Property: the evidence under a figure, and the valuation the agent cannot make

Property work is not the production of a price. It is the construction of an evidence trail under one: a comparable table that shows what a figure rests on, a market assessment that states the range the evidence supports, an abstraction that keeps a lease’s terms intact, and a checklist that says what must be produced before anyone signs. The price is the last thing in the file and the least defensible on its own. This piece sets out the job, the six documents that carry it, the six ways the work fails, the measured cost of a deal — $0.35 of model calls at the corrected rate — and the boundary that matters most in this family: an agent may assemble the evidence, and may not value the property, interpret the lease or sign.

The job

A property transaction produces an unusual artefact: a number a person will defend in writing, under a duty, with their professional liability attached to it. Everything else in the file exists either to make that number defensible or to record why it is not. That is the whole discipline, and it is older than any software in the room.

The work is document-heavy long before it is judgement-heavy. A broker holds a listing, a set of comparables, two leases and a title; an owner holds a developer’s brochure and a service-charge budget; a client holds a brief saying what they want and a checklist someone gave them at the start. Reading that material, normalising it and assembling it into six documents is the job, and it is the part of the job that consumes the hours. The judgement — whether the range holds, whether the lease is acceptable, whether to exchange — belongs to people, and this piece does not move it.

One role carries all six documents: real_estate_assistant. It is one role and six artefacts, not six products. The comparable table is the spine, because every later document either restates it or argues from it. The market assessment restates it as a range. The lease abstract is the same discipline applied to a different document: what the instrument says, term by term, quoted rather than summarised. The checklist is the same discipline applied to a process: what must exist before signature, who owns it, and what is missing. The deck and the proposal are the same evidence arranged for two different readers.

The order matters more than it looks. A market assessment written before the comparable table is an opinion with a table bolted on afterwards. A vendor deck written before the assessment is a price looking for support. A checklist written last is a record of what the transaction turned out to need. Each of those is a common sequence and each produces a file that reads well and cannot be defended.

One instrument built for this job is the studio’s Real estate assistant, which works a property file and produces the six documents described below. This piece is about the work rather than the instrument: where it names the assembly it says so, and every argument here stands whether or not an assembly is used. The role reads the file with documents/extract, builds the tables with documents/data_table, writes the assessment and the proposal with documents/report, renders the deck with documents/deck, delivers PDF or DOCX with documents/office, and quotes clauses through external/graphify over a corpus the office supplies. It has a mailbox and a chat channel and no WhatsApp, because that connector is not built — a WhatsApp enquiry becomes a handover note rather than a reply.

Evidence before opinion

Every number about a property or a market carries a source and an as-of date. That sentence is the whole method, and most of the failures described later in this piece are a breach of it.

Three kinds of statement appear in a property file and they are never mixed. An observed statement comes from a document or a cited page: the lease says the term is ten years from 1 March. A computed statement is arithmetic on observed values, shown so that a reader can redo it: on an illustrative file priced at 1,890,000 across 450 square metres, the figure is 4,200 a square metre because the first number divides by the second. An opinion is a judgement, and it carries the condition that would change it: the range holds while stock stays at this level. A file that labels the third kind as the first is not a sloppy file; it is a misrepresentation waiting to be discovered by the other side’s adviser.

A listing is a claim by a seller, not a fact, and a portal price is an asking price whether or not the portal labels it as one. Neither is a sold price, and no arithmetic turns one into the other. A rent is evidence only when the comparable is genuinely similar — same segment, same area, similar size, condition and tenure — and even then only with the differences stated. Where two sources disagree beyond a small tolerance, both are reported. The average of two disagreeing figures is a third figure that no document supports.

Stale is stale. A statistic older than the question needs is labelled with its date and treated as history, not as today’s market. That is a small discipline with a large effect on a market assessment, because the temptation in a thin market is to reach for the last transaction that fits the story and present it in the present tense.

The six documents

Six artefacts carry the work, and each has one reader and one job. Where an artefact is asked to serve two readers it serves neither, and the pack thickens without getting stronger.

Table 1. The six artefacts, the reader each is written for, what it must contain to be usable, and the part that renders it.

The comparable table is the plan. It decides what the assessment can claim and what the deck can argue, and its absence makes every later document an opinion with formatting.

The market assessment is the first document a client reads, and the one most often written as advocacy. Its usable form is narrow: the comparables, the adjustments applied, the range those adjustments support, what would move the range, and what is missing from the file. Five items, in that order, and the last two are the ones that make it a professional document rather than a sales document.

