Growing commercial-connect.net on the park research
SEO architecture, lead conversion inside the TREC rules, compliance as code, and a data pipeline. Planned 15 Nov 2026.
127park pages
0suite pages
8build steps
105-155hours total
Read section 2 first. I framed the TREC carve-out too loosely in chat and the plan corrected me. A form fill does not make someone your client, so the availability survey cannot be automated. It waits for your call.
Verified, not taken on trust: the footprint detector that would not work is working. I re-ran it myself. It found Brownstone and Grogan's Ridge by building shape alone, plus 13 candidate clusters in one small box that are not in our roster.
Two page types, two rule sets
Park record pagesThird party property facts. All 11 TREC rules. No economics, no availability, non-representation statement, everything dated and sourced.
Service and intent pagesYour services and your own listings. Only 535.155(a) and (d)(1). May say "office condos for lease in Katy" freely.
Capture block on bothClean advertisement. Auto-sends sourced public fact only.
After your callRepresentation established. The availability survey is now a client communication and outside the rule entirely.
!
What is broken right now
Five things are wrong on the live site today. Three of them make SEO spend worthless until fixed.
The contact form is mailto: only. It opens the visitor's email app and captures nothing in Pace. Anyone on webmail gets nothing at all. Every dollar of traffic before this is fixed evaporates. It is build item 1.
The listings grid is client-filtered, so a crawler sees "No active listing matches these filters". Your three published listings are invisible to Google right now.
The submarket site is on ccg-submarket-site.pages.dev, so every link and ranking those 21 park pages earn builds authority for a throwaway subdomain instead of commercial-connect.net.
compliance.py applies all 11 rules to every page, so it would block a perfectly legal service page that says "for lease". The gate change has to land before the intent pages.
The Mahaffey flyer is still live on Showcase under retired Dow Real Estate Group branding with an August 2024 availability date. That is live 535.155(d)(14) exposure on an ended listing. Unrelated to this build, but the whole programme is positioned on your surfaces being clean.
Also needs doing: grep the live pages and JSON-LD for jacob@dowreg.com, and prove the new alias actually receives a test message before it goes in any capture copy.
2
The legal hinge, and a correction to what I told you
I framed the TREC carve-out too loosely earlier. The plan pushed back and it is right.
I said the page stays clean and the thing the visitor receives after identifying themselves can carry availability. That stretches the carve-out. 535.155 excludes a communication to the license holder's current client. A stranger who typed an email address into a form is not a client.
So the boundary event is representation established in a real conversation, not the form submission.
Advertisements, bound by 535.155(a) and (d)(1): every public page, every capture block, the auto-response email, the market report PDF, the park fact sheet, the quarterly digest. None of these carry availability for anyone else's property.
Not advertisements: the availability survey, tour sheets, comps and rate talk, sent by you to someone you currently represent.
What this costs: the availability survey cannot be automated. It waits for your call. That is slower and it is the right trade.
This is the sentence the whole design rests on, so it is worth being conservative about it.
3
SEO architecture
Fold the submarket site into commercial-connect.net, build 127 park pages on the open named-entity terms, and a small set of city service pages on the open broker terms.
Fold ccg-submarket-site into /submarkets/ on commercial-connect.net. Cheap because Pages serves static assets before the catch-all Function runs, so the gated gen.py pipeline survives untouched. Deploy the new URLs, verify, then flip 301s. Never 301 to a page that is not live yet. 6 to 10 hours.
URL shape: /submarkets/ hub, /submarkets/office-condos/<city>/ index, /submarkets/office-condos/<city>/<park>/ record pages. Service pages live on the main site at /office-condos/<city>/ and /commercial-broker/<city>/.
No suite pages, per your rule. A unit gets a URL only when it is your listing. The effort goes into park-page depth instead: unit mix, developer attribution with cross-links to that developer's other parks, turnover count, parcel aerial. All 22 SHB parks interlink through attribution, which is a cross-link structure no portal can generate.
Chase tier 1 and 2 first. Tier 1 is the 127 park names, zero portal competition. Tier 2 is broker and advisor terms ("commercial real estate broker katy tx", "tenant representation"), which the 15-SERP study found carry zero portals. Tier 3 is "office condo parks in <city> tx". Volume is in LAND, so a land services page comes after.
Do not chase head inventory terms. LoopNet is top-10 in all 15 SERPs analysed. A service page may legally use lease language; spending content hours trying to outrank LoopNet head-on is still a losing bet.
Schema: Place on park pages, RealEstateAgent plus FAQPage on service pages, CollectionPage on indexes. Offer and RealEstateListing stay banned everywhere, even on your own listings.
Roughly 36 to 54 hours for the whole SEO workstream.
4
Lead conversion
Replace the dead form with a Pages Function that writes to Pace and a D1 backup, and give each page type a value exchange that is legal to auto-send.
/api/lead Function. Turnstile plus honeypot, write to D1 first so a Pace outage never loses a lead, then upsert the Pace contact, create the opportunity, alert you. Verify by reading the lead back out of both, then delete the test contact. A 200 proves nothing.
