A teardown of one production run on 1543 Washington Avenue, Miami Beach, 16 September 2026. Stage names, counts and timings are read from that run's own telemetry.
The plainest way to use Titleman is also the least explained: you open a room, you type an address and a question, and something comes back. No product picker, no form, no template. This is what is actually happening in the gap.
1. Before anything runs, the subject is pinned down
Twenty-two seconds in, the run had settled what it was looking at:
- Use class retail, status FOUND, resolved from the recorded use code.
- Resolution basis: the register's use code — not the listing text, not an inference from the address string.
- Parcel
0232340191140, with a property passport built against it.
Only then did the state change to in progress. A run that cannot pin the subject does not get to proceed to the expensive part — that ordering is deliberate, because every number downstream inherits this decision.
2. The room picks the engine, you do not
Fourteen seconds after the work began, the run recorded which engine it had dispatched: underwriting. Nobody chose that from a menu. The question implied it, and the room routed it.
That is the whole argument for the plain room over a product picker. A person asking "is this worth what they are asking" does not know, and should not have to know, whether that is an underwriting question, a comparables question or a best-use question. Usually it is three of them.
3. Then ten reasoning stages, most of them at once
The run recorded each stage by name. Stripped of internal prefixes:
- main and investor role — read the question and take a stance
- comps engine — assemble comparables
- transactions — what actually traded
- costs — what it costs to own and to fix
- finance — the capital side
- forecast / news — what is moving around the asset
- underwriting — put the numbers together
- final report — write it
Ten reasoning agents, 13,735 seconds of work — three hours and 49 minutes of it. The answer arrived in 1 hour 16 minutes of wall clock, because 9,357 of those seconds ran on top of each other.
4. What comes back is a canvas, not a paragraph
The output is a board of panels, each one a different view of the same asset. From this run, among others: an asset passport, an investment scorecard, a capital stack, a cash-flow waterfall, a comps scatter, a football field, an exit value forecast, an isochrone and drive-time view, competitor density and a competitive set map, a market pulse and a liquidity tracker, interest-rate context, and satellite surroundings.
Every panel records its own state — ready, or empty, or needs data. A panel with no source behind it is left visibly empty rather than filled with a plausible number. That is the single design decision that matters most in this product, and it is the one that makes the rest of it trustworthy: you can tell at a glance what the run knew and what it did not.
5. What this is, and what it is not
The output is an opinion of value and the analysis behind it. It is not an appraisal — we hold no appraisal licence, and in the United States appraisal is licensed work. Treat it as the document you take into the decision, not the one that closes it.
It returns a written answer and a canvas in the room. A workbook or a deck comes back on some jobs and not others, so if a file is the deliverable you need, say so in the first message rather than the last.
Measured on the Titleman production database, run id 25183, 16 September 2026.