GeneDx Holdings [WGS] — Financial Model Notes

Task 2 · WGS_Model.xlsx · six tabs, live formulas · FY2019A–FY2035E Built 2026-07-27 · notes written 2026-07-29 · investment-memo v1.4.2

Scope note. These notes were written on 2026-07-29 from the workbook and WGS_Valuation.md as they stood on 2026-07-27. No assumption was changed, added or re-run to produce them. Every figure below was re-derived by independent formula evaluation of the workbook (Python formulas package, which re-parses and re-evaluates every formula from scratch rather than reading back cached values) and reconciled against the valuation note. All sixty-odd checked outputs tie exactly. Where a number here is more precise than the valuation note's rounding, both are stated.

1. Structure

The model is deliberately small. It is not a three-statement build: there is no forecast balance sheet, no forecast cash-flow roll and therefore no balance check. That is a design choice with a reason, stated in §7.

Formatting follows the standing convention: the top line of each statement block carries the dollar sign; rows beneath are plain numbers. Per-share and price cells always carry $.

2. Tie-out to the filed statements

Per the v1.4.2 rule — a zero balance check verifies internal consistency, not input accuracy — the actual-year lines are tied to the filed statement, not to internal coherence. The tie-out is asserted on the Historicals tab itself (rows 19 and 20), not only in prose.

Q1-2026 (10-Q, 2026-05-04): Revenue 102,254 ✓ · Gross profit 68,211 ✓ · R&D 19,804 ✓ · SG&A 74,591 ✓ · Impairment 31,287 ✓ · Loss from operations (57,471) ✓ · Net loss (63,316) ✓.

2.1 The discontinuity is modelled as a hard break, not smoothed

Historicals row 2 carries the instruction in the workbook: "FY2019–FY2022 = Sema4 … FY2023 = transition year … FY2024–FY2025 = GeneDx continuing operations only. DO NOT TREND ACROSS 2022/2023."

This is enforced, not merely noted. No forecast driver anywhere in the workbook links to a pre-FY2024 cell. Every Scenarios growth path anchors on Assumptions!$B$20 (FY2025A volume) and Assumptions!$B$23 (FY2025A revenue) — both FY2025 figures. The FY2019–FY2023 columns exist to be read, and to make the break visible, and for nothing else. The apparent revenue "decline" from $234.7m (FY2022) to $202.6m (FY2023) is the legacy Sema4 wind-down and the apparent gross-margin "explosion" from −11.4% to +44.4% is a mix change; neither is allowed to touch a forecast.

Note the FY2019–FY2022 columns carry combined General and administrative (row 10) while FY2023–FY2025 also carry a separate Selling and marketing (row 9) and an SG&A combined line (row 11). The company's line-item presentation changed; the model reproduces both presentations rather than forcing one onto the other.

3. The driver architecture — and why it is volume × ASP

This is the single most important design decision in the model, and it follows directly from what happened.

A conventional model carries a blended revenue growth rate. That model cannot represent 2026Q1, in which exome/genome volume grew +34% while revenue grew +27% — a decoupling that a single growth rate collapses into one number and hides. So every scenario is built as:

E&G revenue  =  E&G volume  ×  E&G ASP        (Scenarios rows 7-8, 29-30, 51-52)
Total revenue =  E&G revenue  +  Other revenue  (rows 10, 32, 54)

with volume growth and ASP forecast independently, year by year. The consequence is that the bear case is not "growth is slower"; it is "volume compounds and price does not", which is the actual mechanism.

Other revenue — the shrinking residual of legacy hereditary-cancer and other panels, the Fabric software business, and episodic biopharma data deals — is entered as a hard dollar figure per year rather than grown off a rate. It fell from $17.0m (25Q4) to $11.7m (26Q1). Modelling it as a percentage of a growing total would have made a decaying line grow.

3.1 The KPI_ASP bridge — the derived series the company does not disclose

ASP is computed, not disclosed. The company reports E&G revenue and E&G test-result volume every quarter in the 8-K Item 2.02 exhibit; it has never used the words ASP, average selling price, pricing, denial or collections in seventeen quarters of releases. Row 8 does the division the company does not do:

Re-derived: H8 = $3,847.95, J8 = $3,295.98, J10 = 67.71 days. Peak-to-trough −14.3% in two quarters.

