Manual

Step-by-step guide using a worked example

This manual walks through every step of the unified 11-step Value-your-company workflow using ExampleCo — a hypothetical B2B SaaS at Series A scale. Click Load ExampleCo below to pre-fill every field; then follow the chapters and observe how each panel reacts. Edits made in one step propagate across the framework via the cross-engine sync (badges pulse when a sync fires); the TRL assessment at Step 7 shares its state with the Parameters lab embed via the same persisted Zustand store.

Open the workflow

00Before you start

The simulator is a computational companion to The Cost Gradient of the Build (de Miranda Neto, 2026). The paper's core thesis: commoditization is layer-specific, not AI-wide. The framework decomposes the knowledge-production stack into seven layers, each moving at its own velocity:

L1 Infrastructure — compute, datacenters. Slowly commoditizing.
L2 Foundation models — frontier model weights + APIs. Commoditizing weakly.
L3 Capability access — APIs + tool integrations. Visibly commoditizing.
L4 Codified work — retrieval, synthesis, prototyping. The core erosion layer.
L5 Hypothesis / judgment — problem formulation. Partially protective.
L6 Institutional — trust, certification, distribution. Most protective.
L7 Cross-border modulator — knowledge-integration coefficient K₇ ∈ [0, 1].

The headline number: the gap between the classical Damodaran valuation and the layered (Appendix A) valuation is the commoditization premium — the paper's §7 quantitative thesis. The Reconciliation view (Chapter 10, accessible via the toggle on the Results step) computes it live.

0b. The four modules

The landing page offers four ways into the simulator. The chapters that follow walk through the second one — Value your company — end to end, because it is the most demanding of the four and exercises every panel in the platform. The same panels, the same paper-canonical defaults and the same Advanced lab back the other three modules; once you have read the Value walkthrough you have read most of the workflows the others ship.

Module	What it answers	Who tends to use it
Explore the framework (/demo)	What does the paper actually say, on real figures? Reproduce every chart from Appendices A–G against the paper's two case companies (NeuroCertify, DataFlow Pro), interact with K₇, AI substitution potential and the TRL parameters.	Investors, researchers, students, policy analysts.
Value your company (/value)	What is my firm worth, and how much of the answer is the commoditization premium? Eleven steps from jurisdictions through funding round to the multi-audience PDF — the workflow this manual documents in detail (Chapters 1–15).	Founders, CFOs, M&A advisors, venture capitalists.
Defend your project (/defend)	Should this internal project be funded against the corporate hurdle rate? The 11 steps trim to 9 — same revenue, costs, layer profile and TRL, but capital comes from a corporate WACC plus a project-risk uplift, the funding round drops out, and the output is a board memo with NPV, IRR, payback and a commoditization risk register instead of a cap table.	Intrapreneurs, product leaders, corporate-venture teams, innovation heads, L&D.
Teach a cohort — coming soon	Can I run an assignment for a class? Instructors will create assignments from preset case studies (NeuroCertify, DataFlow Pro) or custom firms, students clone the assignment, work through Value or Defend, submit, and the instructor compares submissions side by side. The module is on the roadmap; the landing page tile is disabled until shipped.	Professors, graduate students, corporate L&D, executive education.

What sits behind all four: every module reads from the same configuration store and the same Advanced parameters lab. A per-token AI cost set in the Advanced lab (Chapter 11b) flows into the Value migration cash-flow, the Defend project payback and the Explore-mode sensitivity figures without any extra wiring. Same paper, same defaults, four readings — Value for the standalone firm, Defend for the project inside a firm, Explore for the framework itself, Cohort for teaching it.

01Jurisdictions

Step 1 of the unified workflow. Pick every jurisdiction where your firm operates payroll or hires engineers. At least one is required. The framework computes the substitution premium differently in each jurisdiction — Brazil's CLT, France's CDI, and the US's W-2 have different labor-cost multipliers, termination costs, AI-service overheads, and corporate tax rates.

