Address in. Feasibility package out — or an honest refusal.
A developer picks a market, types an address or parcel ID, and gets back the zoning rules that apply, the buildable envelope, a benchmarked cost range and the resulting numbers. A market is a way to browse and scope, not a unit of data: the address is resolved to its own local jurisdiction first — zoning in the US is a city and county function — and where that jurisdiction is not the market chosen, the platform renames both and answers with neither. Where the inputs to answer honestly are not in the database yet, the platform refuses and names every reason instead of returning a plausible-looking figure.
Denver, CO — modelled — the reference market: district standards read out of the published Denver Zoning Code by hand, an ingested parcel register and zoning layer, and a cost band built from real City of Denver permit valuations
Berkeley, CA — reconnaissance only — no ordinance read, no zoning layer or parcel register ingested, no cost basis, and California's state layer unread Missing here: market_ordinance_unread, market_zoning_layer_not_ingested, market_parcel_layer_not_ingested, market_state_layer_unread, market_cost_basis_absent.
Outside the modelled markets — browse scope, not a data unit — an address anywhere the deployment does not model is refused by name, never priced with another market's numbers Missing here: market_not_modelled.
This field asks the engine about whatever you type — the engine runs on our own machine, this page's own server puts the question to it (your browser never reaches it, and nothing is computed in your browser), and every answer says where it came from: LIVE means the engine computed it when this page was built, CAPTURED means the engine could not be reached and the answer is one of the 13 real runs captured from that same engine on 10-07-2026 · 5:29 PM UTC at commit 1c90cb4, replayed verbatim. Under Denver, CO the captured inputs are 2323 curtis st · 2307 curtis st · 2307 curtis st mstr · 2323 curtis · 2307 curtis · 5800 tower road · 1100 congress ave, austin, tx 78701 · 0234219013000 · 0234219033999. Anything else gets an honest message rather than an invented package.
{"address":"1100 congress ave, austin, tx 78701","market":"denver-co"}HTTP 404The engine refused before any package existed: error PARCEL_NOT_FOUND
no parcel matching address '1100 congress ave, austin, tx 78701'
Out of market: this address is not in a market we model
The engine said exactly this, in its own words. It did not guess, and neither will this page.
PARCEL_NOT_FOUNDHTTP 404no parcel matching address '1100 congress ave, austin, tx 78701'
We model one metro area today — Denver, CO — and this is what the product does with everywhere else: it refuses by name instead of answering with a number borrowed from a city whose code does not apply. Zoning in the US is a city and county function, so a new market is added one at a time, and until Austin's rules are read from Austin's code, there is no honest answer for Austin.
The address used for this check was a real one, typed in and answered by the real engine. Everything the engine knows about is listed in the response's own known_parcels, and every one of those is a Denver parcel or an obvious fixture placeholder.
engine_versions · run took ms · project_id · computed live by the engine when this page was built, against the deployment's own database
Denver, CO — modelled — the reference market: district standards read out of the published Denver Zoning Code by hand, an ingested parcel register and zoning layer, and a cost band built from real City of Denver permit valuations. These runs were captured from the deployment's own data, which is Denver, CO's — the market layer refuses every other scope by name. market=denver-co travels in the URL with the address.
Every figure, reason code, row id and citation below was returned by the running API or derived from the live database — this page does not type its numbers by hand. A LIVE answer was computed by the engine when this page was built; a CAPTURED one was captured from that same engine on 10-07-2026 · 1:26 PM UTC at commit 409ad7a, and every answer says which of the two it is. Three things here are written by us rather than generated, and we would rather name them than let the sentence above cover them: the acceptance count in the pill (nine checks in scripts/acceptance_denver.py, which the snapshot generator refuses to write past if any of them fails), the count of district codes holding no relief-free parcel, and the prose — including “What is not built yet”, which is a list of what we have not done. The market list in the dropdown is not written by us at all: it is the engine's own GET /markets response, captured with the runs, so the coverage claim on the page — one market modelled, one in reconnaissance — is the engine's claim, not our wording.
That one envelope parcel is 2323 Curtis St — a single register record in C-MX-5. Its figures are what the district's rules allow on the site, not a survey of what is there today: the note directly above the numbers says what the register records on it.
One real address, run end to end
This is the acceptance case: the owner's own address, run through the real pipeline and checked by an automated harness. It is deliberately the hard case — a pre-2010 district with an overlay, waivers, a use-form sentinel and no published frontage — so the refusal is the feature, not a failure.