DSO is computed with actual day-counts per quarter, not a flat 90. Row 10 uses 91 / 91 / 92 / 92 / 90 / 91 / 92 / 92 / 90 days for 24Q1 through 26Q1 respectively (= AR × days ÷ total revenue). On a flat-90 convention the 26Q1 reading would be unchanged at 67.7 but 24Q4's would move; the actual-day convention is used so the trend is not a calendar artifact. This is the one earnings-quality metric in the whole file untouched by the discontinuity, the impairment or purchase accounting, which is why it carries the weight it does in Gate 1.

3.2 The guidance decomposition — cells B17:B32, the variant in eight rows

This is the arithmetic the memo turns on, and it uses only company-stated inputs:

The whole variant is B32. The −12% guidance cut moved volume by 4pp (34% → "at least 30%") and E&G revenue by 14pp (34% → "at least 20%") — essentially the entire cut is price — and the reset guide still embeds a +5.7% H2 ASP recovery for which no mechanism has been stated in the release, the 10-Q, or any subsequent 8-K. If H2 ASP simply equals H1 ASP, FY2026 revenue lands near $467m, below the $475m guidance floor — which is exactly where the Base case lands, by construction rather than by coincidence.

Two conservatisms in this decomposition, stated against my own interest. The company guided "at least" 30% and "at least" 20%; the model takes both at the floor, which maximises the implied step-up. And Q2's $100m E&G revenue and ~30,000 tests are the company's own guide, not actuals — if Q2 beats on ASP, B28 rises and B32 shrinks. The variant is therefore at its widest as modelled, and 3 August narrows or widens it with real data.

4. Scenario assumptions — every driver, and where it came from

Bull 25% / Base 45% / Bear 30%. Weights are a judgement, and the bull weight is the one the trade note says it cannot independently defend.

Probability-weighted target $66.71 (+10.2%) = 0.30 × 39.15 + 0.45 × 65.06 + 0.25 × 102.79.

4.1 Three modelling choices worth naming

1. Opex is a dollar path, not a ratio. Adjusted operating expense is entered as an absolute figure in every year of every scenario. Had it been set as a percentage of revenue, the bear case — the one with the lowest revenue — would have automatically enjoyed the lowest opex, which is precisely backwards: the observed 2026Q1 event was opex +48–57% against revenue +17%, i.e. cost that did not flex with the shortfall. The bear case therefore carries the highest opex ($332m FY26 vs the bull's $318m), which is the honest representation of inverted operating leverage.
2. The Base case is deliberately below the guidance floor. $466.8m against a $475–490m guide. This is not an accident of the drivers; it is the variant, expressed as a number.
3. There is no adjusted-EPS line in the workbook. The FY2027 adjusted EPS figures quoted in the valuation note ($(1.31) bear / $0.51 base / $2.95 bull) are derived from the tab's adjusted-EBIT outputs, not computed in-cell. Stated so the workbook is not credited with more than it contains.

4.2 The exit multiples — the weakest link, named as such

The target price in every scenario is FY2027 revenue × exit EV/Sales + net cash ÷ 30.5m shares. The exit multiple is therefore doing enormous work, and the memo says so: 74% of the disagreement with the Street's $81.78 target is multiple, not numbers. The anchors:

• Peer median P/Sales 5.36x; WGS at 4.17x on TTM revenue, 3.71x EV/Sales on FY2026 consensus, 2.98x on FY2027 consensus.
• Base 3.40x sits between WGS's current FY2027 consensus multiple and the peer group's low end (CDNA 4.55x, NEO 2.29x, FLGT 1.74x, MYGN 0.62x).
• Bear 2.20x is the NEO/FLGT zone — where the peer set puts names whose realised economics broke.
• Bull 4.75x is below NTRA (14.78x) and GH (17.86x) but above everything else.

