Questions, answered.

Nine: multifamily, retail, industrial, office, hotel, self-storage, build-to-rent, data center, and ground lease. Each one is a real engine with that class's own conventions, not one model with the labels changed. You pick the asset class and strategy when you create the deal, and the tabs, inputs and defaults change to match.

Both, on any class that can be built. Development gives you a budget with per-line draw curves, capitalized construction interest solved as a fixed point, a lease-up funding order, a permanent takeout sized to market or to loan payoff, and residual land value solved backward from a target development spread. That last one tells you what you can pay for the dirt.

Yes. Available room-nights times occupancy times ADR gives room revenue, then food and beverage and each other operated department carry their own revenue and their own expense ratio to departmental profit. Undistributed expenses and franchise fees (charged on room revenue, not total revenue) take you to GOP, the two-part management fee and fixed expenses take you to EBITDA, and the FF&E reserve sits below EBITDA to reach NOI. That last placement matters: burying the reserve in operating expenses or omitting it overstates NOI, which overstates the valuation and the debt coverage at the same time.

Three things that a generic model gets wrong. Concessions are a function of length of stay rather than a guessed percentage, so one free month on a thirteen-month average tenancy is 7.7 percent of gross rent. Other income (signup fees, admin fees) is earned per lease event and annualized by turnover, not multiplied by twelve. And the in-place to stabilized vacancy glide is explicit, because storage vacancy really does run near twenty percent before stabilization.

The cost stack. Horizontal work (grading, utilities, streets) is quoted per lot and runs first; vertical homes are quoted per home and start while later lots are still being graded. Those are separate budget lines on separate schedules, and lots are not the same count as homes. Pencil also models the lease-up property-tax phase-in, where a ground-up project is assessed on partial improvements until the assessor catches up. On a typical bill that is real money in exactly the years the deal is tightest on coverage.

Yes. Roll any set of deals into one vehicle and you get three valuation lenses side by side (discounted cash flow, direct capitalization, replacement cost) against price and all-in basis, a principal-weighted debt rollup, unlevered and levered returns, and the minimum debt service coverage and debt yield across the entire hold rather than year one. A portfolio can also carry a single promote over the combined cash flow, which is the correct structure for a fund. Summing each deal's own promote overstates the sponsor whenever a winner carries a loser.

For most private and mid-market buyers, yes: Pencil does the lease-level work, the recoveries and the returns, and the people you send it to can open it in a browser instead of needing a license. For institutions whose investment committee and portfolio systems consume Argus specifically, treat Pencil as interoperable rather than a replacement. You can import an Argus report pack and rebuild the model from it, including per-lease step schedules and recovery structures, and compare Pencil's cash flow against Argus's own.

For the underwriting itself, yes. Both read a rent roll and a T12 and standardize them. Pencil then builds the full model on top rather than handing back a normalized spreadsheet, covers nine asset classes rather than multifamily alone, and keeps every figure traceable to the line item it came from. The honest limit is data. Radix also runs a market-benchmark and research business, and Pencil does not sell data. If what your subscription is really buying you is Radix's rent comps, keep it and use Pencil for the modeling.

Rent rolls and operating statements in Excel, CSV or PDF, including scanned ones. Also commercial and retail rent rolls, rehab trackers, rentable-item directories, and Argus report packs. Multi-row headers, charge-code layouts and subtotal rows are expected rather than fatal.

Every original line is kept beside the structured row it produced, as an audit trail you can open. When a number in the model looks wrong, you can see the exact source row it came from rather than guessing at the parse.

No. Classify a general ledger line once and Pencil maps it the same way on every future statement for your workspace. You can also define your firm's exact operating statement layout, including line names, groupings and subtotals, and have the pro forma reproduce it.

You fix it in place and the model recomputes. Nothing about an import is final, and no imported value is locked. The classification you correct is also what Pencil remembers for next time.

No, and this is the design premise. Every input and every output is a named line item in a dependency graph, 124 of them in a standard deal, and you can see and change any of them. Purple means editable, everywhere in the product. Change one and only its dependents recompute, in order.

Pencil is built and tested on hundreds of institutional-quality models, by a capital markets broker who underwrites and closes deals for a living. Every release runs an automated suite that checks the engine line by line, not just the headline return: gross operating profit, departmental expense, capitalized interest, promote dollars. And because every input and output is a visible line item rather than a hidden cell, you are never asked to take the result on faith. You can trace any number to its source, change it, and watch what moves.

