Phase Space AI

Valuation

Catalyst Pharmaceuticals [CPRX]

Catalyst Pharmaceuticals [CPRX] — Valuation (RE-UNDERWRITTEN 2026-07-29, v2)

Supersedes archive/CPRX_Valuation_2026-07-29_superseded.md. Framework: MEMO_BRIEF v2. No position verdict is issued.

---

0. The governing fact, re-verified from primary sources

Catalyst Pharmaceuticals no longer exists as a listed security. The prior memo asserted this; it was challenged and it holds. Four independent confirmations:

The price closed at $31.44–$31.50 for the fourteen sessions before the last — the flat, no-variance tape of a settled cash merger. The 12-month "target" is a realised, settled number: $31.50 per share in cash.

This has a hard methodological consequence. The brief's method requirement 1 — compute the terminal margin from Alpha Vantage's normalized statements "so it is reproducible" — is impossible on this name. AV purges delisted tickers: INCOME_STATEMENT, BALANCE_SHEET, CASH_FLOW, OVERVIEW, EARNINGS and EARNINGS_ESTIMATES all return an empty object with HTTP 200 and no error. Every figure below is therefore rebuilt from SEC EDGAR XBRL and the filed 10-K/10-Q text. This is itself a finding: the new data layer has a survivorship hole, and an "empty response" from AV is not evidence about a company.

Everything below is run twice — as the framework requires, and as a no-deal counterfactual. The counterfactual is where the methodological content lives, and it is where the answer changes.

---

1. Company state

Declared: STATE A — mature and structurally stable. Evidence grade B, not A.

B is inapplicable (no exogenous cycle). C is inapplicable (85–90% gross margin, 44.8% operating margin — not thin). D is inapplicable ($597m TTM revenue, $214m of FY2025 net income). A is the only admissible state; the regime-change failure is carried as an evidence-grade reduction to B, which under valuation.md rule 5 constrains position size rather than the operating assumption.

---

2. Inputs, rebuilt from EDGAR — and the share-count error

2.1 DEFECT — the prior memo used BASIC shares where the instrument requires DILUTED

reverse_dcf.py --shares is documented as "diluted shares, millions." The prior memo supplied 122.381m basic. The filed diluted weighted average for Q1-2026 is 126,903,391 — 4.52m shares, 3.7%, higher.

The EPS cross-check proves which is right and shows exactly why the check the brief warns about is weak:

Magnitude: EV understated by $142.5m (+4.6%). Corrected mcap $3,996.2m − net cash $755.9m = $3,240.4m. This pushes the required CAGR up by ~1.4pp and therefore cuts against the conclusion below, which is why it is stated first.

2.2 Split-basis check — negative, verified rather than assumed

AV SPLITS returns an empty array, but AV has purged the ticker, so that is uninformative. Verified from EDGAR instead, over three independent periods:

Share count moves monotonically and smoothly (120.9m → 123.3m → 122.2m, the decline being the $200m buyback authorised 2025-10-01, 1,740,713 shares repurchased through 2026-05-07). No split discontinuity. No split-basis mismatch in the window used. All revenue figures are dollar amounts and split-invariant.

2.3 Net cash — the prior memo was RIGHT

$755.859m of cash and equivalents at 2026-03-31, of which $574.153m is US Treasuries classified as cash equivalents (10-Q Note 3, amortized cost $574.142m, unrealised gain $11k). No funded debt: LongTermDebt is absent from every period; total liabilities of $134.913m are payables, accrued revenue allowances and a $2.681m operating lease. Contra the brief's expectation that net_cash was wrong on ~20 of 22 names — on CPRX it was right, to the thousand.

---

3. Terminal EBIT margin — 48.0%, and where the −20.9pp came from

3.1 The origin of the error

The screen recorded terminal_margin_capped_from: 0.438 and set terminal_margin: 0.229, basis "max(own, industry median) — CAPPED at industry p75." The gap is exactly 20.9pp.

The coordinator has since traced the mechanism: coverage_scan.py keyed its "sector peer median" on a sic field that does not exist in SEC companyfacts, so every company fell into one bucket and the "sector median" became a universe-wide median of 14.35%; 70 of 111 records carry 14.4%. The 22.9% was an industry-p75 statistic computed over that artifact. It was not an estimate of anything about Catalyst, and no attempt is made below to reconcile against it.