The comps bracket the answer without deciding it, and the honest reading is that this row is an opinion with a range around it, not a measurement.

4.3 Base-rate check on each scenario — carried in the workbook, rows 69–72

Anchored on Chan, Karceski & Lakonishok (2003): growth persistence beyond chance is close to nonexistent, and the modal modelling error is extrapolating the recent rate.

• Base, FY2026–30 revenue CAGR 17.3% off a $427.5m base — roughly a top-third outcome at this scale. Critically it does not extrapolate FY2025's +41%: it assumes decay from 30% to 13%, which is the reference-class prior rather than an override of it.
• Bull, 27.0% CAGR — explicitly top-decile, and named as such in-cell. It requires both 25%+ volume compounding and an ASP recovery to $3,870. Research §6.3 shows those two pull against each other: the marginal patient added at higher penetration is by construction in a worse-reimbursed channel. The bull case is internally in tension, which is why the weight is 25% and not higher.
• Bear, 12.7% CAGR — approximately the median five-year outcome for a company at this revenue scale. This is not a disaster case. It is the base rate with the price problem simply not fixed. Worth stating twice: the "bear" here is an ordinary outcome, not a catastrophe.

5. Cost of capital

Two structural notes on the WACC build, stated because they matter more than the inputs.

1. The debt weight uses the $100m loan face, not market value, and the equity weight uses B9 market cap — giving weights of 94.9% equity / 5.1% debt. At this leverage the WACC is ~99% a cost-of-equity number; the 10.0% cost-of-debt estimate moves it by roughly 1bp per percentage point. The unsourced input is immaterial, which is worth knowing before worrying about it.
2. No tax shield is applied to the cost of debt. With a large NOL position and no cash taxes modelled before FY2030, an after-tax adjustment would be fictitious. This is conservative (it raises the WACC slightly) and deliberate.

Net cash bridge (Assumptions B6:B10): cash + securities + restricted $171.7m (31-Mar-26, company-stated) less the $100.0m Blackstone term-loan face = net cash $71.7m. Market cap $1,846.5m (60.54 × 30.5m diluted FY2027E shares). Enterprise value $1,774.8m.

Note the share-count convention: market cap uses the FY2027E diluted 30.5m, not the 29,666,318 basic shares outstanding at 31-Mar-26. Using the forecast diluted count consistently in both the market cap and the target-price divisor is internally consistent and is the conservative choice on the target — but it does mean the stated $1,846m market cap is ~2.8% above the spot-basic figure. Stated so the reconciliation is not left to the reader.

6. DCF — reported, and deliberately not used

Ten-year explicit forecast, perpetuity terminal, SBC deducted as a real cash-equivalent cost (it was $9.0m in Q1-2026 alone, ~8.8% of revenue), no cash tax until FY2030 on the NOL carryforwards.

Note the working-capital line is negative in every year — a use of cash. That is the DSO problem carried into the forecast rather than assumed away: a business collecting more slowly funds its own growth, and the model makes it pay for it.

Re-derived output:

Why it is reported and weighted 5%, not used. Terminal value is 74.8% of enterprise value, and the first three explicit years contribute −$82.7m of negative PV — so the value net of the loss-making years is effectively all terminal, which is the sense in which the valuation note calls it ~100% terminal. A $13-to-$46 spread on an identical cash-flow stream cannot discriminate, and a method that cannot discriminate must not carry the price target. The workbook says so in-cell (DCF A26). This is the same archetype already logged in CALIBRATION_WATCH.md for TXG / TWST / NTRA / GH. Handling it by disclosure rather than by quietly dropping it is the required treatment.

The one genuinely useful DCF output is the inversion. Solving for the discount rate at which each scenario's stream equals today's $1,775m EV: the house base case implies 7.3%, the house bull case implies 15.0%. A 7.3% cost of capital is implausible for a ~96%-realised-vol, loss-making, single-product small cap. The stock is priced for the bull case, correctly discounted — a valuation-negative observation which, per references/trade-construction.md, is on its own not a thesis.

