relative to a baseline snapshot, with field-level before/after detail.

With --trend on a snapshot directory, a Trend view appears: sparkline charts of descriptive aggregates (component counts, orphans, undocumented components, reference edges) across the directory's snapshots, plus a per-interval change log. The window is capped at the 60 most recent snapshots.

A trend directory must hold snapshots of a single implementation. If it mixes CJA and AA snapshots, pass --platform cja|aa to select one (or point at a single-platform directory); without it the run stops rather than guess. If it mixes data views or report suites, the run stops as well. This mirrors --compare-to, which refuses both a platform and an instance mismatch, so neither view ever diffs unrelated inventories. To compare or chart across different data views or report suites on purpose (for example staging versus prod drift), pass --allow-instance-mismatch; the run then proceeds with a warning. Platform mismatches are always rejected. The report shown alongside the trend is the newest usable snapshot in the directory.

• Restorable report links — the catalog's filters, sort, view, and open detail panel are encoded in the URL hash. Within an authorized report location, copy the address bar to restore the same filtered view.

Performance budget

The output is CI-gated against the budgets in docs/PERFORMANCE.md. Build time and HTML size are enforced at every published tier (100 / 500 / 1,000 / 2,000 components). Browser-measured budgets are enforced at the 1,000-component tier (initial render < 1s, filter/search < 150ms) and the 2,000-component tier (< 2s, < 300ms), plus a 700ms cap on the graph view's main-thread block. These guarantees cover up to 8,000 reference edges; denser valid reports use an explicit graph opt-in and sit outside the published size/latency envelope. Functional browser tests cover Chromium and WebKit. A separate Chromium-only performance gate measures all four component tiers; this is not a timing guarantee for every branded browser.

Troubleshooting

Exit 0 means the report was generated. Exit 1 means a runtime failure such as an output-write error. Exit 3 means the input or invocation was invalid, including generator failures and timeouts. Exit 2 is not used.

Before sharing a snapshot, JSON sidecar, report, terminal output, or bug reproduction, redact customer names, component content, IDs, owners, source paths, credentials, and other organization-sensitive data. Prefer a minimal synthetic reproduction in a public issue.

Stability

From 1.0.0, semantic versioning covers the surface below. Anything not listed is internal and may change in any release.

CLI. The argument set: the positional path (snapshot file, snapshot directory, or - for stdin), --dataview, --rsid, --platform, --at, --compare-to, --trend, --allow-instance-mismatch, --output, --title, --exclude-orphans, --max-graph-nodes, --json, --quiet, --version. Removing or repurposing any of these is a major bump; adding flags is a minor one.

Exit codes. 0 success, 1 runtime error, 3 invalid input. 2 is never used.

The data payload. The JSON embedded in every report and the --json sidecar share one schema, published at docs/payload-schema.json (JSON Schema 2020-12) and validated in CI against every payload shape produced by the bundled fixtures. Removing or retyping a field is major; adding optional fields is minor. The segment_trees / formula_trees node internals are documented in the schema as loosely specified. Current-generator and private-corpus validation is a separate, recorded release gate; see docs/RELEASING.md.

Performance budgets. The tier table above is a guarantee, not a goal: loosening a budget is a breaking change; tightening one is minor.

Warnings (snapshot generator newer than the tested version; 5,000+ component reports) are informational and never make a valid snapshot fail.

Develop

Requires Python 3.11+ and uv.

uv sync                # Set up environment
uv run pytest          # Run tests (auto-generates the large fixture on first run)
uv run ruff check      # Lint
uv run ruff format     # Auto-format

uv run python scripts/generate_examples.py   # Regenerate examples/
uv run python scripts/perf_check.py          # Run the perf gate
uv run python scripts/check_markdown_links.py
uv run python scripts/check_workflow_policy.py
uv build --out-dir dist/packages
uv run python scripts/package_smoke_check.py dist/packages/

See also

• sdr-grader — deterministic, rule-based linter for the same input format.
• cja_auto_sdr — generates CJA snapshots.
• aa_auto_sdr — generates AA snapshots.

Documentation

• docs/ARCHITECTURE.md — module layout, one-way data flow, design principles.
• docs/ADAPTER_GUIDE.md — how the CJA and AA adapters work, and how to add a new platform.
• docs/PERFORMANCE.md — performance budgets and how they're enforced.
• docs/EMBEDDED_DATA_FORMAT.md — the JSON payload format embedded in the HTML output.
• docs/PRODUCT_CONTRACT.md — supported inputs, stable surfaces, limits, and compatibility policy.
• docs/RELEASING.md — candidate, corpus, repository-control, publication, and announcement gates.

Community

Contributions are welcome within the project's intentionally narrow scope. Read CONTRIBUTING.md before opening a pull request. Report security issues privately as described in SECURITY.md; do not put vulnerabilities or customer snapshot data in a public issue. Participation is governed by the CODE_OF_CONDUCT.md.

License

MIT — see LICENSE. The output bundles D3 v7, vendored under the ISC license; see THIRD_PARTY_LICENSES for the full notice.