{"address": "5800 Tower Road, Denver, CO 80249"}status: manual_review_requiredHTTP 201Refused for 7 named reasons
rules_not_transcribedhighthe numeric standards for this district (setbacks, lot coverage, FAR, height in feet, minimum parking) have not been transcribed into zoning_rules yet, and no US city publishes them as data. No limit is invented here: the district identity below is what the data supports.
former_chapter_59_districthighthis is a Former Chapter 59 (pre-2010) district. It sits outside the 2010 Denver Zoning Code building-form standards, so the modern standards do not govern the site — applying them would be confidently wrong.
overlay_districthighthe parcel is inside an overlay district. An overlay changes the standards that apply, and its provisions are not modelled, so the base-district numbers (even once transcribed) would not be the whole answer.
waivers_or_conditionshighthe zoning layer records waivers and/or conditions against this site. Site-specific relief like this is granted by ordinance and must be read before any standard is applied.
zone_use_form_sentinelhighthe zoning layer gives this site the use-form code '999', its sentinel for 'not a 2010 building form'. A building-form standard cannot be selected from it.
lot_frontage_unknownhighthe parcel register does not publish a street frontage, and the front setback cannot be applied without one. Supply the frontage (or a surveyed site plan) rather than accept a footprint derived from a guess.
cost_baseline_not_sourcedwarningthe construction cost basis that priced this response is ours, not market data: it is labelled 'derived' and we derived it from City and County of Denver — Commercial Construction Permits (daily extract from the Accela database) — 317, retrieved 2026-09-29. The city published the data underneath the figure; nobody published the figure. Its limits and its citation travel with it on this response (cost.baseline, and the cost_baselines record in the provenance), and it is a screening range for advancing or killing a site — not a bid, a budget or an engineering figure. Reported rather than refused because the basis is specified, cited and labelled: the envelope and the money may be computed, and the flag stays visible next to them.
Nothing numeric was returned
- envelope
- null
- cost
- null
- feasibility
- null
- sensitivity
- null
The cost basis for this run is our own derivation from a cited dataset, not a published figure. It is reported as a flag rather than a refusal: the basis is specified, cited and labelled, and the label travels with the number.
The exact numbers still missing for this district
Read back from the response — the standard must supply all of them, or the district gets no row at all:
max_dwelling_units_per_acremax_farmax_height_ftmax_lot_coverage_pctmax_storiesmin_front_setback_ftmin_lot_area_sqftmin_parking_ratiomin_rear_setback_ftmin_side_setback_ftEvery record that reached the answer is cited
| Table | Row | Record | Quality | Source |
|---|---|---|---|---|
| parcels | 8333 | 0010208160000 — 5800 N TOWER RD | sourced | City and County of Denver — Parcels (ODC_PROP_PARCELS_A layer 245) · layer 245 · read 09-27-2026 |
| zoning_districts | 3626 | C-MU-20 — Former Chapter 59 Zone | sourced | City and County of Denver — Zoning (CCD_Zoning layer 27) · layer 27 · read 06-02-2026 |
INPUT PROVENANCE (read off the records this response carries, not off a dataset label): of the 2 inputs that reached a figure, 2 are sourced (the parcel record; zoning_districts row 3626). No input in this response is a placeholder: nothing here is invented, and every labelled input above is either sourced or derived from a cited dataset. Estimates are ranges, not guarantees.
Real counts from the primary pilot scope
These are the numbers that decide which district can produce a first answer, and they are generated by a script against the live database rather than typed into a slide.
Why parcels are held back from an answer
A parcel carrying any of these is refused rather than modelled, because each one changes what the code means for that site.
- 988
former_chapter_59_district - 4,391
overlay_district - 218
planned_unit_development - 737
waivers_or_conditions - 1,866
zone_use_form_sentinel
Street frontage: measured for C-MX-5, with two things we will not smooth over
The Denver parcel register publishes no frontage field of any kind, so every frontage in the database is ours: the street-facing lot edge, measured against the city's own street centrelines, with the source, the tolerance and the method labelled on every single value. For C-MX-5, 395 of 455 parcels now carry one; 60 carry none and a named reason (no_parallel_street_edge), because a zero would be a wrong number wearing a number's clothes. Range so far: 2–465.49 ft, median 122.42 ft.
