This is the implementation specification for a decision surface in late-stage private markets. The architectural argument for why the surface is being built, who is racing to host it, and what becomes consolidated around it lives in the companion paper, Bloomberg moment: The race to host the decision. This paper sets out what a decision surface actually is.
The market a decision surface serves is partially built. The transaction tape, the first layer of late-stage private market pricing infrastructure, has been assembled over the last three years across primary references like Forge Price, Caplight Data, Hiive, and Nasdaq Tape D; index references like Hiive50, Notice50, the Forge Private Market Index, and Setter30; aggregate references like Setter Capital reports; fund-mark cross-checks like PitchBook and CB Insights; and broker tapes like Rainmaker and MVP. These products have made the market meaningfully more transparent. That transparency is the foundation the rest of this paper is built on.
The decision surface is the second layer. It is the structure that buyers, sellers, GPs, LPs, auditors, and regulators reference when they make and defend the hardest pricing calls in this market. It does not replace the tape. It does what the tape was never designed to do: produce inspectable, configurable, scenario-decomposed views of fair value that hold up in front of an IC, an LPAC, or a regulator.
The methodology rests on four pillars: traceable data, consistency, persistent company memory, and inspectable, decomposed outputs. Each pillar exists because an institutional decision requires it. An IC approval, an LPAC sign-off, an audit review, a 409A determination. Each of those reviewers applies a specific test, and the methodology has to pass all of them.
This paper sets out the four pillars, the methodology end to end across seven steps, and how the resulting surface interoperates with the tape it complements.
A defensible late-stage private fair value has to survive a specific set of reviewers. Each one applies a different test.
The four pillars are the conditions that have to hold for all five of these reviewers to accept the output. They are not independent. They compose. A methodology that is traceable but not consistent produces work that cannot be replicated. A methodology that is consistent but not inspectable produces work that cannot be defended. All four have to hold together.
Every quantitative input that flows into a fair-value output must be linked back to a primary source: a filing, a transcript, a dated transaction record. The link is preserved through every transformation. Inputs derived from other inputs carry provenance chains back to the original source.
This is the condition auditors and 409A teams test first. It is also the condition generative-AI workflows fail most visibly. They synthesize from a corpus and reconstruct citations after the fact, producing plausible-sounding references that cannot be verified. That is the structural reason AI-generated research does not satisfy institutional traceability standards.
In Arcanis, every datapoint in a company knowledge base carries a structured provenance record: source document, page or timestamp, ingestion date, extraction method, and version. When data is transformed, the transformation is logged. The chain is complete and auditable at every link.
The same inputs must produce the same outputs, every time. The valuation logic has to be encoded as deterministic computation, not generated freshly each run. Two analysts running the methodology on the same inputs at different times must produce identical outputs.
This is the condition LPAC chairs and fairness-opinion reviewers test. It is also the condition that disqualifies generative AI as the core of a pricing methodology. Stochastic models produce outputs that vary by run. That variance is fine for hypothesis generation. It is structurally incompatible with auditable valuation.
Generative AI plays a real role at the edges of this methodology, in data ingestion, surfacing, and narrative explanation. It does not participate in the valuation math. The valuation math is encoded in deterministic models compliant with IPEV guidelines and ASC 820 (FASB's fair value measurement standard for US GAAP reporting).
A late-stage private company is not a one-time research target. The same name comes back every quarter, for a re-mark, a follow-on, a tender, a continuation vehicle, a co-investment. The methodology must produce outputs that accumulate, not ones that start from scratch each time.
This is the condition VC GPs running tenders and continuation vehicles test. It is also the condition that distinguishes a methodology from a research output. A research output is consumed and discarded. A methodology produces a durable knowledge base that compounds in value each time a new run adds to it.
In Arcanis, every company has a company-specific data lake structured by topic vector: revenue, competition, valuation, market, legal, and others. Every research run is saved into the knowledge base and versioned. Subsequent runs query precedent runs first, filling gaps and flagging changes rather than rebuilding from scratch.
The deliverable must be inspectable, decomposed, and stress-testable by the recipient. Every assumption lives in an open Excel cell the user can change. The output is a set of independently-priced scenarios, not a single number with an explanatory narrative.
This is the condition IC members test when they have to defend a price. They cannot defend a single number with no decomposition. They can defend a set of priced scenarios with explicit drivers, explicit risks, and a clear methodology for how the price was derived from both.
In Arcanis, every fair-value output is delivered as an inspectable Excel artifact alongside the company knowledge base. All scenario assumptions, valuation inputs, comparables, and multiples are in open cells. The recipient stress-tests independently, changing drivers and seeing what happens to the price without asking the vendor. This includes the scenario weights themselves. The reviewer can change which scenarios they weight, how heavily, and see the resulting fair value update in real time.