The lease abstract is where precision is cheapest to achieve and most valuable. Its rule is that every term is quoted to its clause, because an abstract that paraphrases is an interpretation wearing a table’s clothes.

The checklist is the artefact that prevents the expensive telephone call three days before exchange. It is not a list of documents a jurisdiction requires in general; it is the list for this transaction, from the file and the client’s own list, with a status and an owner on every row. Where the file does not say who owns an item, the row says so, and that gap is itself a finding.

The deck and the proposal are the outward-facing pair, and both are constrained by the evidence behind them. A pricing argument in a deck that cannot be traced to a line in the comparable table is a claim about the seller’s hope. A proposal that promises an outcome is a document the client can hold against the office later, and none is written here.

Building the comparable table

Four steps, and the second is the one that decides whether the table is evidence or decoration.

Normalise. Same units, same basis, same date. Area in square metres rather than in whatever the listing used. Rent per annum, and stated as gross or net of service charge, because a headline rent gross of service charge and a headline rent net of it are not comparable numbers, and the difference between them is often material. A price from eighteen months ago is dated, not current.

Adjust. For each difference between the comparable and the subject — size, condition, floor, outdoor space, parking, date — state the adjustment, its direction, its size and its reason. The adjustment is the argument of the table. A comparable with no adjustment is either identical, which is rare, or unexamined, which is common.

Rank. Against the client’s stated objective, not the analyst’s taste. A buyer seeking income and a buyer seeking a home rank the same four properties differently, and a table that ranks by price has answered a question nobody asked.

Conclude. A range, the risk attached to it, and the condition that would change the conclusion. A single number with no range is a valuation attempt, which is the next section’s subject.

Three properties is a thin table; five is workable; the count is less important than the rule that every line carries a source and an as-of date. Where a comparable is an asking price rather than a transaction, it is labelled as an asking price and given less weight, and the assessment says so in words rather than adjusting it silently by a percentage nobody can defend.

The market assessment is not a valuation

A valuation is a licensed act in most jurisdictions. It is performed by a qualified person, signed, and carries that person’s liability; in many jurisdictions the capacity to do it for a lender is regulated, and the signer’s indemnity insurance is priced against it. Nothing in a message confers that standing, and no assembly confers it either.

What the role produces is the evidence a valuer works from: the comparables, the adjustments, the range those adjustments support, the method visible so that a reviewer can redo the arithmetic. It labels its own sentences. This one is arithmetic over the supplied comparables; that one is a judgement with a stated condition. The labelling is not modesty. It is what makes the document usable by the person who does hold the licence, and it is what stops a reader mistaking a supported range for a professional opinion.

The distinction has a practical consequence for the office. A market assessment is the document an agent shows a client to explain why the asking price is where it is. A valuation is the document a bank accepts. The two are produced by different people under different duties, and a file that lets the first drift into the second has created a liability it did not intend.

Reading a lease without opining on it

An abstraction is a transcription with structure: parties, term, commencement, rent, review, service charge, break, reinstatement, and the rest, each quoted to the clause it comes from and each with the page it was read from. Thirty-one terms is typical for a commercial lease and rather fewer for a short residential tenancy; the number is not the point.

The point is that the abstract does not say what the lease means for the tenant. Whether a break clause is operable on the date the tenant intends, whether a review mechanism produces the increase the landlord claims, whether a reinstatement obligation is enforceable as drafted — those are questions of interpretation, they turn on the whole instrument and on the law of the jurisdiction, and they are answered by a qualified person who has read it. An abstract that answers them has stopped being an abstract.

The failure mode is specific and common: paraphrase that strengthens. A clause permitting entry on notice becomes a clause requiring notice; a rent review on an upwards-only basis becomes a rent review. Each step is small, each reads as clearer than the original, and each is a statement the document does not make. The prevention is mechanical — quote, cite, and mark anything that is a summary as a summary.

What the abstract buys is that nothing is lost between the lease and the person advising on it. A file with thirty-one terms quoted and dated lets a lawyer answer in an hour what would otherwise take a morning, and lets the agent answer the questions that are not legal ones — when the term ends, what the rent is, what the service charge covers — without pretending to the ones that are.

The pre-exchange checklist

The checklist is the process document, and it is the one that earns its place in a transaction that has gone quiet. For a sale or a rental it sets out the ordered steps and, against each, what the agent does, what the client provides, what a third party produces — lawyer, notary, bank, registry, authority — and what can stall it. Each item carries a status and an owner.