Why this is Step 1: the jurisdiction gate informs every downstream step. The corporate tax rate on the Capital step (5) auto-fills from the primary jurisdiction (US 21%, FR 25%, BR 34%). The country risk premium pre-fills similarly. The Funding round step (7) shows Carta benchmarks calibrated for US firms by default; other jurisdictions get an adjustment hint.

Try it now — Open Step 1 — Jurisdictions

02Identity & segment

Step 2 asks you to pick one of five business segments. The rest of the workflow reconfigures around the choice — revenue shape, cost lines, KPI panel, multiples, sensitivity drivers, and layer profile suggestion all change.

Subscription / SaaS

Three sub-variants: single plan, multi-tier, cohort-based. MRR/ARR-driven.

Marketplace / Transactional

Two-sided. GMV × take_rate. Network-effect modulated.

Hardware / Manufacturing

Units × ASP. BOM-dominated COGS, ASP erosion.

Services / Consulting

Consultants × utilization × billable rate. Most L4-exposed.

License / Perpetual

Deal-based + maintenance tail. Hybrid recurring.

ExampleCo is a Subscription / SaaS firm (single-plan variant). The wizard auto-fills the rest of the form with paper-canonical SaaS defaults — you only override what's firm-specific.

Try it now — Open Step 2 — Identity

2b. When each segment fits

If you're unsure which segment to pick, this table shows the decision rule + what each segment tends to look like in valuation.

Segment	Pick when…	Typical EV/Rev
Subscription / SaaS	≥80% of revenue is monthly/annual recurring per customer + retention is measurable
	6× (post-2022 reset)	L4=0.35
Marketplace	You take a % cut of transactions you don't fulfill yourself
	0.4× of GMV / 5× of net revenue	L4=0.20, L6=0.30
Hardware	Physical units shipped; BOM ≥30% of revenue; ASP-driven
	2.5× of revenue / 12× EBITDA	L4=0.15, L6=0.40 (highest)
Services	People-driven; revenue = hours × rate × utilization
	1.8× of revenue / 1.5× bookings	L4=0.60 (highest)
License	Deal-based with attached maintenance contracts; renewals matter
	3.5× of revenue / 6× maintenance ARR	L4=0.30, L6=0.25

Hybrid firms: if your firm is 70% SaaS revenue + 30% services revenue, pick SaaS as the primary segment and note the services line as part of S&M / customer success in the cost structure. Adding multi-stream support would require a new HYBRID revenue variant — open as a future iteration.

The L4 column is the key: high-L4 firms get the largest commoditization premium discount in the Reconciliation step (Chapter 10). Services firms see −50% or more; hardware firms see less than −10%. The pattern follows the paper's §7 thesis exactly.

03Revenue

For ExampleCo (SaaS, single plan), the wizard asks for 8 fields:

Field	ExampleCo	Meaning
Starting customers	200	Paying customers at t=0
ARPU monthly	$149	Avg revenue per user. Total MRR ÷ customers = $29.8K initial MRR ≈ $358K ARR
New customers month 1	30	Acquisition seed for the projection
Customer growth rate monthly	5%	M/M growth of new acquisitions, NOT total customers
Monthly logo churn	2%	≈ 21.5%/yr — mid-market SaaS. Below 1% = enterprise; above 4% = SMB
GRR	90%	Best-in-class floor. Below 80% = bucket leaks faster than you can fill
NRR	115%	World-class is ≥120%. Must be ≥ GRR (validated server-side)
Annual price increase	3%	Inflation-pace. Applied yearly to ARPU

Why MRR/ARR/NRR matter: ARR is the headline SaaS metric. NRR > 100% means existing customers grow faster than they churn — the company can grow even with zero new acquisition. Public SaaS comps in the 8-12× EV/ARR range typically have NRR > 115%.