The four conditions above describe properties of the methodology itself. They presuppose an input layer that is as comprehensive as the available information allows. Late-stage private markets are characterized by information asymmetry. Some parties know more than others, and some information genuinely cannot be obtained from outside the company. A defendable methodology cannot pretend the asymmetry away. What it can do is systematically gather every publicly available signal across every relevant company vector: filings, expert transcripts, professional databases, public web, dated commentary, transaction records, and NDA-protected materials when access is permitted. The goal is to resolve as much of the asymmetry as public information allows, and to make explicit which inputs were considered, which were excluded, and what was beyond reach. This is a precondition for the four pillars, not a fifth alongside them.
Institutional customers in this market currently get their fair-value reference from four structural sources: a transaction tape, a generative-AI workflow, an internal team's own model, or the Arcanis methodology described in this paper. Each does something well. The table below shows what an institutional customer gets at each stage from each source.
| Stage | Transaction tape (Hiive, Forge, PM Insights) | Generic AI | Internal team model | Arcanis |
|---|---|---|---|---|
| Data | Real prints — the strongest signal of where the market cleared. Provenance is inherent. | Synthesized; provenance reconstructed. Citations cannot be independently verified. | Sourced by the analyst, generally traceable but not standardized across the institution. | Source-traced ingestion of filings, transcripts, databases, and direct company data. Every input carries a full provenance record. |
| Research | Not the tape's job — it reports prices, not methodology. | Stochastic; outputs vary across runs. Structurally incompatible with auditable valuation. | Built deal by deal in spreadsheet, generally consistent within a team but not standardized externally. | Standard valuation models, IPEVcompliant, encoded as deterministic computation. Same inputs produce same outputs every time |
| Output | A reference price and bid-ask depth. | Narrative outputs without inspectable cells; suitable for synthesis, not for IC defense. | Inspectable to the team that built it; harder to share, audit, or defend externally. | Inspectable Excel. Independentlypriced scenarios with explicit drivers and risks in open cells the recipient can change. |
| Decision | "Here is what just cleared." Decisiongrade for execution. | "A plausible view." Not defensible to an IC or LPAC. | Defensible internally; less so when shown to an LPAC or auditor unfamiliar with the team's conventions. | "Here is what should clear under each scenario, and here is what the last round was underwriting." Decision-grade for underwriting. |
| Over time | Print history accumulates. Strong signal on covered names. | Sessions ephemeral; each run starts from scratch. No institutional memory. | Knowledge lives with the analyst. Often rebuilt when staffing changes. | Company knowledge base updates continuously. Each research run builds on the last. Coverage compounds. |
Each of the first three sources is strong on one or two of the four pillars and silent on the others. The Arcanis methodology is designed to hold all four at once. This is what makes it a decision surface rather than a tool. A surface is what consolidates around itself; a tool is something a buyer uses for one specific job.
Arcanis scenarios are constructed from company fundamentals, not from secondary-market prices.
This is deliberate, and it is the load-bearing design choice of the methodology. If scenarios were derived from secondary-market prices, if a Hiive bid fed the model and the model produced a scenario set calibrated to that bid, the surface would add no information. It would just be the tape, repackaged. The investor would be comparing the tape to itself.
Instead, the methodology produces two independent reference points for the investor to compare: the secondary-market prices observable on the tape, and the scenario prices Arcanis derives from company fundamentals. When the two converge, the market is pricing the company consistently with its underlying operating picture. When they diverge, the gap is information. Either the market is mispricing the company, or the model is wrong. Both are worth knowing.
For platform operators, this is the structural reason a decision surface is additive to a tape rather than redundant with it. The surface does not sit downstream of a tape, re-processing its prices into derived outputs. It sits beside the tape, producing independent signals that become more useful the more they are compared to tape data. A platform that hosts the tape and embeds the surface holds both layers; the platforms that host only the tape carry one.
In addition to the scenario set, Arcanis derives a Baseline for every covered company. The Baseline estimates what the last primary round implicitly priced in: the revenue growth and profitability trajectory the lead investor was underwriting when they set the round price.
The Baseline is independent of secondary-market prices. It is derived from the last primary round's terms and the company's fundamentals at the time of the round. It serves as a reference anchor: it makes explicit what the company would have to do to have been fairly priced at that round, and it allows a secondary buyer to ask a real question: has the company kept up with what the last round was underwriting, or has it fallen behind?
The Baseline is not a price target. It is an estimated view of what the last round priced in, surfaced so it can be evaluated alongside the scenarios it stands beside.
The methodology runs in seven steps, each feeding the next.