Two rules keep it honest. The first is attribution: what a jurisdiction requires is attributed to that jurisdiction and to a named instrument, with the date the instrument was read. A local rule is never generalised into a fact about property everywhere, and it is never stated without its source. The second is silence. Where no source is supplied for a requirement, the row records the question and who must answer it rather than answering from recollection. A checklist with an honest gap is worth more than one with a confident fill.

The checklist is also not a compliance opinion. It records what has been produced, what is missing and who owns the gap. Whether a document is compliant, sufficient, or effective for its purpose is a conclusion a qualified person draws; the agent’s contribution is that the question arrives at that person with the evidence attached and the gaps marked.

The deck and the proposal

The deck is written for a room: the property, the evidence, the pricing argument, the process and the timing. Its discipline is that every claim about price traces to a line in the comparable table, and that nothing promises a return. No guaranteed yield, no invented buyer interest, no hidden cost moved into a footnote.

The proposal is written for one reader deciding whether to instruct: the brief restated in the client’s own terms, the options with their trade-offs, the mandate terms, the fee basis, the next step. Two sentences do the damage in proposals, and both are banned here. One promises an outcome — a sale at a figure, a tenant by a date. The other implies a service the office does not provide, which in this family usually means regulated work.

Both documents inherit their quality from the table underneath them. A deck built on an unexamined comparable set reads exactly as confident as one built on a sound set, which is why the table is checked before the slides are written and not after.

Six ways this work fails

The failures below are ordinary. Each is a way a property file that looks complete turns out not to be defensible.

The clause recalled rather than read. A summary cites a lease or a regulation in the document’s voice, but the words come from the model’s sense of how these documents usually run. The other side produces the actual clause and the summary dies, taking the rest of the file’s credibility with it. The prevention is procedural: every clause quoted comes from a document supplied, and where nothing is supplied, nothing is quoted.

The asking price used as evidence. A table is built from portal listings and presented as market evidence, with no adjustment for the gap between asking and achieved, and no date on the figures. It is the commonest table in the business and the easiest to break: one question about a sold price and the range collapses. Asking prices are used as a cross-check, labelled, and weighted below transactions.

The adjustment nobody can redo. A comparable is adjusted by twelve per cent with no reason given, or the adjustment appears only in the conclusion. The table then supports any number the author wanted. An adjustment is defensible when a reviewer can redo it from the stated reason and reach the same figure.

The abstract that interprets. A break clause is summarised as the tenant’s right to leave, which is a legal conclusion about operability. The lease says what it says; the abstract quotes it, and the adviser decides what it does.

The checklist that answers itself. A jurisdiction’s requirements are filled in from recollection because the file contains no instrument for that jurisdiction, and the row looks complete. The transaction then stalls on a document nobody knew was needed. Where the source is absent, the row records the question and the owner of the answer.

The deck that promises an outcome. A vendor deck states a return, a timeframe or a buyer appetite that no evidence supports. It wins the instruction and creates the dispute. The pricing argument comes from the table; the process comes from the checklist; the outcome is not stated by anyone in this file.

What the agent may not conclude

Five boundaries hold here, and they are properties of the arrangement rather than promises about behaviour.

It cannot value. A valuation is a licensed act, signed by a qualified person whose liability attaches to it, and often regulated when it is produced for a lender. The role produces the evidence and the range that evidence supports. It does not produce a valuation, and it does not describe its assessment as one.

It cannot say what the document means for the client. Extraction and organisation are the agent’s work; interpretation is not. Whether a break is operable, whether a review is enforceable, whether a clause survives the jurisdiction’s mandatory rules — each is a question for a qualified person, and the abstract is built so that person can answer it quickly.

It cannot draft operative language or mark anything ready to sign. Documents are produced from the office’s approved templates for the jurisdiction in the file, with fields filled from the file and every unfilled field marked as needing one. No clause is invented, no statutory reference is guessed, no notice period is recalled. Every legal document remains a draft until a qualified person at the office has read it, and the lawyer’s name goes on the instrument.

It cannot state one jurisdiction’s law as universal. What a jurisdiction requires is attributed to that jurisdiction, with the instrument named and the date it was current to. Where the file supplies no source, the checklist carries the question rather than an answer.