Try it now — Open Step 3 — Revenue

04Costs

COGS (Cost of Goods Sold) for SaaS is segment-specific — hosting + third-party data + customer support + payment processing. OpEx (S&M + R&D + G&A) is universal across all five segments.

Line	ExampleCo	Layer-4 exposure?
Hosting	8%	L1 — partially compressible by efficiency gains
Third-party data	3%	L2 — commoditizes with foundation model prices
Customer support	4%	L4 — AI-substitutable (tier-1 chat)
Payment processing	2.5%	L1 — fixed Stripe/Adyen %
S&M	40%	L4 — partly substitutable (SDR/ad ops)
R&D	25%	L4 — engineering, the deepest substitution
G&A	15%	L4 — back-office, modest substitution

ExampleCo's gross margin = 1 − sum(COGS) = 100% − 17.5% = 82.5% — in line with the public SaaS median of ~78%.

Why the Layer-4 column matters: in the Reconciliation step (Chapter 10), you'll apply the paper's L4 compression to these cost lines. The L4-exposed lines (S&M, R&D, G&A) drop; the L1/L2/L6 lines stay put. That asymmetry generates the commoditization premium.

Try it now — Open Step 4 — Costs

05Capital structure (WACC + terminal)

The CAPM build-up. WACC = w_E·Ke + w_D·Kd·(1−t) where Ke is the cost of equity from CAPM:

Ke = Rf + β_L · ERP + CRP + size_premium
β_L = β_U × [1 + (1 − tax) × (D/E)] (Hamada re-levering)

For ExampleCo: Rf=4.2%, ERP=4.6%, β_U=1.20 (Damodaran SaaS), CRP=0 (US-domiciled), tax=21%, D/E=0 (typical SaaS). The wizard shows a live WACC preview as you edit — for these inputs it lands at ~9.72%.

The badges ↔: notice how Rf, ERP, β, terminal growth, and corporate tax carry small chips next to their labels saying things like "↔ macro_context (§B.5)". These are live sync points — editing here propagates to the four-path framework store; editing the framework Configuration panel propagates back. The chip pulses for a moment whenever a sync fires.

The g < WACC guardrail: the terminal growth rate must be strictly less than WACC. Try setting g = 30% — the backend will reject the payload immediately. This is a real-world classic mistake; the wizard catches it.

Try it now — Open Step 5 — Capital

06Layer profile

Step 6 — the dedicated page for layer exposure + K₇ + AI substitution potential. The L4 share (codified work) is the single most predictive driver of how much commoditization premium the framework applies. Auto-fills from the segment registry suggestion when you arrive; tune to your firm.

Layer-4 share slider — the AI-substitutable core. Services typical: 0.60. SaaS: 0.30-0.40. Marketplace: 0.20. Hardware: 0.15. The other six layers auto-renormalize proportionally so the sum stays at 1.0.
K₇ presets — globalized 1.0, current 2026 0.7, fragmented 2030 0.4. Same-bloc collapse threshold near K₇ = 0.45; inversion premium contracts ~38% when frontier-model dependencies cross blocs (paper §4.1).
AI substitution potential — fraction of L4 work AI can substitute. Paper §6.4 default 0.60. Combined with L4 share: compression = layer4_share × ai_sub × amplifier_base.

Why this step is upstream of Results: the layer profile is what makes the layered (Appendix A) DCF different from the classical Damodaran. Without an explicit layer profile step, the paper's §7 commoditization premium would be opaque. By putting it before Funding and Sensitivity, we make it clear that capital decisions and risk analysis are downstream of the firm's exposure structure.

Try it now — Open Step 6 — Layer profile

07TRL assessment

Step 7 — the multidimensional readiness profile. Five layers scored live as you answer evidence-based binary questions: TRL (technology, ISO 16290 + NASA Handbook), MRL (manufacturing, DoD Deskbook), CRL (commercial, ARENA), IRL (integration, Sauser et al.), optionally ML-RL (Lavin et al. when the tech triage flags AI/ML), plus a regulatory-gap pass over ISO 26262 / 21448 / 8800 and a Hype-cycle position.