3.2 CPRX also carried data_quality_ok: false, OVERRIDDEN downstream

CPRX was one of four names where a data_quality_ok: false flag was overridden by a downstream consumer recomputing the flag rather than reading it. So the record was scored after failing its own data-quality test. Every input above has consequently been rebuilt from primary sources, and one load-bearing input (the share count, §2.1) was in fact wrong — vindicating the flag that was ignored.

3.3 The bridge, reconciled on actuals then projected

m_EBIT,T = m_gross,T − R&D − SG&A − other. Q1-2026 actuals, from EDGAR:

The bridge closes on the actuals: 90.3 − 1.78 − 33.0 − 6.5 = 49.0% = reported Q1-2026 operating margin ($73,232 / $149,390), to the decimal. Hard constraint: 48.0% ≤ 88.0% ✓.

Why 48.0% and not 43.8% (the prior memo's figure). The prior memo used the FY2025 actual and was directionally right against the screen, but FY2025 predates the royalty step-down and so understates run-rate economics by roughly 5pp of gross margin. Q1-2026 is the first quarter that contains it, and it printed 49.0%. Setting the terminal margin at 48.0% takes the demonstrated post-royalty structure, credits almost none of the amortisation roll-off and almost none of the SG&A leverage, and allows for AGAMREE royalty dilution. It is 3.2pp above the TTM 44.8% and 1.0pp below the most recent quarter. A terminal margin above the TTM here is not aggression — it is refusing to average across a step change that has already happened, which is the mirror image of the error that produced the 22.9%.

---

4. The revenue-quality work — organic vs acquired, run as the AIOT decomposition

This is the highest-value section and it is where both the screen and the prior memo were incomplete.

4.1 Per-product revenue, ten quarters, from primary filings

Net product revenue, $000. Q4 columns are derived as (fiscal year − sum of Q1–Q3) and both derivations tie to the filed totals exactly.

Annual: FY2022 $213,938 (100% FIRDAPSE) · FY2023 $396,502 · FY2024 $489,327 · FY2025 $588,807.

4.2 The AIOT test

Precedent: AIOT screened at +44.7pp when 63.5% of revenue was acquired and legacy revenue was down 1.0%. Same decomposition, FY2022 → FY2025, on the $374,869k of revenue added:

61.5% of the revenue added was acquired, confirming the prior memo. But the AIOT test turns on the legacy line, and this is where CPRX and AIOT diverge completely:

CPRX passes the test AIOT failed. The screen's 40.1% total CAGR is not a valid organic hurdle — but neither is the conclusion that the growth is acquisition-manufactured. FIRDAPSE compounded at 18.8% on its own for three years and accelerated to +18.1% in the latest quarter. That is a franchise, not a roll-up.

4.3 What both the screen and the prior memo missed — the ex-FYCOMPA rate

The prior memo reported total YoY of +5.6% and guided +7.0% and stopped. Neither figure describes the forward business, because both are depressed by a wasting asset that is nearly gone.

Q1-2026 vs Q1-2025 decomposition of the +$7,943k (+5.6%):

The core portfolio grew 28.2%. Reported growth was 5.6% because a generic-eroded product gave back $21.9m. FYCOMPA is now 9.2% of revenue and falling ~60% a year. Its remaining drag on total growth is roughly 4pp in FY2026, ~1.5pp in FY2027, and nil thereafter.

Consequence for the implied-path test. The framework compares required against demonstrated, and "demonstrated" must describe the assets that will exist over the horizon. The correct demonstrated reference is the core portfolio's rate with an explicit, dated, decaying FYCOMPA drag — not the blended 5.6%, and not the screen's 40.1%. Reporting +5.6% as the demonstrated growth of a company whose surviving products grew 28.2% is the same class of error as the terminal-margin cap: substituting an artifact of composition for a fact about the business.

4.4 Receivables and DSO — the tell the brief asks for

Not computed in the time available. Recorded as a gap. Directionally, the risk is low: net cash rose from $517.6m to $755.9m over five quarters and Q1-2026 operating cash flow was $59,587k against $63,734k of net income (94% conversion) — a receivable inflating ahead of revenue is hard to reconcile with that. But the DSO series itself was not run and is not asserted.

