Method
Public records become structured intelligence through four core artifacts, seven stakeholder lenses plus synthesis, and one source pipeline. The civic record moves before the market reacts because the decisions are made on schedule, in public, and recorded. The observatory’s job is to read at the rate the record moves and surface the patterns the record makes visible.
A municipality across time. One city, one comprehensive plan, one planning board, one corridor. The dossier reads twenty-four months of decisions and surfaces the structural position — the regulatory regime, the development pressure, the philosophy of the board, the named patterns visible at the city scale.
A cross-municipal economic axis. The corridor reads decisions across boundaries — the same highway, the same aquifer, the same school district. It surfaces the organism that no single city dossier could see. US-27 South Lake reads four cities along one spine.
One harvested record turned into one citation-grade artifact. The meeting itself is the unit. Each agenda item passes through plain-English summary, signal extraction, source trail, delta from prior, and why-it-matters. The reading is what an analyst would write the morning after.
A named pattern observed across the corpus. Not a city, not a corridor — a recurring shape in the regulatory motion that deserves its own essay. The Self-Storage Canary, The Grandfather Window, The Six-Month Board Flip. Briefs read across the field; dossiers and corridor signals feed them.
Intelligence is not a synthesis with audience filters applied. It is a multidimensional entity projected onto cognitive surfaces. Each projection is a deliberate, crafted artifact. Seven stakeholder lenses surface the same evidence through different mechanisms — the developer asks one question of a parcel; the policy attorney asks another. The eighth projection — synthesis — holds the seven in superposition and names the dialectic.
| Lens | Reads for |
|---|---|
| Synthesis | The whole-picture view. Holds all positions in superposition. Names dialectics rather than flattening them. The default. |
| Developer | Entitlement risk. Where opposition is organizing. Which corridor has density-policy clarity for the next 18 months. |
| Investor | Pre-market signal advantage. Capital allocation timing. Where infrastructure unlocks basis-point edge. |
| Broker | Parcel-level narrative. The argument the broker can make to a client about this asset, this corridor, this scarcity. |
| Attorney | Precedent and argument-construction. Board-behavior history. Where state preemption binds. Which dispositions survive review. |
| Business | Site selection. Operational expansion. Where the corridor fits this build, what the entitlement risk is, what the timing looks like. |
| Resident | What is about to change in this neighborhood. Plain-English bureaucratic motion. Which decisions affect the streets the resident drives. |
| Civic leader | Cross-municipal coordination. Regional policy alignment. The fiscal sustainability of the growth choices a city is making. |
AI agents calling the MCP server may pass natural-language synonyms — legal,lawyer, counsel and regulator route to attorney; realtorand intermediary to broker; civic, government andofficial to civic-leader; homeowner and citizen to resident;operator and site-selector to business; builder to developer. Aliases resolve server-side to the canonical lens before the projection lookup.
Lens selection is single-select. The reader narrows when they want a focused frame; otherwise synthesis carries the unified narrative. Combinatorial multi-select fails as a model — stacking projections produces redundancy, not richness.
Every place, corridor, pattern and forward call carries a confidence grade. It is not a rating of the place, a recommendation, or a forecast. It states one thing: How much corroboration sits behind a reading. Assessed by the operator against the evidence standard below — not computed.
| Grade | Evidence standard |
|---|---|
| low | Single source; speculative; opportunity for misread. |
| medium | Two corroborating sources; pattern recognized but not crystallized. |
| high | Three+ corroborating sources; pattern named and exhibited; reasonable forecasting confidence. |
| very-high | Statutory or formal-record-grade evidence; deterministic from the corpus. |
Free-text period the reading speaks to (e.g. "12-36 months", "imminent", "long-term"). A stated expectation, not a forecast model.
Until August 2026 every record also carried a signal score from 0 to 100, and a direction —rising, stable, cooling. Both are gone, and the reason is worth stating rather than quietly redirecting.
The score was published as a composite of five weighted inputs. Nothing computed it. It was read out of the record's own front matter, where an author had written a number, and across 129 records it ranged from 56 to 95 with a median of 78 — a nought-to-a-hundred scale whose bottom half was unreachable, because nobody publishes a dossier and calls it a twelve. The direction was worse: it was defined as a five-point move across ninety days, and no record has ever stored score history, so the comparison it described could not be made.
This publication's claim is that it read the actual document. That claim does not survive publishing a measurement nobody took. The confidence grade above is a judgment, and it is labelled as one.
The machine-readable version of everything above is published at/schema/signal-model.json (version 2026-08-15), and this page is rendered from that same file — so the definition a reader sees and the one an agent pins to cannot drift apart.
| Step | What happens |
|---|---|
| 01 · Harvest | Agenda packets, staff reports, meeting minutes, ordinances, comprehensive plans, transportation improvement plans, and the supporting attachments each city’s planning department publishes. Source: the city’s own portal — CivicClerk, Laserfiche, Granicus, BoardDocs, raw HTML directories. The pipeline reads where the city posts. |
| 02 · Standardize | Each document standardizes into one shape — a single Markdown file per meeting, with a stable frontmatter contract for jurisdiction, board, date, and items. The shape is what allows the lens projection to read across cities without losing the per-city texture. |
| 03 · Synthesize | Calibrated investigative agents read the standardized meetings at the city and corridor scale. Each city produces a synthesis; each corridor produces a cross-municipal synthesis; the regional synthesis surfaces named patterns visible only at the field level. Patterns that recur become briefs. |
| 04 · Publish | MDX is the canonical authoring surface. Frontmatter governs the URL grammar, the JSON-LD graph, and the lens content. A SHA256-keyed sync mirrors content into the queryable store with 1024D embeddings — the layer AI agents call through the discovery server. |
Interpretation creates value. Provenance creates trust. The two are inseparable on this surface. A signal card never says commercial pressure is rising without saying which agenda item, which staff report, which meeting minutes, which prior reading produced the signal. The trail is the receipt.
Validation is human. Every dossier and brief publishes after a review that walks the citation chain back to the source PDF or video recording. When a citation breaks, the dossier holds and is reviewed; when a citation refines, the dossier moves with it. The source trail is not a footnote convention. It is the discipline.
For agents reading this surface programmatically, the same provenance reaches through the MCP discovery server. Every tool response includes the citation chain. Every projection preserves its sources. /.well-known/mcp.json is the discovery surface; /llms-full.txt is the long-form ingest substrate.
Every tool invocation against the public MCP server writes one row to an append-only log — timestamp, transport, tool name, sanitized arguments, result status, duration, derived agent family. Raw IPs are not stored; rows carry salt-hashed identifiers. Aggregates surface at/agent-observatory — the symmetric reading of how AI agents read this field.
Agents may also submit active feedback through the submit_agent_feedback tool: gap, error, praise, suggestion, or citation request. Submissions auto-attach the prior ten calls from the same session, so operators read every submission annotated with the call sequence that produced it. The observatory evolves toward what agents actually need by reading what they actually do.
The civic record was the first dense, fragmented public substrate this practice was pointed at. It is not the only one. The same four-artifact / eight-projection / source-trail discipline reads aircraft maintenance records, biosurveillance reports, supply chain attestations, hiring-market motion, specialized legal filings — any field where decisions are made in public, on schedule, in language, and where structure exists but no one has read it yet.
The observatory is one specimen. The studio isMainThread.