Engineering Delivery Dashboard - Visibility Before Burn-Down
Three layers before you celebrate velocity
Burn-down charts without a causal chain mistake activity for progress. Sprint reviews argue from memory. A green gadget next to a broken user path is leadership theater - until you can interrogate why defects return and who resolves them.
Here's what this piece covers: the visibility-before-burn-down thesis, what the 2026-07-28 Confluence metrics draft locked, the three-layer operating model, the named pattern, and three habits for staff meetings. Same spine as the values treatise: see first, then burn down.
Part of the Field Notes: agentic delivery series.
1. The leadership problem
Dashboards without root-cause analysis become activity theater. Teams invent local widgets. Directors compare incompatible definitions of "throughput." Nobody can answer whether regression is new work or the same class returning.
Wrong: "Track velocity."
Right: "Jul 2026 delivery model: Jira layer for shared visibility → recurrence pipeline for chronic targeting → PR analytics for resolution velocity - one causal chain."
The anti-pattern is green dashboard, broken journey - KPIs disconnected from ship criteria on the master flow.
2. What the docs show
docs/confluence-agentic-delivery-metrics-section.md (2026-07-28) is the leadership reference: verified vs planned vs team-measured confidence labels, plus the engineering delivery dashboard section.
Requirement ticket for the Jira layer landed Done in Jun 2026 (internal productivity-dashboard ticket - public framing: metrics spec written; gadgets built). Dashboard 12039: productivity, throughput, in-progress time, created vs resolved, regression vs new, repeat regression.
Agentic footprint called out in the same draft (verified from repos): 53 standards, 20 framework skills, ~47 SPEC contracts, Graphify on 4 repos (~42K nodes). SIT catalog 63 workflows; E2E tag registry 76 tags. Those numbers are adoption inventory - not proof of uniform use (adoption % still a gap).
3. What I built: three layers
| Layer | What it is | Role in burn-down |
|---|---|---|
| Jira dashboard | Operational gadgets every sprint | Visibility - shared view |
| Recurrence pipeline | Clustering over 3,403 issues, git-linking, chronic detection | Leverage - root-cause targeting (article 20) |
| GitHub PR analytics | Merge cycle, reviews, throughput | Resolution velocity (CLI → HTML dashboards) |
Not three unrelated initiatives - one system. Visibility without recurrence is a vanity wall. Recurrence without PR velocity cannot staff the fix. PR metrics without a shared Jira view restart the argument every sprint.
LAYERS = ("jira_visibility", "recurrence_leverage", "pr_resolution")
def ready_to_burn_down(org):
return all(org.has_layer(name) for name in LAYERS)
def staff_meeting_agenda(org):
return [
org.dashboard.snapshot(),
org.recurrence.top_chronic(n=3),
org.pr_metrics.resolution_delta(),
]Confidence discipline from the metrics table: label verified (repo/Jira), planned (PO KPI), team-measured (e.g. cycle-time compression - confirm baseline before interviews). Do not mix them in one headline.
4. Named pattern: visibility before burn-down
Visibility before burn-down - three layers on one causal chain; KPIs leaders can interrogate before celebrating velocity.
| Artifact | Who uses it | Decision it supports |
|---|---|---|
| Dashboard 12039 | EM, directors | Shared health every sprint |
| Recurrence summary | Leads | Which chronics to staff |
| PR HTML/TSV | Eng managers | Who resolves what, how fast |
| Confidence column | Anyone citing numbers | Verified vs planned vs gap |
What broke: citing a stale open-defect count from a July export in an interview. Live queue links beat screenshots. Uniform adoption % across outcome teams was not quantified - inventory ≠ habit.
5. What I'd do as a lead
Three habits - steal one this month:
- One dashboard per org. Do not let every team invent widgets with different definitions of "throughput."
- Review dashboard + top three chronics in the same meeting. Visibility without targeting is theater; targeting without a shared view is tribal knowledge.
- Refresh live defect-queue links. Never cite a frozen count when the filter still exists.
Further reading
- Field notes: agentic delivery - series intro
- Metrics spine: Script to CLI · Wrong metrics · README as contract · Auditing agent analytics · HTML dashboards
- Prior: Outcome-aligned teams · Chronic regression
- Next: Blocker matrix
- Planning anchor: Confluence metrics draft · 2026-07-28