---

5. Product exclusivity — every date, with its source, or an explicit "not disclosed"

All from the FY2025 Form 10-K, filed 2026-02-25 (accession 0001193125-26-071525) unless stated. Nothing here is inferred.

FIRDAPSE (amifampridine — LEMS) — 55.5% of revenue, growing 18.1%

This is the single most important set of facts in the thesis and the prior memo did not state the Teva date at all. Three of four ANDA filers — including Teva, the largest generic manufacturer in the world — are contractually barred until 2034/2035, and two of the three conceded both validity and infringement. FIRDAPSE's generic exposure is not a 2026 event; it is one remaining litigant against patents running to 2037.

AGAMREE (vamorolone — DMD) — 24.6% of revenue, growing 66.6%

FYCOMPA (perampanel — epilepsy) — 9.2% of revenue, falling 61.3%

Acquired from Eisai, closed 2023-01-24. It was bought with roughly one patent-year of tablet exclusivity remaining, and three ANDA filers' Paragraph IV certifications on the tablet formulation had already gone uncontested by Eisai before the acquisition (10-K). The prior memo's "'497 patent expired 2026-07-01" is correct; the '571 expiry of 2025-05-23 — the one that actually triggered the erosion — was not stated.

Nothing omitted above is fabricated. The one genuinely missing input is the AGAMREE PTE length, which does not yet exist as a determination.

---

6. The implied-path test

reverse_dcf.py. Solved for: revenue CAGR. Held fixed: terminal EBIT margin 48.0%, exit multiple 11.3x, WACC 10.0%, 5 years, net cash $755.859m, diluted shares 126.903m, TTM revenue $596.958m.

6.1 The exit multiple, derived then cross-checked

EV_T/EBIT_T = (1−t)(1−g/ROIC)/(WACC−g), with t ≈ 22%, ROIC ≈ 46% (NOPAT $208.8m on invested capital of roughly $450m of intangibles plus working capital), WACC 10%:

Three references converge, which is unusual and worth recording:

Base set at 11.3x — the own-history median, which sits inside the identity's range. The screen's 23.4x was 87% higher than the price an informed strategic acquirer actually paid in cash with full diligence and control.

6.2 Results

Runs A and B reproduce both prior outputs exactly, which localises every disagreement in parameter choices, not arithmetic.

Run G isolates the terminal-margin error's true cost: at a defensible exit multiple, the screen's 22.9% terminal margin would have demanded a 27.6% revenue CAGR. The screen reported 9.3% only because a second error — a 23.4x exit multiple above the name's own all-time p90 — ran in the opposite direction and was larger. The reported 9.3% was roughly the product of two errors cancelling. Neither was flagged.

6.3 The margin — demonstrated − required

Verdict on the merits, pre-deal: PASS.

Three of the five references clear 10.0%, including both of the two that describe the assets which will still exist in five years. The two that fail are the two contaminated by FYCOMPA. A price requiring 10.0% against a legacy franchise compounding at 18.8% for three years and a core portfolio growing 28.2% is not a demanding price.

This reverses BOTH prior conclusions, in different ways. The screen's PASS at +30.8pp was right by accident — two large errors of opposite sign, and a denominator (40.1%) that was invalid. The prior memo's INDETERMINATE-with-a-−3.6pp-note was too harsh: it correctly rejected the 40.1% but then adopted a blended guidance figure depressed by a product in terminal decline, and never computed the ex-FYCOMPA rate.

Verdict as recorded: INDETERMINATE. Not on the merits — the price input is an extinguished security. No implied-path statement about a delisted stock is meaningful, and per the framework a missing input is INDETERMINATE and never FAIL. The pre-deal merits are recorded separately as PASS, because that is the finding the framework needs for calibration.

Implied compression: trading at $31.49 the EV/TTM EBIT on diluted shares is 12.11x (68th percentile of own history) against a solved exit of 11.3x — implied compression of 0.81x, or 6.7%. Against the screen's 23.4x exit the implied expansion was +93%, which no part of the screen's output flagged.

Terminal value is ~100% of modelled EV by construction, so the reverse DCF is the mandatory primary long-horizon output.

