The run, and the number it gave itself
Run 23220 was a broker opinion of value on a Tampa industrial warehouse. It was executed on an internal staff test account against a publicly identifiable address, so no client's material was ever in it. The run returned a verdict of PASS, and beside that verdict it returned two figures most software will not put in front of a reader: a confidence score of 45 out of 100, and a risk rating of High.
Forty-five out of a hundred is a weak score, and it is the score that went into the report. It was not rounded up, softened into a phrase like "moderate-to-good", or dropped from the summary while the verdict stayed. It is the number the evidence supported.
The run also stated why it scored itself down. It had found two sale comparables. It had hit a zoning code conflict it could not resolve, and left it visible instead of picking the reading that made the file tidier. It was missing occupancy and condition data on the subject. And it named what would raise the score: a zoning letter, and either a land comparable or a cost estimate.
Two sale comparables is the crux. A standing rule in this product line is that below three comparable sales, a point value becomes a range. A system that knows that rule and then prints a confident point value anyway is not being helpful. It is being wrong in a way that is hard to see.
The dishonest version of the same run
There was a version of that report that would have read better. Put 95 out of 100 at the top, call the risk moderate, mention the zoning question in a subordinate clause near the end, and the output looks like the work of a firm that knows what it is doing.
It would rest on the same two sale comps and the same unresolved zoning conflict. Nothing about the property would be better understood. The only difference between the 45 version and the 95 version is how much uncertainty the reader is allowed to see.
Titleman's internal review read that draft and passed it specifically because the moderate verdict was presented as the selling point rather than dressed up. The reviewer's language was that manufacturing false confidence on thin evidence would have been the dishonest version of the same document.
The same instinct governs naming. Titleman holds no appraisal licence, and in the United States valuation is licensed work. The product is a Value Opinion. It is not called an appraisal anywhere, because it is not one.
The review that passes this is the review that blocks other things
A confidence score is only worth reading if the thing producing it can say no. Titleman's outputs go through a gate: a reviewer that did not write the artifact returns PASS, PASS with a disclosed gap, or BLOCK. The reviewer is separate from the author by construction, so nothing reviews itself.
That gate blocks work. One recent verdict concerned a planned Tampa industrial page. The reviewer did not take the previous reviewer's word for anything; it re-ran the check from scratch and found zero content artifacts attached to the task — no page draft, no data file, nothing. It traced the chain upstream and found the structural dependency had failed terminally, meaning it cannot unblock. It counted four dead review gates spawned around the same row, with zero publications produced across the whole chain. The verdict was BLOCK, not retryable: a PASS there would close a row that produced nothing.
That page does not exist. It is not live, not available, and not linked here, because there is nothing to link to. The analytical work behind it is real and finished; the page built on it is not. Those are two different claims, kept apart on purpose.
A gap disclosed instead of filled in
A third verdict is the middle case. Copy was drafted for a post about publicly filed net operating income, citing two Tampa Bay properties whose NOI appears in SEC-filed CMBS exhibits. The review passed it as a draft and recorded that two lookups remain before it can post: matching the correct accession number to each property, and naming the reporting period for each NOI figure. The draft carries the exact line to insert once those resolve, rather than a guess or a vague placeholder.
The same draft records why it replaced an earlier planned piece. The original angle was loans maturing soon. The data did not support it — every Tampa maturity in the slice falls between 2028 and 2032, none within 24 months — so the angle was dropped and the reason written down, instead of the topic quietly changing.
Where agreement is evidence, and where it is not
A separate released report compared four brokerages' Q2 2026 Tampa industrial figures: Cushman & Wakefield at 7.4% vacancy, Matthews at 8.1%, Savills at 9.3%, and CBRE at roughly 10% availability. The report does not average them. It states that CBRE is measuring availability, which is a genuinely different metric, so treating all four as one disagreement would be a mistake. Asking rents were carried the same way, each tagged to its publisher: Matthews $12.86 psf, Avison Young $9.63 psf, Savills $9.05 psf.
The report puts a number on why the spread matters: 190 basis points of vacancy spread is about a $5.4M value swing on a 100,000 sf building at a 7% cap rate. It also states plainly that nothing in it is Titleman's own measurement and nothing in it is an appraisal.
Two separately reviewed artifacts, written in different passes, cited the same figures for the same quarter and matched exactly. That is a consistency signal worth having. Four brokerages publishing different vacancy numbers is not — they are measuring different inventories, and reporting it as consensus would be the error.
Wrong in the flattering direction
The same discipline caught a modelling error that was entirely ours. An underwriting workbook for a 416-unit Tampa multifamily property — anonymised here — first returned a 53% levered IRR, a 6.36x equity multiple, and a 2.98x debt service coverage ratio. Those numbers were the error announcing itself.
The cause was a basis mismatch. The draft had defaulted the purchase price to the property's county-assessed value of $35.21M, which is an administrative figure used to compute property tax, not a price anyone would pay. The arithmetic connecting price to income was correct; it was reporting the gap between two different bases as investment return.
Corrected, the model returns a levered IRR of 8.66%, confirmed at 8.7% by an independently written second implementation of the same model. Year-1 NOI of $3,681,558, an operating expense ratio of 48.09%, and a loan sized by the DSCR test at $39,862,003 rather than the higher LTV figure of $41,617,613 — meaning cash flow, not collateral value, is the binding constraint, and the required equity of $24,165,095 is larger than a headline LTV would suggest. The equity multiple is 1.47x.
Four automated checks now live inside the file, each of which fired on the bad first draft: a going-in yield outside roughly 3.5–9%, an exit cap priced tighter than the going-in cap, a DSCR above 2x, and an operating expense ratio under 40%, which is implausibly low for Florida product once insurance is counted. The workbook also states on its own face that no rent roll and no trailing-twelve operating statement for that property were ever seen, so every rent, the vacancy rate, every expense line and both cap rates are assumptions. It is a checked structure for underwriting the property, not yet underwriting of it.
How to read a number from us
Every figure in a Titleman deliverable is classified as sourced, assumed, or computed, and the classification travels with the figure. The governing rule internally is short: an unverified number does not go outside. A human signs off before anything is committed on the company's behalf.
The practical consequence is the one this article opened with: when the evidence is thin, the number that comes back is low, and it stays low. A reader acting on an opinion of value should want it that way. A product that returns 95 whether it knows or not tells you nothing by returning 95.