35 jurisdictions, 64+ consecutive runs, not one missed day
How Prosody Labs built a self-updating intelligence dashboard for Opportunity Zone 2.0 — and what it shows about working fluently with AI. Live, public, operating today.
At a glance
| Engagement type | Internal R&D / capability demonstration |
| Domain | Federal & state economic-development policy |
| What we built | An autonomous, daily-updating web dashboard tracking state Opportunity Zone nomination processes |
| Coverage | 35 jurisdictions monitored at time of writing (36 as of June 2026 — and growing) |
| Operating record | 64+ consecutive daily runs, fully unattended |
| Human time per week | Minutes, spent reviewing — not researching |
| Live at | prosodylabs.prosodyconsulting.com |
The challenge
When the federal government re-authorized Opportunity Zones — “OZ 2.0” — it set a national standard but pushed the real work down to the states. Each state was tasked with nominating eligible census tracts, and each was free to design its own nomination process: its own deadlines, its own eligibility requirements, its own portals and points of contact.
For investors, developers, and the advisors who serve them, that created a genuinely hard tracking problem. There is no single national clearinghouse. The relevant information lives in fifty different places — governors’ press releases, state economic-development agency pages, legislative bulletins — and it changes constantly as states announce, open, extend, and close their windows. Miss an internal state deadline and you miss the chance to influence which tracts get nominated at all.
Monitoring this by hand means a person revisiting dozens of government sites every week, noticing what changed, and re-assembling it into something comparable across states. It is exactly the kind of work that is too important to skip and too tedious to do reliably by hand.
The approach
Rather than treat AI as a one-off research assistant, we treated it as an operator. The goal was not “help me look this up once” but “own this beat and keep it current indefinitely.” That reframing — from prompt to process — is the heart of what we mean by AI fluency.
The system runs as a scheduled agent every morning. On each run it works through a defined research discipline, updates a single source-of-truth dashboard, preserves an audit trail of what it could and couldn’t access, archives a dated snapshot, and publishes the result to the web. No one kicks it off. A person’s only job is to glance at the changes.
The discipline that makes the output trustworthy is official sources first. The agent is instructed to check each state’s own government and agency sources before falling back to secondary reporting, and to be explicit about provenance. When a source can’t be reached — a government site blocks automated access, a link rots, a page times out — the agent doesn’t quietly drop it. It writes the URL and the reason to a running log so the gap is visible and can be revisited, rather than silently becoming a blind spot.
What we built
The deliverable is a single, self-contained dashboard — one searchable, sortable HTML page — that a non-technical user can actually use under deadline pressure.
Each state is a row, compared across the columns that matter for action: current status (open, closed, pending), the internal state deadline, the governor’s submission deadline to Treasury, the number of eligible tracts, a plain-language summary of requirements and process, direct links to official resources, and a last-updated stamp. A visitor can type a state name to filter instantly, or click any column heading to sort — for example, to see every state with an open window ranked by how soon it closes.
Three design choices turn a static table into a living instrument:
- Change is visible. New and edited entries are flagged so a returning user sees what moved since last week instead of re-reading the whole board.
- History is preserved. Every run writes a dated archive — 64+ of them so far — so the record of how each state’s process evolved is never overwritten. That archive doubles as the system’s own change history.
- Gaps are honest. The blocked-URL log is a deliberate transparency feature. It is far better for a policy tool to say “we couldn’t reach this source on this date” than to imply coverage it doesn’t have.
Around the dashboard sits the quiet infrastructure of a real product: web analytics so we can see usage, automatic publishing to the live Prosody Labs site on every run, and a weekly email to stakeholders that goes out only when something material actually changed — signal, not noise.
How it works — the AI fluency part
What makes this a fluency demonstration rather than just a script is that the judgment lives in instructions, not in brittle code.
It’s an agent on a schedule, not a scraper. A traditional scraper breaks the moment a state redesigns its website. This agent reads pages the way a researcher would — adapting to new layouts, distinguishing a real announcement from boilerplate, reconciling a press release against an agency page. The daily cadence (a 5 a.m. run) means the dashboard is current before anyone needs it.
The rules of good research are written down and enforced every run. Official-source-first, log-what-you-can’t-reach, flag-what-changed, archive-before-overwriting — these are persistent operating instructions the agent follows autonomously, the same way you’d onboard a diligent analyst. The standard doesn’t degrade on run 64 the way human attention would.
It manages its own failure modes. Government sites block bots; links die; pages stall. Instead of crashing or producing a silently incomplete board, the system records each failure with a timestamp and reason and carries on — leaving a clean trail for review and retry.
It closes the loop to publication. Research, structuring, quality-checking, archiving, web publishing, and stakeholder notification are all stitched into one unattended pipeline. The output isn’t a chat transcript someone has to act on; it’s a maintained website and an inbox-ready summary.
The deeper point: the skill on display isn’t prompting a model for a clever answer. It’s designing a durable workflow — deciding what the standard is, what to do when sources fail, what to preserve, and who to tell — and then trusting a capable agent to run it. That’s the difference between using AI and being fluent with it.
Results
- 35 jurisdictions tracked at time of writing — 36 as of June 2026 and growing — compared on a consistent, decision-ready set of fields. See the live dashboard for the current count.
- 64+ consecutive daily runs with no manual intervention, building a continuous record of how state processes evolved.
- Weekly research effort collapsed from hours to minutes — humans review changes instead of hunting for them.
- A self-documenting audit trail of every unreachable source, so coverage gaps are known rather than hidden.
- A live, public dashboard that updates itself and notifies stakeholders only when it matters.
See it live, today, at prosodylabs.prosodyconsulting.com.
Why it matters for our clients
This project is a self-funded illustration of a pattern we can deploy against many problems: any domain where the truth is scattered across dozens of changing sources and someone is currently paying with their time to keep up.
Regulatory and policy monitoring, competitive intelligence, grant and RFP tracking, compliance-deadline watching, multi-vendor or multi-jurisdiction status boards — these all share the same shape. Prosody Labs designs the standard of quality, encodes the judgment, and hands the repetitive vigilance to an agent that doesn’t get bored, doesn’t skip a week, and tells you honestly what it couldn’t see.
If your team is spending real hours each week reassembling the same picture from the same scattered sources, that’s a candidate for this approach.
Built by Prosody Labs. The Opportunity Zone 2.0 Monitor runs daily and is published at prosodylabs.prosodyconsulting.com. To discuss building one for your domain, request a consultation.