It has no standing of its own. A role named for a job is a job description, not a qualification, a licence or an agency. Nothing here is regulated, nothing signs, nothing posts to a portal or a listing site, and none of it is legal, tax or valuation advice. Where the office’s own rules on fair treatment apply — no filtering, ranking or describing by protected characteristics, no inflated yields, no invented interest — the role is bound by them and the pack shows it.

The cost of a deal, measured

This publication takes the amount from the 402 challenge in front of the call and treats a catalogue or a price list as a way to check a figure, never as a way to supply one. The figure that has circulated for a DeepSeek drafting call is $0.00875. It is a catalogue value, and it is not the amount the endpoint asks for: the measured price of deepseek-v4-flash on Base is $0.00700 under exact for a short prompt, read from the live handshake on 30 October 2026 outside the service’s peak windows. The same call under a long prompt, and the upto ceiling at max_tokens: 1000, is $0.00955.

Priced at the measured rate, one complete deal’s document set — the six artefacts in Table 1, fifty paid model calls — comes to the block in Table 2. The call counts are planning estimates; the price is not.

Table 2. One deal, six documents, at the measured DeepSeek rate on Base. Source: live 402 handshake, 30 October 2026; call counts are planning assumptions stated below.

The arithmetic is short. Fifty calls at $0.00700 is $0.35000 — $0.35 for the document set of one complete deal, which is $0.05833 an artefact across the six. At the rate the catalogue shows, $0.00875, the same fifty calls would read $0.43750; the measured rate puts them at $0.35. Nothing in that total is priced except the model call.

The same price carries through a year at the office’s pace. Table 3 puts the three useful scales beside each other — one deal, a year of forty-five, and the balance an office starts with — and sets the metered ceiling next to the first two.

Table 3. The measured rate at three scales, with the metered ceiling beside it. Source: live 402 handshake, 30 October 2026; forty-five deals a year is an assumption.

The annual figure is the one worth remembering: 2,250 calls is $15.75, not $19.69. The higher number is the same arithmetic at the catalogue rate of $0.00875, which the handshake does not ask for. At the ceiling the year costs $21.49, which is $5.74 more — 36 per cent, the gap between a short prompt on exact and a long one, or a metered call at a thousand authorised output tokens.

Two more readings of the same figures. Forty-five deals across a year of two hundred and fifty working days is $0.063 of model calls a working day. And 0.50 USDC, the opening balance credited to a workspace, is seventy-one calls: one deal’s document set with twenty-one calls left over.

The one metered stream

The 402 challenge in front of a model call is the only priced thing in this arrangement, and that fact is worth stating plainly because it is where this role differs from a subscription.

The six documents are renders. The comparable table and the checklist are documents/data_table; the assessment and the proposal are documents/report; the abstract is documents/extract and a table; the deck is documents/deck. None of those parts carries a published per-call price in the catalogue, because they run inside the workspace rather than over the wire: there is no challenge to read, no amount to measure and no price to report. Strip the model out of the role and the six artefacts still render from a structured file. What leaves with the model is the reading, the adjustments and the judgement — not the grid, the abstract, the checklist, the report, the deck or the proposal.

That is the structural point, and it is not a claim that an unattended role does the work. What it does mean is that the role’s cost line is the drafting, not the documents; that the document path is priced by the engagement rather than by the call; and that a low model figure is evidence about the meter rather than about the fee.

The honesty that belongs beside it: no published per-call price is not free. It means the price is not metered per call in the catalogue — a different fact, and a weaker one. Nothing here says the renders cost nothing to run, and this publication has no measurement of what they do cost.

What the figure leaves out

Retrieval over the office’s corpus carries no published per-call price either, and comparables and clause citations are retrieval work: twelve comparables an assessment, thirty-one terms a lease, a further set for the decks and proposals. The mailbox that delivers the pack and the chat channel that reports engagement state carry no published per-call price. The WhatsApp connector does not exist, so there is nothing to price and nothing to omit on that row.

That is an omission, not a saving. The totals above understate the work by everything the workspace does besides calling the model, and the honest reading of $0.35 is that it is a floor for the model calls of one deal, not a price for the deal. Where a supplier publishes no per-call price, the correct entry is not an estimate; it is the statement that no price exists.

The largest cost is absent from the tables altogether. Preparing the file — comparables with their dates and sources, the leases, the title, the client’s brief and the office’s own checklist — is a person’s hours. Reading the output against the documents it cites is a person’s hours. The valuation a lender requires, the legal advice the transaction requires and the survey a condition question requires are other people’s fees, and this publication holds no figure for any of them.

