Phase Space AI

Financial Model Notes

Harrow [HROW]

Harrow, Inc. [HROW] — Financial Model Notes

2026-07-29 · All figures $000s unless stated · Source: SEC EDGAR CIK 0001360214

This document records every derived figure used elsewhere in the memo, with its construction, so each can be re-checked independently. Where a figure could not be obtained it says so.

---

1. Revenue — quarterly, as filed and as derived

Filed quarterly figures come from the XBRL RevenueFromContractWithCustomerExcludingAssessedTax duration facts. Q2 and Q4 product-level figures are derived by subtraction from the 9-month and full-year disclosures, and each derivation is shown.

TTM revenue at 2026-03-31 = 272,303 − 47,831 + 44,203 = 268,675. Matches the screen exactly.

Product-level derivations (the ones the memo leans on)

From the Q3-2025 10-Q (9-month), the FY2025 10-K (full year) and the Q1-2026 10-Q:

Tie-out: Q2-2025 components sum to 18,336 + 18,641 + 5,212 + 85 + 21,468 = 63,742 = filed Q2-2025 total ✓. Q4-2025 components sum to 35,883 + 25,905 + 12,250 + 82 + 14,972 = 89,092 = derived Q4-2025 total ✓.

Note on the FY2024 branded reconciliation: the FY2024 product lines (49,303 + 28,061 + 37,836 + 995 = 116,195) exceed the disclosed Branded segment total of 116,115 by $80k. The discrepancy is in Harrow's own disclosure, is 0.07% of the segment, and is not adjusted here. Recorded rather than smoothed.

---

2. TTM income statement at 2026-03-31

Constructed as FY2025 − Q1-2025 + Q1-2026 on every line.

Prior-year TTM (through 2026-03-31 minus one year), for the margin-change test: FY2024 8,822 − Q1-2024 (6,928) + Q1-2025 (11,232) = 4,518 on revenue 199,614 − 34,587 + 47,831 = 212,858 → 2.1%. Operating-margin change = +5.2pp. Revenue growth = 268,675/212,858 − 1 = +26.2%.

Prior-prior TTM revenue (through 2025-03-31 minus one year) = 130,193 − 26,103 + 34,587 = 138,677 → prior TTM YoY = 212,858/138,677 − 1 = +53.5%. Growth acceleration = 26.2 − 53.5 = −27.3pp.

Where the screen went wrong

The screen reported op_margin_pct = 11.2 and ev_ebit = 55.2. 11.2% is FY2025 operating income divided by FY2025 revenue, while revenue_ttm in the same record is the TTM figure. Mixing an FY numerator with a TTM denominator produced an EV/EBIT of 55.2x (= 1,662.3 / 30.1). On a consistent TTM basis:

EV/EBIT = 1,670.8 / 19.667 = 85.0x, and operating margin = 7.3%.

The screen's own data_quality_ok flag was already false.

---

3. Balance sheet and net debt at 2026-03-31

Debt — a single instrument. $300,000 face of 8.625% Senior Notes due September 2030, carried at 292,087 net of 7,913 of unamortised issuance costs. Maturity schedule: nil in remainder-2026, 2027, 2028, 2029; 300,000 in 2030. Upsized by $50,000 during Q1-2026 (net proceeds 48,526 at 100.25%, underwriting discount 1,474).

The screen's figure is the carrying-value calculation to within $480k. Face value is the correct input to enterprise value: unamortised issuance cost is a deferred expense, not a reduction in the principal that must be repaid. Excluded from all three: operating-lease liabilities of 8,792 (FY2025) and a 7,000 accrued milestone.

Derived credit metrics: - EBITDA (TTM) ≈ EBIT 19,667 + amortisation 17,894 + PP&E/software D&A ≈1,936 = ≈39,497 - Cash interest run-rate = 8.625% × 300,000 = 25,875 → EBIT/interest = 0.76x - Net debt / EBITDA = 205,356 / 39,497 = 5.2x; gross debt / EBITDA = 7.6x - Tangible equity = 28,706 − 181,054 − 332 = (152,680)

---

4. Enterprise value and multiples

---

5. Own EV/Sales multiple history — construction

968 daily observations, 2022-03-16 → 2026-07-28. For each trading day: EV = close × shares outstanding at the most recent reported quarter + us-gaap:LongTermDebt as filed at that quarter − cash at that quarter; divided by the TTM revenue that was publicly available on that date (quarter end + 42 days for a 10-Q, + 75 days for a 10-K, so no look-ahead). Series in data/evs_history.json.