The engine refuses to grant a higher level when lower-level criteria are unmet — combats the self-assessment upward bias that makes most informal TRL scoring useless. A downgrade notice appears whenever a level is rejected; the affected SRL (system readiness, computed) falls accordingly.

Mode selector — Quick (~25 questions, binary), Standard (~50, evidence optional), Audit (~75+, evidence MANDATORY on every "yes"). Pick once; the questionnaire reconfigures.
Triage — three pill groups (tech type, sector, safety criticality). Filters the questionnaire and the regulatory norms shown, so an aerospace + safety-critical project sees ARP 4754A / DO-178C; a SaaS shows the AI/ML lane via Lavin et al.
Live profile panel — radar chart over 5 axes; cross-diagnostics (industrialization gap, certification blocker, hype-bubble risk); capital-needs alerts mapping the profile to typical funding-stage brackets (seed / Series A / Series B+) with budget order-of-magnitude language.
The toggle — once you have a profile, the welcome bar lets you flip Apply TRL adjustment to valuation. When ON, the engine adds a TRL-derived ERP uplift, regulatory-gap country-risk uplift, and an MRL/CRL size-premium uplift to the WACC inputs BEFORE the firm-valuation request ships. The Results step renders the per-line breakdown.

Two places, one store: the TRL questionnaire also lives at /value/deep-dive/trl (Parameters lab → TRL assessment). Both pages render the same components against the SAME persisted Zustand store — edit an answer here, refresh the deep-dive, the answer is there. No sync layer, no copy-pasting.

What the assessment is NOT: a calibrated financial discount. The TRL → WACC adjustments are heuristic priors documented in docs/trl-methodology.md (table at §3.2). They're defensible directionally (low-TRL technology DOES carry more uncertainty), not statistically anchored. Re-tune them for your sector if you have better evidence.

Why TRL matters at Step 7: the profile informs both the discount rate the firm-valuation engine applies AND the funding-round narrative on Step 8 (a TRL 4 firm raising Series B looks suspicious; a TRL 7 firm raising seed leaves money on the table). Surfacing it between Layer profile and Funding round is the natural place — you've told the framework what the firm IS; now you tell it where the firm STANDS.

Try it now — Open Step 7 — TRL assessment

08Funding round

Optional — skip if you just want EV and equity value without a per-share price. For ExampleCo: $30M pre-money, raising $5M, 10M founder shares, 10% post-money option pool.

Stakeholder	Ownership %
Founders	≈ 71.4%
Option pool (PRE-money convention)	10%
Series A investors	≈ 14.3%
Note holders	0%	(no convertibles in this round)

Why option pool is sized pre-money: standard VC practice — the pool comes out of founder dilution, not investor dilution. If you flip it, the investor effectively pays for the pool.

Try it now — Open Step 8 — Funding round

09Sensitivity

Skip with defaults (±20% triangular around each driver's base value, 10k Latin-Hypercube simulations) unless you have specific anchoring views. The wizard exposes segment-aware drivers — for SaaS, you get a different set than for hardware.

Segment	Top tornado driver (typical)
Subscription / SaaS	cac
Marketplace	sales_marketing_pct
Hardware	cogs_bom_pct	BOM is ~55% of revenue → huge sensitivity
Services	cogs_consultant_compensation_pct	Consultant comp IS the product
License	deals_growth_rate_monthly	Deal velocity drives the maintenance ARR tail

Try it now — Open Step 9 — Sensitivity

10Results + reconciliation

The Results step computes the deterministic DCF on every input change (debounced 500ms). For ExampleCo with paper-canonical defaults, you should see:

EV — DCF: in the low millions USD (Series A scale, 60-month horizon)
EV — Multiples × Revenue: peer median 6× × $358K ARR ≈ $2.1M
EV — Multiples × ARR: same anchor as above
Implied EV/Rev TTM: the cross-check — should be in a similar range to peer median if your inputs are sane
WACC: ≈ 9.72%
Terminal % of EV: typically 60-80% — confirms the standard finding that "terminal value is what matters"

Five live charts:

Monthly projections — revenue, EBITDA, cumulative cash
WACC waterfall — Rf → +βL·ERP → +CRP → Ke → blend
Cap table donut — post-round ownership
Monte Carlo histogram (after "Run Monte Carlo") — P5/P50/P95
Tornado bar chart (after "Run tornado") — drivers ordered by swing

Rule of 40 + Magic Number: two SaaS KPIs the wizard reports. Rule of 40 = YoY rev growth % + EBITDA margin %. ≥40 is "public-grade efficient growth." Magic Number = (ΔARR × 4) / S&M last quarter. ≥1.0 means sales spend recovers in ~12 months.

Reconciliation toggle. The Step 9 page has a toggle at the top that switches the body content between two lenses on the same firm:

Segment view — your firm-valuation DCF + multiples + segment-aware KPIs + the 5 live charts
Reconciliation view — six EVs side by side:
- SaaS DCF baseline — your wizard's output
- SaaS DCF (L4 compressed) — same engine after applying paper §6.4 + Eq. B.13 to the L4-substitutable cost lines
- Classical Damodaran — textbook single-rate DCF anchor
- Layered (Appendix A) — seven-layer DCF with TRL-modulated discount + per-layer risk-premium vector
- Two-phase (Appendix B) — phase-conditional WACC under post-AI double-valley dynamic
- Dual-channel (B.2.6) — adds the revenue-retreat correction (Eq. B.12-B.15)

Two headline deltas in the reconciliation view:

L4 compression Δ = SaaS DCF baseline − SaaS DCF compressed. Negative when compression frees margin (typical SaaS: cuts S&M+R&D+G&A by ~10-15%, lifts EV).
Commoditization premium = Layered − Classical Damodaran. The paper's §7 headline. Positive when L6 (institutional) protection dominates; negative when L4 (codified) exposure dominates.

Pattern across segments: services (L4=60%) gets the largest negative commoditization premium (−54% in our default calibration). Hardware (L6=40%) gets the smallest (−8%). The ordering is exactly what the paper predicts: high-L4 firms get re-priced; high-L6 firms hold their value.

Try it now — Open Step 10 — Results

11Two valleys + risk

The framework's post-AI double-valley charts (paper §6.5 + Appendix B). The classical Gartner hype cycle has one valley (months 6-18). The post-AI version has two:

Valley 1 (months 4-10): the classical disillusionment dip — investors realize the prototype isn't the product.
Valley 2 (months 22-36): the commoditization valley — competitors close the technical gap by integrating frontier-model capabilities. Revenue retreats; the second dip is often deeper than the first.

The paper's Appendix B reformulates CAPM, WACC, EVA, ROI, and Gordon perpetuity under this double-valley dynamic. The classical constant-parameter formulas leave the second valley invisible. The two-phase + dual-channel paths in the framework make it explicit.

When to care about Valley 2: if your firm is L4-heavy AND in a fast-commoditizing category (SaaS, services, license), the Valley 2 drag is a material risk. Hardware and marketplace firms (L6-heavy) feel it less. Use the Reconciliation view to see how much of the EV decline lives in the dual-channel layer.

Try it now — Open the Hype-cycle figures

11b. Per-token AI cost (Advanced lab)

The framework needs one number to talk about the cost of AI tooling: how much your firm spends, per developer, per year, on the assistants and copilots that make a substitution scenario possible. The paper's May 2026 reference figure is US$ 12,000 per developer per year — a heavy-use profile under early-2026 Anthropic Sonnet list prices. That single aggregate then feeds the Migration cash-flow, the payback month, the break-even quarter, and the inversion premium.