When this is the wrong instrument

Six cases where the work is real and an assembly is not the way to do it, and one where the arithmetic settles it rather than a principle.

When a licensed valuation is required. No assembly produces one. If a lender, a court or a regulator needs a valuation, the instrument is a qualified valuer and the role’s output is an input to their work, not a substitute for it.

When the lease is the whole question. A tenant asking whether to accept a break clause, a review or a reinstatement obligation needs advice, not an abstract. The abstract makes the advice cheaper; it does not make it unnecessary, and starting with the abstract when the question is legal delays the answer.

When there is no comparable set. The table is built from comparables the office already holds. An agent cannot invent inventory, and an assembly given no comparables produces either an empty table or a fabricated one. In a thin market the honest output is a list of what is missing and what it would change.

When the file is one property and one opinion. A single flat, one buyer, one asking price: an experienced agent reaches the answer in ten minutes and the document set is overhead. The assembly earns its place where the file has several documents, several comparables and a reader who will test them.

When the client wants a number that will not be challenged. A file is sometimes assembled to support a figure already decided. That is the one brief this work cannot serve, because the table is the argument and the argument has to survive the other side’s adviser.

When the work is already a dispute. Once a transaction is contested, statements about what a lease means or what a property is worth become evidence in a process with rules of its own. The instrument is a lawyer, and documents prepared for that process are prepared under their direction.

The seventh case is arithmetic. Where the deal is small and the fee is small, $0.35 of model calls is irrelevant to the decision and the hours around it are the whole cost. The assembly is worth fitting where an office runs several files at once and wants the evidence trail consistent across them; for one small deal, doing it by hand is cheaper and faster.

What we measured and what we did not

The prices in this piece were measured, not listed: the amount in the 402 challenge, read on Base on 30 October 2026 outside DeepSeek’s peak windows, which are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday. The model identifier was read from GET /v1/models. Catalogue values were used to check a figure and never to supply one, and no figure here is derived from a catalogue.

The call counts are planning estimates, and they are labelled as such in Table 2 and Table 3: eight calls a comparable table, eight an assessment, six a lease across two leases, six a checklist, ten a deck, six a proposal, forty-five deals a year. They are not observations of a real deal, and they will be wrong for any particular one.

What this piece did not do matters as much as what it did. It did not run a deal through the assembly, and it describes no client, no property and no completed transaction, because it has none. It holds no figure for the accuracy of an abstraction, no comparison against a valuer’s range, and no time-to-complete figure, because none has been measured. The refusals described above — the valuation, the interpretation, the operative clause, the universal rule and the standing the role does not have — are stated in the role’s specification and enforced by the parts it has and the ceilings it spends within. They are reported here as specified rather than tested in a transaction, and a reader who needs to rely on one should test it.

Amounts move and a table is a snapshot. The same call costs different amounts on different chains, under exact and upto, across DeepSeek’s prompt bands and inside its peak windows, and the ceiling moves with the max_tokens an agent authorises. The challenge in front of the call is the price. Where a figure here is wrong, it will be corrected and the correction stated, in the manner of every piece on this site.

Where this goes

• The role: studio.x402.press/roles/real_estate_assistant — the agent, its six documents and its running cost
• The service it buys: x402.press/services/deepseek/ — the canonical price, seven chains
• The review: x402.review/2026/10/31/deepseek-review/ — the measured handshake and what it does not promise
• The catalogue: studio.x402.press — fifteen entries, thirteen roles and two desks

The clusters in this family

• Comparable evidence
• Market assessment
• Lease abstraction
• Sale agreement documents
• Vendor presentation deck
• What an agent may draft, and what must be signed

Method note

Prices read from the live 402 handshake on Base on 30 October 2026, outside DeepSeek’s peak windows; the catalogue figure of $0.00875 was checked against the handshake, found to be stale, and is not used anywhere in this piece except where it is named as stale beside the correction. Call counts are planning assumptions, stated in Table 2 and Table 3 and repeated in the arithmetic. Nothing in this piece is legal, tax or valuation advice: the agent extracts and organises document terms and the evidence behind a figure, and a qualified person advises. No client, property or transaction is described, because none was observed.

x402.review is written and edited by N.P. Vincent. The publication is connected to x402.press, which sells the services reviewed here; that interest is disclosed on every piece that touches a service, and it is stated here rather than left to inference.

Filed under Jobs · Tags: property, lease, comparables, valuation, x402