By calendar year (min / median / max): 2022 — 2.71 / 3.44 / 5.13 · 2023 — 4.12 / 5.96 / 10.10 · 2024 — 4.03 / 6.78 / 14.30 · 2025 — 5.04 / 6.87 / 8.75 · 2026 — 4.79 / 6.44 / 9.21.

Known limitation: the 2025-09-30 debt input uses LongTermDebt as filed at 347,239, which double-counts the redeemed notes in flight during the September-2025 refinancing. This overstates EV on roughly 60 trading days by up to ~6%, biasing the historical percentile against the current observation (i.e. it makes today look slightly cheaper than it is). Not corrected, but stated.

---

6. Reverse DCF

Run with ~/.claude/skills/investment-memo/assets/reverse_dcf.py. Inputs: --spot 39.32 --shares 37.2751 --net-cash -205.356 --revenue 268.675 --years 5 --solve cagr --terminal-margin 0.162 --exit-multiple {X} --wacc {W}. Full grid in HROW_Valuation.md §5.

Base case (12.0x, 16.2%, 10%, 5y): required revenue CAGR 38.8%.

WACC build-up (used as a sensitivity, not the base): equity $1,465m / debt $300m = 83%/17%. Cost of debt 8.625% pre-tax, taken untaxed because Harrow has large NOL carryforwards and pays negligible cash tax ($93k in FY2025). Cost of equity 4.3% risk-free + 1.75 beta × 5.0% ERP = 13.1%. WACC = 0.83 × 13.1% + 0.17 × 8.625% = 12.3%. The base run uses 10% for comparability with the screen; 12.3% raises the requirement to 41.7%, i.e. it makes the conclusion worse, not better.

---

7. Peer comparator construction

Peers pulled programmatically: EDGAR companyfacts for FY2025 revenue, prior-FY revenue, OperatingIncomeLoss, cash and LongTermDebt; shares from the dei:EntityCommonStockSharesOutstanding cover fact; price from Alpaca SIP close 2026-07-28. Output in data/peers.json.

Excluded as anchors, with reasons: CORT (extracted FY2025 revenue $761m and 5.9% EBIT margin are inconsistent with the company's known scale — an XBRL tag-selection artifact, not used); OCUL (effectively pre-revenue, EV/Sales 24.2x is not informative); AXSM and TARS (negative EBIT, so no EV/EBIT); BLCO, PCRX, SUPN, EOLS, AMPH (growth of −1.7% to +11.6%, outside the bracketing requirement).

Anchors used: ANIP, COLL, HRMY, ADMA — growth 19.6%–43.8%, bracketing the required CAGR; all US branded specialty pharma; all with positive EBIT. Median EV/EBIT 9.4x, median EV/Sales 2.05x.

---

8. Mention-frequency construction

Every quarterly earnings 8-K from 2023Q3 to 2026Q1 was fetched from EDGAR and all ex99* exhibits concatenated (press release + Letter to Stockholders + any supplement), then case-insensitive regex counts run per term. Word counts are reported alongside so density can be checked. Output in data/mentions.json.

Coverage: 11 consecutive quarters, complete — no gaps. The 2023Q2 accession returned no ex99 exhibits and is omitted rather than substituted.

Stated limitation: this measures written disclosure, not earnings-call Q&A. No transcripts were used, so no claim is made about management's behaviour under questioning.

---

9. Consensus

Alpha Vantage EARNINGS_ESTIMATES, retrieved 2026-07-29 (quota was available). Raw response in data/av_hrow.json.

Implied Q4-2026E = 348.47 − 44.20 − 70.35 − 101.71 = $132.21m, i.e. +48% above the all-time-record quarter of $89.09m. Recorded because it is the load-bearing assumption in the Street's number.

---

10. Figures explicitly NOT produced

• No Excel model. The memo's valuation output is a reverse DCF plus a multiple-history target; a three-statement build would add no information the implied-path test does not already carry, and the time-box directed effort to accounting quality first.
• No forward DCF. Terminal value exceeds 60% of EV by a wide margin, so the reverse DCF is mandatory as the primary output and a forward DCF would be a restatement of the same terminal assumption with a false-precision interim build.
• No F-score or gross-profitability level score. Under the INFLECTION archetype these are demoted to context by the Criteria; they were not computed.
• No chart pack. Time-boxed out in favour of the accounting-quality and valuation work, per the brief's stated priority order.
• No cross-sectional momentum percentile. No broad-universe snapshot exists in this worktree; the raw 12-1 figure (+20.7%) is reported and the percentile is INDETERMINATE.