Primary sources are ingested into the company-specific knowledge base: filings, expert transcripts, professional databases, public web sources, transaction records, and NDA-protected direct company information where available. Ingestion is fully automated. Every source is logged with a provenance record at intake. Nothing enters the knowledge base without a traceable origin.
In parallel, the methodology runs the standard analytical modules an institutional analyst would run on the company independently: competition analysis, peer benchmarking, precedent transactions, discounted cash flow modeling, and comparable company analysis. These analyses are run identically across every covered company. The output is a structured set of analytical inputs that feeds scenario construction and valuation.
The methodology identifies the full set of risks bearing on the company, including those surfaced through expert opinion, alongside the upside and downside factors currently in force. Each risk and factor is categorized: priced in at the last primary round, or emerged since.
This categorization is load-bearing for what follows. A risk that was already priced in at the last round is structurally different from one that has emerged since. The same is true of upside and downside factors. The mapping produces an explicit, auditable picture of what the institutional buyer is actually taking on at today's price versus what the last primary round was implicitly underwriting.
A late-stage private company has a high-dimensional future. Dozens of risks bear on it. Multiple upside and downside factors interact. The same risk can amplify or cancel another depending on which factors fire alongside it. No human can hold all of this in mind simultaneously, let alone defend a price built on it.
Step 4 collapses that high-dimensional space into a small number of human-understandable scenarios, typically three to five, that an IC member can read, reason about, and defend.
Each scenario describes one independent strategic path the company could plausibly take. Independence, or orthogonality, is the design constraint. Two scenarios should not be variants of each other or different points on the same trend. They should describe genuinely distinct futures, each with its own internal logic, its own driver mix, and its own risk profile.
Concrete example. "Company executes the current plan and exits via IPO at the modelled run-rate" and "company executes the current plan and exits via IPO at a slower run-rate" are not orthogonal. They are two points on the same vector. "Company executes the current plan and exits via IPO" and "company is acquired by a strategic buyer at a control premium" and "company misses its plan and raises a flat or down round before exit" are orthogonal. They describe different paths with different risks and different value mechanics.
Orthogonality is what makes scenario weights meaningful. If scenarios overlap, weighting them double-counts the same outcome. If they are orthogonal, the weights describe the buyer's view of which independent future is most likely.
The risks and factors mapped in Step 3 are the inputs to scenario construction. Arcanis groups risks and factors by which strategic path they belong to. A risk that materially affects the IPO path but not the acquisition path belongs to the IPO scenario; a factor that affects all paths goes to all of them. Inter-dependencies between risks, one risk amplifying another, or one factor cancelling a risk, are resolved at the scenario level, where they are visible and defendable.
Arcanis produces a default scenario set for every covered company. The default set covers the most plausible orthogonal paths and is constructed by the same encoded logic across every company in coverage, so the methodology produces comparable scenario architectures across names. Analysts then review the default set, verify each scenario against the risk and factor map, and configure it as appropriate, refining drivers, sharpening risks, and where the company's situation calls for it, adding or removing scenarios.
This is not a sensitivity table. A sensitivity table varies one input by a percentage to produce a grid; the outputs are not coherent strategic paths and they are not orthogonal. Step 4 produces coherent strategic paths that are orthogonal by construction.
Each scenario carries a default probability weight produced by the methodology. The weights collapse the orthogonal scenarios into a single weighted fair value: a probability-weighted view of where the company should clear today.
The default weights are not the answer. They are a defensible starting point. Different viewers, a buyer with a long time horizon, an LP focused on downside protection, a GP weighing a tender at a specific price, will weight the same scenario set differently and reach different fair values. This is not a flaw of the methodology. It is the structural reason the methodology is configurable.
At IC preparation time, the reviewer can change the weights and watch the fair value update. The orthogonal scenarios stay the same; only the weighting changes. The IC sees three things at once: what the methodology's default weights produce, what the team's reweighted view produces, and what the secondary market is currently pricing in. The gap between any two of those three is information.
This is the methodological reason scenarios beat single-number determinations. A single number forces the analyst to commit to one weighted view of the future and hides every other view inside it. A weighted scenario set surfaces the weighting as an explicit, configurable input, and lets every reviewer see the price under their own view of the future without rebuilding the analysis.
Each locked scenario is priced using standard valuation models compliant with IPEV standards. The same models are applied uniformly across scenarios and across companies. The weights from Step 4 are applied to produce a probability-weighted fair value alongside the per-scenario prices. The weights are configurable; the per-scenario prices are deterministic. The same locked scenario produces the same price on every run.