Pinning a flat dollar figure for the whole horizon misses what the paper itself flags in §10 Limitation 4: list prices keep moving, providers and contracts vary widely, and a heavy user of Anthropic Opus 4.7 doesn't pay the same as a self-hosted open-weight stack. The Per-token AI cost group in the Advanced parameters lab lets you describe that aggregate bottom-up — from the same numbers you read off your provider contract — and the simulator rebuilds the yearly figure on the fly.

What you set:

Input tokens — list price per million (USD). What the provider charges to read a million tokens of context.
Output tokens — list price per million (USD). What it charges per million tokens the model writes back. Typically 4-5× input across the major catalogues.
Input tokens per developer, per month (millions). Roughly how much context a developer pushes through the model in a month — agentic assistants burn through more than people expect.
Output tokens per developer, per month (millions). Generally a fraction of the input volume — code generation, planning notes and refactor diffs together rarely top a fifth of what the model had to read.
Provider you're modelling. A free-form label saved with the scenario; doesn't change calculations.

Four preset buttons snap you to common 2026 contracts (Anthropic Sonnet 4.6, Opus 4.7, Haiku 4.5, a self-hosted open-weight stack). The panel shows What this works out to live, and the same derived number then appears, locked, in the migration-global box that used to ask for the flat dollar figure — the page tells you so explicitly. Clear all token-pricing fields and the box turns back into a plain dollar input.

What changes downstream: the simulator's own smoke test on the paper's 50-engineer / 60% substitution / Brazil reference firm makes the link visible. With the legacy US$ 12,000/year aggregate the firm sees ~US$ 983k/year of net saving and breaks even in Q14. With the Sonnet 4.6 preset (≈ US$ 14,400/year) net saving drops to ~US$ 863k and break-even slips to Q16. With a self-hosted open-weight stack (≈ US$ 3,200/year) net saving climbs to ~US$ 1.43M and break-even comes in at Q10. With the Opus 4.7 preset (≈ US$ 21,600/year) the substitution scenario stops paying for itself — the firm never breaks even within the 20-quarter horizon. The choice of contract shows up in the inversion premium too.

What stays in place: the per-year price trajectory (the paper-cited ~10×/year decline since 2023, exposed separately in the AI service pricing trajectory group) and the per-country AI service overhead continue to stack on top of whatever per-token cost you set here. Per-token pricing only fixes the year-0 aggregate — everything that the framework already does with that aggregate is preserved.

Try it now — Open the Advanced lab

13Common pitfalls

Five real-world traps the validators catch. Each is a production bug the wizard has already encountered (or anticipates).

⚠β = 0.0137 (the percent-decimal confusion)

Cause: Someone enters Damodaran's β as 1.37% instead of 1.37 — off by two orders of magnitude. WACC collapses to ≈ Rf, EV explodes.

How the wizard catches it: Pydantic clamps unlevered_beta to [0.3, 3.0]. The wizard rejects 0.0137 immediately.

How to avoid: Damodaran's SaaS β is around 1.20. If your input has a decimal point in front, you almost certainly meant the integer version.

⚠NRR < GRR

Cause: GRR (retention ignoring expansion) must be ≤ NRR (retention including expansion). If you enter NRR=0.85 and GRR=0.90, the math is inconsistent — you can't have less revenue with expansion than without it.

How the wizard catches it: Pydantic model_validator rejects NRR < GRR at request time.

How to avoid: Best-in-class B2B SaaS: GRR ≥ 90%, NRR ≥ 120%. If you don't track them separately, start with GRR = NRR (no expansion).

⚠g ≥ WACC

Cause: Terminal growth rate must be strictly less than WACC, or the Gordon growth formula blows up (denominator → 0).

How the wizard catches it: ValuationRequest validator computes WACC from inputs and rejects g ≥ WACC. The wizard's WACC step shows a live warning.

How to avoid: g ≤ long-run nominal GDP growth (≈ 2.5-3.0%). WACC for SaaS lands around 9-12%. Plenty of headroom.