6.4 Sensitivity — on the exit multiple

The verdict is invariant across the entire defensible exit-multiple range (10.4x–12.49x) on the core and FIRDAPSE references, and invariant in the other direction on the guidance reference. The judgement is therefore not in the multiple — it is entirely in which growth reference is the honest one, which is exactly where §4.3 puts it. Stating that is more useful than a flip point.

---

7. The 12-month target

7.1 Realised, not forecast

There is no distribution around this number. Any expected return computed off the $31.49 screen price is +0.03%, not the +30.8pp margin the screen reported.

7.2 No-deal counterfactual, rebuilt on corrected inputs

Built as valuation.md requires: near-term base → named dated events → the name's own multiple history with the percentile stated. No peer median is used anywhere.

Base. Consensus is unobtainable — EARNINGS_ESTIMATES returns empty for a delisted ticker, recorded as INDETERMINATE rather than as absence of coverage. Company guidance is used, which for a 12-month horizon is the better instrument: FY2026 total revenue $615–645m, issued 2026-02-25, with "promoted product revenue expected to grow by 20% or more."

Named dated events inside 12 months, each in the Catalyst Calendar: - FIRDAPSE upstream royalty expiry, January 2026 — blended 18.5% → 6.0%, already in the Q1-26 print (gross margin 87.3% → 90.3%) - Hetero Paragraph IV trial, from 2026-03-23, D.N.J. — the last unsettled FIRDAPSE ANDA - 30-month stay expiry, 2026-05-26 - FYCOMPA '497 patent expiry, 2026-07-01 — continuing erosion - AGAMREE Phase 1 immunosuppression-dose readout, 2026-06-30 (delivered) - AGAMREE Canadian launch by sub-licensee KYE, post Health Canada approval - AGAMREE PTE determination by USPTO — date not determinable; not asserted

Margin held at the Q1-2026 actual 49.0%, not the TTM 44.8%, since the royalty step-down is permanent. This is the largest single judgement in the counterfactual and is flagged as such.

(Per share = EBIT × multiple + net cash $755.859m, ÷ 126.903m diluted shares.)

No-deal 12-month target: $33.35 — own-history median 11.26x (50th percentile, stated) on guided FY2026 revenue at the midpoint. +28.6% to the unaffected close of $25.94. Range across the guidance band and the p25–p75 multiple band: $26.41 – $43.88.

Prior memo: $32.15. The +$1.20 difference is the corrected margin (49.0% vs 44.8%) partly offset by the corrected diluted share count.

7.3 The sanity band — an acquirer's diligenced valuation

The framework's own-history method lands within 5.5% of the price a strategic buyer paid in cash after full diligence, with the multiple taken mechanically from the name's own median and the revenue from published guidance. Nothing was tuned to reach that. It is the strongest single validation of the 12-month instrument in the record, and the prior memo's version of this check (−2.0%) was closer only because two of its inputs were wrong in offsetting directions.

It also reframes the transaction. Angelini's "21% premium" was a premium to a depressed price — CPRX traded at the 41st percentile of its own EV/EBIT history on 2026-04-22. Against its own median multiple on guided earnings and the corrected margin, $31.50 was a ~5.5% discount to fair value. Shareholders received a control premium relative to where the stock had drifted and a small discount relative to what the business was worth on its own historical terms. And the buyer paid 12.49x — above the identity-warranted 10.4–11.9x — which is what an informed strategic pays when it can see the FIRDAPSE settlement wall running to 2035.

7.4 Item B16

Targets here are above both the unaffected price (+28.6%) and, marginally, the deal price. Recorded for the calibration tally.

---

8. Mention-frequency — prepared remarks vs Q&A

EARNINGS_CALL_TRANSCRIPT. Only the 2025Q4 transcript survives; 2026Q1 returns an empty array — the same delisting purge as the fundamentals. A time series is therefore not constructible, and that is reported rather than substituted for. Recorded as a data gap caused by delisting, not by absence.

2025Q4 call, 45 utterances, 7 prepared / 38 Q&A:

The FYCOMPA row is the finding. Under the brief's own evidentiary rule — a claim management volunteers is weaker evidence than one it concedes under pressure — the inverse also holds: a topic management raises fifteen times that no analyst will engage with has already been written off by the Street. That independently corroborates §4.3: the market was valuing the core portfolio and treating FYCOMPA as a wasting asset, while the screen and the prior memo were both computing growth rates that included it.