7. What the model deliberately does NOT contain

8. Known weaknesses, in descending order of how much they matter

1. The exit multiple carries the target. 74% of the disagreement with the Street is multiple, not numbers (§4.2). This is the weakest part of the negative view and the memo grades Gate 2 MARGINAL because of it.
2. The scenario probabilities are a judgement. 30/45/25. The trade note states plainly that the 25% bull weight cannot be independently defended, and that the long-side E[R] of +10.2% "exists only because the bull case carries a 25% weight I cannot independently defend."
3. The pre-tax cost of debt (10.0%) is an estimate — Blackstone terms are undisclosed. Immaterial to the WACC (§5), but it is an unsourced input in a model whose discipline is tie-out.
4. Adjusted EPS is not computed in-cell (§4.1, item 3) — yet the FY2027 adjusted EPS gap of −54% is what carries Gate 2. The gate's load-bearing number is derived outside the workbook. This is the most uncomfortable item on the list and it belongs at this position, not buried.
5. No forecast balance sheet, hence no balance check (§7).
6. Q2-2026 E&G revenue and volume in KPI_ASP column K are the company's guide, not actuals — labelled 26Q2G. Half of the H1 ASP that the entire variant is measured against is therefore a forecast. 3 August replaces it with a fact.
7. The FY2025 exome/genome base ($360.3m on 97,271 tests) is company-disclosed but the FY2025 ASP of $3,704 is derived, as is every ASP in the file. If the company's volume definition changes — a real risk given the reflex product creates a second result on one patient — the derived series breaks and the comparison to history breaks with it. Nothing in the file rules this out.

9. Verification performed

• Every formula re-evaluated independently with the Python formulas package, which re-parses the workbook from scratch rather than reading cached values. All checked outputs across Assumptions, KPI_ASP, Scenarios, DCF and FactorScorecard reproduce the figures published in WGS_Valuation.md exactly — including WACC 14.22%, H2-required ASP $3,505.58, the +5.73% step-up, all three target prices ($39.15 / $65.06 / $102.79) and the $16.44 DCF.
• Two real formula bugs were found and fixed by that check at build time (recorded in WGS_Valuation.md): an H1-2026 aggregation pointing at the FY2024 columns, and an implied-EV formula multiplying the revenue-growth row instead of the revenue row. Both would have passed a zero balance check. They were caught only because the outputs were re-derived rather than read back from the string that wrote them.
• Actual-year lines tied to the filed statements, not to internal coherence (§2).

10. Data provenance

• SEC EDGAR XBRL company facts, CIK 0001818331 — actual-year income-statement and balance-sheet lines.
• 10-K FY2025 (accession 0001818331-26-000015, filed 2026-02-23) — FY2024/FY2025 statements, Infinity treatment, segment absence.
• 10-Q Q1-2026 (accession 0001818331-26-000041, filed 2026-05-04) — Q1 statements, Note 6 (Fabric impairment), MD&A ("a 5% decrease in average reimbursement rates"), third-party payor reserves, AR.
• 8-K Item 2.02 Exhibit 99.1 × 17 quarters (2022Q1–2026Q1) — quarterly E&G revenue and volume (the entire KPI_ASP input set), guidance, and the mention-frequency corpus.
• 8-K 2026-02-27 (Blackstone $100m term loan, $6.6m extinguishment loss); 8-K 2026-06-15 (Gardner appointment).
• Alpaca Markets — daily bars (252 sessions to 2026-07-27) for beta, realised volatility and correlations; options snapshots for the Gate 5 feasibility test.
• Alpha Vantage EARNINGS_ESTIMATES — one call, cached at data/av_earnings_estimates_WGS_2026-07-27.json. The shared 25/day cap was respected; a 17-quarter EARNINGS_CALL_TRANSCRIPT sweep was not available, which is why the mention-frequency corpus is press releases and carries no prepared-remarks vs. Q&A split.
• stockanalysis.com — consensus targets, short interest, float, 5-year beta. Aggregator data, flagged as such.
• ClinicalTrials.gov API v2 and PubMed E-utilities — the unpublished-scoping corpora behind the §7 exclusions.