⚠ΣCOGS ≥ 100% of revenue

Cause: Adding up all COGS lines crosses 100%. Implies negative gross margin — possible early-stage but rejected by the engine.

How the wizard catches it: Each segment's costs schema has a model_validator that rejects sum ≥ 1.0.

How to avoid: The COGS panels show a live gross-margin number at the top — watch it. SaaS typical: 15-25% COGS; hardware: 60-70%.

⚠Segment mismatch

Cause: identity.segment = 'hardware' but revenue.revenue_model_type = 'SUBSCRIPTION_SINGLE'. The schemas would still match individually but the cross-discriminator is inconsistent.

How the wizard catches it: ValuationRequest._segment_consistency validator catches this and rejects with a clear error message.

How to avoid: The wizard's Step 1 (Identity) calls applySegmentDefaults() which pre-fills revenue + costs with the matching segment shapes. Stick with the wizard and this never happens.

Why we validate aggressively: a spreadsheet DCF model that swallows 0.0137 silently produces an EV off by 10×. The framework rejects 7 categories of inputs that real-world models routinely accept. The cost is a brief error message; the benefit is that every valuation that runs to completion is at least internally consistent.

12Multi-audience PDF

The framework workflow at /value/report generates a multi-audience PDF with the same numbers rendered six different ways — one section per audience role. The same firm config, the same four EVs, six different framings.

Section	What it surfaces
Investor brief	EV bands, four-path reconciliation, MOIC vs IRR, K₇ sensitivity, downside cases
Founder brief	Equity at exit, dilution path, Berkus / VC method anchors, runway implications
Regulator brief	Fiscal-bloc implications (App. D.6), jurisdictional substitution premium, employment effects
Researcher brief	Layer-decomposed risk premium components, K₇ sensitivity table, calibration choices
Technologist brief	L4 / L5 / L6 exposure of the firm, AI-substitutable cost lines, build-vs-buy posture
Journalist brief	The headline number, the commoditization premium in dollars, top-3 risks in plain language

The PDF is academic-grade: embedded figures are vector SVG (re-scalable, paper-quality), the typography mirrors the working paper, and a per-figure download menu lets you export individual charts as PNG/SVG outside the report. See the figure-download menu in the chart corner.

What's NOT in the PDF (yet): the firm-valuation wizard's segment-aware KPIs (NRR, Magic Number, Utilization, etc.) aren't in the multi-audience report — those live in /value/valuation/results and are exported as a separate scenario JSON via the "Export scenario JSON" button on that page.

Try it now — Open the multi-audience report

14Saving & sharing scenarios

Three persistence layers, in increasing order of permanence:

Auto-save (localStorage) — every field you edit is written to your browser's localStorage immediately. Close the tab, reopen later: the workflow picks up where you left off. The save indicator (✓ Auto-saved locally) is always visible at the top of the workflow.
Export scenario JSON — the Export scenario JSON button on the same save bar packages the entire workflow state (jurisdictions + identity + revenue + costs + capital + layer profile + funding round + sensitivity + last computed response) into a single {firm-name}-scenario.json file. Hand this file to a collaborator or store it in version control; re-import via the Import button on the same bar.
(Future) Database + auth — a planned addition ships server-side persistence with user accounts and shareable URLs. Today's export-JSON contract is the bridge: the same shape will round-trip into the future DB.

Two stores under the hood. The wizard state lives in firm-valuation-v1 and the framework state in value-store in localStorage. The cross-engine sync (introduced in Chapter 5 via the WACC sync chips) keeps them aligned. Clearing browser data (Application tab → Storage → Clear) resets the workflow.

Change history. Every field you edit is also logged to the History tab (top-right corner of the workflow chrome). The history shows previous → next per field with timestamp + source store; capped at 500 entries; can be paused or cleared. Useful when a number drifts off where you wanted it and you need to trace back what you changed.