Fields that earn their place: name, email, phone (required on requirement forms, optional on the digest), role (looking for space / own a unit here / own elsewhere), size and lease-or-buy, and timeline mapped to a concrete expected_decision_date. That date is the revenue forecast; a pipeline without it shows activity, not money. Not collected: company, budget, current address.
Park page carries two doors. Tenant door: "tell me your size and timeline and I will run the availability picture across the whole park and its neighbours". Owner door: "own a unit here? I will walk you through how your park has actually been trading". Confidential, no obligation, no value opinion.
Auto-sent immediately: the auto-response, and the city Market Report PDF or park fact sheet. Those carry only dated sourced aggregate public fact, so they are clean advertisements and they prove the research is real.
Never auto-sent: availability, rates, lease terms, anything about a specific third-party unit, any price opinion, any SMS.
15 minute response window in business hours, before 9am next day otherwise. A lead untouched at 24 hours re-alerts. The auto-response line that does the work: the report "deliberately does not include current availability, because availability moves weekly and I put a live survey together for the people I am actually working with, so it is accurate when it reaches you." That turns the legal constraint into the sales story.
16 to 25 hours. This ships before any new page exists.
5
Compliance and brand as code
Extend compliance.py with a page_type flag so each page type gets the right rule set, and add rules for the things that have actually bitten.
page_type lives in data, never inferred from the path: survey / index / hub / service / own_listing. An unflagged page defaults to the FULL survey rule set, so forgetting the flag can only over-enforce, never under-enforce.
Service pages waive R3 and R4 (the economics and availability bans) and gain S1 to S3: no park name near an availability token, no inventory claim not generated from the Hub's live count, and a rendered-CSS check that your name and CB&A Commercial render at half the largest contact font per 535.155(a).
L1 makes the mailto: defect unrecreatable: any mailto: in a built page fails the build. Consent text must match the versioned constant byte for byte.
L2 runs the auto-send templates through the same availability scanner as survey pages, because they go to non-clients. A template edit that adds availability language fails before it can send.
B1 catches the gold trap: colour may never resolve to #ffc000 or the fill token. Enforced by token name AND literal hex, because the Tamina leak was one selector using the fill token as text.
prove-gates plants five leaks into a copy of dist and demands five non-zero exits. A guard nobody has seen fail is not evidence of anything.
CB&A signs off at template level, not per page. 127 park pages from one gated template present one compliance surface; a per-page review will not happen.
11 to 15 hours. Blocks nothing shipping today and unblocks every new page type.
6
Data: detector 4 is solved, and I verified it myself
The footprint detector that would not work is working. I re-ran it independently rather than take the agent's word for it.
The bug: the envelope was sent as bare lat/lon with no spatialReference, so FEMA's service read degrees as Web Mercator metres and queried an empty box near null island. It also returns HTTP 400 on resultRecordCount, so you page with resultOffset.
Verified by me just now: 713 commercial footprints over a 4km box in 0.6 seconds. Running the cluster detector on it found Brownstone Office Condos and Grogan's Ridge by shape alone, plus 13 more candidate clusters in that one small box that are not in our 106-park roster.
Why this matters: it is the only detector that sees a park before any appraisal record exists, and it works identically in Brazoria, Galveston and Waller, which routes around the three counties with no usable parcel data.
Needs tuning before trusting it: at 85m link distance it merged Woodgrove into Brownstone, and "Entertainment and Recreation" and "Wholesale Trade" uses are false positives. Candidates go to a review queue for one aerial glance each, never straight to the roster.
Montgomery stays semi-manual. MCAD does not split regimes into unit accounts, so owners come from the county clerk deed index, 15 to 30 minutes per park, quarterly. Montgomery park pages must say "from recorded deeds as of <date>".
Fort Bend and the three missing counties get bulk roll requests under Tax Code 25.195, which obliges the district to provide the roll electronically at nominal cost. The bulk roll is a different product from the web layer and usually carries full legals.
The first revenue-bearing output is not coverage, it is the Harris quarterly diff. An owner change in the 1 to 4 unit segment is a warm outreach trigger at the exact moment it is warmest.
28 to 42 hours. Storage moves to D1 (dow-hub-db) so the site and the dataset cannot drift apart.
7
Build order
Roughly 105 to 155 hours total. The order is not negotiable in the first three steps.
#
Step
Hours
1
Lead capture Function, Pace + D1, retrofit the 7 existing pages
County expansion: bulk requests, Montgomery clerk procedure
6 to 10 plus waiting
Steps 1 to 3 gate everything after them. Step 1 because traffic currently lands on a form that captures nothing; step 2 because no service page can pass the gate until it exists; step 3 because every week it waits is authority spent on the wrong domain.
Decisions I need from you
Fold the submarket site into commercial-connect.net?
Which PaceCRM location receives website leads?
Who is the backstop when you cannot answer in 15 minutes?