- Some "parcels" are not sites. The smallest readings are honest geometry on land nobody can build on: 0227129055000 (3721 N MARION ST APPRX) is 2 ft; 0235514036000 (2200 E 18TH AVE APPRX) is 2 ft; 0234921021000 (605 E 16TH AVE) is 15.4 ft. The edge genuinely exists and genuinely faces the street, so the measurement is right and the site is wrong. We did not invent a minimum plausibility bound to hide them — that would mean retuning a tolerance we derived from the data, and every number above would have to be re-measured.
- Many "lots" are stacks. 178 of 455 parcels share their stored boundary with at least one other parcel — 292 distinct boundaries among them, the largest group being 62 parcels on one boundary. These are condominium and unit stacks, so each unit inherits the parent footprint's frontage: honest as geometry, misleading as "the lot". Both this and the strips above now refuse with their own named reason —
boundary_shared_with_other_parcelsandrecorded_boundary_not_a_site— rather than being modelled from a frontage that measures the parent footprint. The 2307 Curtis St run on this page is one of them.
Measurement is C-MX-5-only so far: 395 of 12,407 corridor parcels and 0 of 9,141 in the airport scope. The rest still refuse with lot_frontage_unknown — which is why the acceptance case above still refuses. 5 frontage values in the database are still fixture test rows, not measurements, and are labelled as such.
C-MX-5, the district those runs came out of
C-MX-5, in the central corridor: the district the first numbers come from, spread across the district rather than concentrated on one block. It is not the largest code in scope — C-MX-12 holds 456 parcels to C-MX-5's 455 — but all 456 of C-MX-12's carry an overlay and none of them is relief-free, while 204 of C-MX-5's are.
What is not built yet
Listed plainly, because the value of this page is that it does not overstate where the product is.
| Piece | State | What it is waiting on |
|---|---|---|
| Address → parcel → district → refusal | works | Nothing. This is the flow shown above, checked by the harness on every run. |
| Zoning rules for C-MX-5 | works | The GENERAL building-form row was read out of the published code by hand and is loaded live with its citation, effective date and the cells the code states absent. It prices every run the search box answers. C-MX-5's other building forms are carried verbatim rather than averaged into the answer. The site-conditional Protected District rows are computed, not carried: every ingested zoning polygon is classified against the code's own enumerated list, the distance from the site's own zone lot polygon to the nearest protected polygon is measured to that measurement's own stated tolerance, and the parcel is refused where the 175 ft cap bites — or where the condition cannot be established. |
| Street frontage | in progress | Measured for C-MX-5 only: 395 of 455 parcels, 60 with a named reason and no value. The rest of the corridor still has none. The two data-quality cases above no longer publish a frontage at all — they refuse with their own named reasons, boundary_shared_with_other_parcels and recorded_boundary_not_a_site. |
| Cost benchmark | in progress | A derived band read from real City of Denver commercial construction permits, labelled 'derived' and carried with its citation on every response. The modelling assumptions underneath it — average unit size, net-to-gross efficiency, storey height, parking areas — are still fixture rows, which is why the cost-bearing packages above are labelled 'mixed' and carry a high-severity fixture flag. |
| Buildable envelope | works | Computed for C-MX-5 today: units, floor area, storeys, height, footprint, frontage and parking all come back in the runs above, each with the constraint that bound it and every caveat attached. |
| Feasibility numbers + risk flags | works | The cost band, the $/unit figures, the sensitivity band and the risk flags all come back with the envelope. Missing limits are reported as data flags, never filled in — and the high-severity fixture flag stays on the package until those assumption rows are sourced. |
| Markets | in progress | 3 markets are listed and all but one refuse by name. Denver, CO is modelled — the reference market, the only one with transcribed district standards, ingested layers and a cost basis. Berkeley, CA: reconnaissance only — no ordinance read, no zoning layer or parcel register ingested, no cost basis, and California's state layer unread; Outside the modelled markets: browse scope, not a data unit — an address anywhere the deployment does not model is refused by name, never priced with another market's numbers. The list itself is the engine's own GET /markets response, captured with the runs, so a market can only be described as the engine describes it. |
| Search in the browser | in progress | The market dropdown and the field at the top answer a fixed set of 13 runs produced by the real engine and captured verbatim, each filed under the market it was asked in, and the page says so in the answer itself. A market whose data this deployment does not hold cannot be answered at all — the engine refuses it by name, and the page shows that refusal rather than a package borrowed from Denver. Answering an arbitrary address needs the API hosted, which we are not paying for yet. |
| PDF report, share link, accounts, saved projects | not started | After the first computed envelope. |