Try it now — Open the change history

15Operating modes — local, cloud, and the local vault

The simulator can be used three ways. They are equivalent for the framework itself; the difference is who can read your data and where it lives.

Local mode (default, no account) — scenarios, TRL answers, change history, and parameter overrides live in your browser's localStorage and nowhere else. Works offline. The operator collects nothing. Limitation: anyone with access to your browser profile can read the same localStorage; clearing browser data destroys everything.
Cloud-synced mode (account, opt-in) — same plus the data also lives on the operator's Postgres so you can resume on a different device. The data-collection minimum is documented in /privacy §2.
Local mode + vault (no account, encrypted at rest) — local mode with a passphrase. Every persisted blob is AES-GCM-256 encrypted with a key derived via PBKDF2-SHA-256 (210 000 iterations) from the passphrase. The passphrase never leaves the device; we have no copy, no recovery path. Designed for shared computers.

On a shared computer: activate the local vault (shield icon in the header). Pick a passphrase you can remember but that nobody else would guess. The vault auto-locks after 30 min idle, so a walked-away computer locks itself before the next person sits down. If you forget the passphrase, your data cannot be recovered — there is no backdoor.

Switching modes: creating an account never destroys your local data; the sidebar adds a one-click "Upload N local scenario(s) to my account" button. Signing out keeps the local data intact. Activating the vault encrypts the existing plaintext in place; destroying the vault decrypts it back to plaintext (if unlocked) or wipes it (if you forgot the passphrase).

Backup: the scenarios sidebar has Download / Upload icons in its header that export every local scenario to a single JSON file (or restore one). Works for both local and cloud entries (the cloud ones already have server-side retention; the JSON is mostly useful for local-mode users who want a portable backup).

Glossary

ARPU: Average Revenue Per User per month. Total MRR ÷ active customers.
MRR / ARR: Monthly / Annual Recurring Revenue. The headline subscription metric.
GRR / NRR: Gross / Net Revenue Retention. GRR ignores expansion; NRR includes it. Best-in-class B2B SaaS: GRR ≥ 90%, NRR ≥ 120%.
CAC / LTV: Customer Acquisition Cost / Lifetime Value. LTV/CAC ≥ 3× = healthy; ≥ 5× = world-class.
Rule of 40: YoY revenue growth % + EBITDA margin %. ≥ 40 = strong.
Magic Number: (ΔARR × 4) ÷ S&M last quarter. ≥ 1.0 = sales spend pays back within ~12 months.
DCF: Discounted Cash Flow valuation. PV of future FCFF + PV of terminal value.
FCFF: Free Cash Flow to the Firm. NOPAT + D&A − CapEx − ΔNWC. Note: NOT EBITDA.
WACC: Weighted Average Cost of Capital. Discount rate for FCFF in DCF.
CAPM: Capital Asset Pricing Model. Ke = Rf + β·ERP. β_L is the levered beta after Hamada re-levering.
ERP: Equity Risk Premium. E[Rm] − Rf. Damodaran implied: ~4.6% (2025).
Terminal value: PV of cash flows past the explicit forecast horizon. Often 60-80% of total DCF EV.
Berkus method: Pre-revenue valuation by scoring 5 risk dimensions. Each capped at $500K. Total = company value.
VC method: Backwards: target IRR × exit value ÷ time to exit = required ownership %.
K₇: Knowledge-integration coefficient ∈ [0, 1]. Paper §4.1. 1.0 = globalized; 0.7 = current (2026); 0.4 = fragmented (2030).
L4 compression: Paper §6.4 + Eq. B.13. AI substitutes L4 codified work; the corresponding cost lines drop. Re-prices the firm.
Commoditization premium: Layered (App. A) − Classical Damodaran. The paper's §7 quantitative thesis.
Double valley: Paper §6.5 + App. B. The post-AI Gartner hype cycle has TWO valleys: classical disillusionment (months 4-10) + commoditization valley (months 22-36).