Every input, scenario, and output is written to the company knowledge base with version history. On every subsequent research run, the system queries the precedent knowledge base by section: revenue, competition, valuation, market, and others. Existing analysis is reused where still valid. Gaps are identified and filled. Changes from prior runs are flagged explicitly. The knowledge base compounds in value with every run.
The methodology's output is delivered as an inspectable Excel artifact alongside the company knowledge base. Every assumption sits in an open cell. The recipient stress-tests independently, changing drivers and seeing what happens to the price, without asking the vendor. The knowledge base is delivered as a structured document that can be read, searched, and cited.
The four pillars apply across the methodology end to end. They are not independent quality criteria but a composed test that the methodology passes, or does not. Traceability, consistency, persistence, and inspectability hold at every step, or they do not hold at all.
The methodology is designed to work next to a transaction tape, not in place of one.
A tape provides real-time market signal: where the marginal buyer and seller are clearing trades right now. That signal is the right reference at the execution decision, what level a buyer can actually transact at, and structurally the wrong reference for underwriting decisions, what scenarios this price embeds and whether a buyer should be underwriting them.
The methodology provides a different set of capabilities: scenario decomposition, long-tail coverage for names that have not recently traded, the Baseline as a primary-round reference anchor, and a per-company knowledge base that accumulates across research runs. These are the things a tape cannot provide, by construction. They come from a different method applied to a different set of inputs.
The two layers do not feed each other algorithmically. They sit side by side. The investor reads the tape for execution-grade signal, reads the scenario surface for underwriting-grade signal, and combines them at the decision layer. If the tape is clearing below the scenario range, the market is pricing in more risk than the model implies, or the model is wrong. Either way, the gap is information.
This is not a novel architecture. Public equities have had it for fifty years: sell-side analyst coverage is the underwriting layer, the consolidated tape is the execution layer, and they coexist because they answer different questions. Late-stage private markets are now in the process of building both layers. The tape arrived first.
The architectural argument behind this, why the consolidation race is now underway and what the implications are for different market participants, lives in the companion paper Bloomberg moment: The race to host the decision.
Late-stage private markets have not yet standardized their pricing infrastructure. Public equities did this decades ago: the consolidated tape on the execution side, sell-side analyst coverage with disclosures and compliance standards on the underwriting side. The private market has now built the first half. The second half is being built now, and the platforms that host it are racing to set the standard around themselves.
This paper describes one approach to that second half. The conditions it sets out, traceable data, consistency, persistent company memory, and inspectable, decomposed outputs, are not Arcanis-specific requirements. They are the conditions any institutional-grade decision surface has to meet, regardless of who builds it. The goal is a market where underwriting methodology is as standardized as the tape that sits next to it, and where the standard, once it forms, is shared across the platforms that host it.
Arcanis is decision surface infrastructure for late-stage private companies. Built and used inside an active VC secondary fund. Available to host platforms via API, embed, white-label hub, and inspectable Excel.
The architectural case for the consolidation race this surface participates in is set out in the companion paper: Bloomberg moment: The race to host the decision.
PitchBook 2025 Annual US VC Secondary Market Watch. Wellington Management VC Outlook 2026. Cambridge Associates 2026 Outlook. Industry Ventures interviews with Hans Swildens (World of DaaS, Transacted, TechCrunch). IMD Business School April 2026 analysis. Manhattan Venture Partners (MVP) April 2026 "Secondary is Primary" panel on the growth of secondaries, with Tom Callahan / NPM, Eric Yi / Citi, Jared Carmel / MVP, and Jason Saltzman / CB Insights. Jefferies H1 2025 Global Secondary Market Review. Equitybee 2025 VC Liquidity Tracker. ILPA Continuation Funds Guidance 2023. SEC Private Fund Adviser Rules August 2023. SEC Investor Advisory Committee September 2025 recommendations. NewView Capital, GSA Ventures, StepStone, Equidam, Sacra, and Chronograph published research. CFA Institute private markets transparency survey. Founders Circle legal guide to secondaries. Charles Schwab and Forge Global press releases (November 2025, March 2026); Goldman Sachs and Industry Ventures press releases (October 2025, January 2026); Morgan Stanley and EquityZen press releases (October 2025, January 2026). Bloomberg LP / Bloomberg Terminal historical and revenue data via public reporting.
This note is market commentary intended for institutional and professional audiences. It is not investment advice and should not be relied upon as the basis for any investment decision. Forward-looking statements involve inherent risks and uncertainties. Actual outcomes may differ materially. Categorizations of named participants are observational and subject to revision in light of input from those participants. © 2026 Arcanis. All rights reserved.
Citation: Prokofyev, A. (2026). The Decision Surface: A methodology for scenario pricing in late-stage private markets. Arcanis Research. https://arcanis.com/research/the-decision-surface/