Yes. Any number of tiers on internal-rate-of-return or equity-multiple hurdles, greater-of gates, general partner catch-up, sponsor co-investment, multiple limited partner classes, and preferred return that compounds annually, monthly or not at all. Sponsor fees are modeled where they actually sit: fees at close are funded by equity and raise the hurdle basis, annual and disposition fees reduce distributable cash. You see the limited partner's gross-to-net bridge, so the fee load is visible rather than implied.

Multiple tranches with real lifecycles, interest-only periods rolling into amortization, sequential refinancings that pay off the loan they replace, four day-count conventions, variable-rate forward curves, and assumed loans that pay their original contractual payment instead of a re-derived one. Sizing can run on loan-to-value, loan-to-cost, debt service coverage or debt yield. Pencil reports lender yield including the origination fee, the balloon at maturity, weighted average life and duration.

There is a Monte Carlo simulation on development deals: thousands of trials across rent growth, exit cap rate, hard cost, absorption pace and schedule, with rent and cap rate correlated the way they move in practice rather than independently. You get the distribution, the probability of loss, and a tornado chart that says which assumption is actually driving the result. For everything else there are named scenarios and an exit-cap sensitivity grid.

Yes. Fifty levels of undo, and named versions you can save, compare and restore. A scenario is a full model state rather than a copied spreadsheet.

It reads your documents into structured data, classifies general ledger lines, and can drive the model on request: change assumptions, build a renovation program, size a loan, configure a waterfall. What it does not do is produce a number you cannot trace. Every action it takes lands on the same visible line items you edit by hand, and you can see and change the result.

Yes. Pencil exposes the model over the Model Context Protocol, so you can connect it to the assistant you already use and drive your deals from there. The same tools the in-app agent uses are available to it, 22 of them, from creating a deal to importing a rent roll to reading a full model summary.

No. Agent actions are explicit and reported: you see what changed and the effect on the headline metrics. Nothing is edited silently in the background.

Only PDF extractions, meaning a rent roll or operating statement that has to be read page by page. Spreadsheet imports are unlimited on every plan: Excel and CSV rent rolls, T12s, rehab trackers, commercial rent rolls and rentable-item directories never count. Neither does manual entry, model editing, or exporting.

No, and that is the point. A recipient opens the link with no account and can change your assumptions in their browser to watch the returns move. A read-only link competes with a printed offering memorandum. A modelable one competes with sending someone a model file.

Never. Their edits live in their browser. Your copy is untouched and you are not notified of individual changes. The banner on the shared view tells them so, so nobody has to guess whether they are about to alter someone else's work.

Yes. Saving a copy creates a deal in their own account, frozen at the numbers you sent. If you later change your model, their copy does not move, so their analysis never shifts underneath them. Their copy carries no access to your original, and sharing it onward creates a new link to their deal rather than widening access to yours.

Yes. Choose which tabs a given link can open, set an access code, and set an expiry. Sensitive material is withheld on the server rather than merely hidden in the interface, so a share that is not allowed to show the partnership economics never receives them.

Yes, and there is no equivalent when you email a file. For each link you see how many recipients saved their own copy and which of your assumptions they changed most. A file that leaves your building tells you nothing. A link tells you that four buyers all pushed the exit cap.

Not to open your link, and not to re-underwrite it in the browser. Saving a copy into their own account requires an account. A subscription is for running your own pipeline in Pencil: importing your own deals, using the agent, exporting.

A branded seven-sheet workbook: Cover, Pro Forma, Assumptions, Debt, Rent Roll, Historicals and Returns. Operating rows, debt service, exit value and the return calculations are real formulas that recalculate natively when you change a driver. No add-in, no macros, nothing to install.

Yes. If a lender, partner or committee requires their workbook, Pencil can populate that file from your deal rather than handing you a different one to reconcile. The tie-out is checked, so you know the filled template agrees with the model it came from.

Your workspace becomes read-only and everything stays visible and exportable, including the full live-formula workbook. Your data is never held hostage. Resubscribe and you continue where you left off.

Everyone in a workspace shares deals, general ledger memory, chart-of-accounts templates, comparables and portfolios, and works the same pipeline on a board or a list. You invite by link and assign roles, and you pay for the seats you use.

In a dedicated Postgres database with row-level security. Workspaces are isolated at the database layer rather than by application code alone, and that isolation is tested: a harness attacks every interface as one tenant against another tenant's resources and fails the release if anything returns data.

No. The trial runs 14 days with full team features, unlimited spreadsheet imports and 10 PDF extractions, plus a seeded demo deal you can explore and break. You subscribe inside the app when you are ready.