Phase Space AI

Financial Model Notes

DexCom [DXCM]

DexCom [DXCM] — Financial Model Notes

Data provenance

TTM to 2026-03-31

Defects found

1. AV EARNINGS_ESTIMATES has no coverage for DXCM. It returns {"Information": …} on a premium key with requests paced below 5/second, while the same call succeeded for BSX. Retried 6 times with backoff. Consequence: the Consensus Criteria is BLANK for DXCM, which per criteria.md blocks nothing. The 12-month target's near-term base therefore uses company guidance ($5.16–5.20bn) rather than Street consensus — arguably a better input, but the substitution is recorded rather than silent.

2. Share-count discrepancy, 1.9%, reported not silently adopted. Three sources disagree: - AV BALANCE_SHEET.commonStockSharesOutstanding: 393.6m - AV OVERVIEW.SharesOutstanding: 385.9m - Implied diluted from the Q1 2026 call ($216.3m net income ÷ $0.56 EPS): 386.3m

The two independent sources that agree (Overview and the company's own EPS arithmetic) point to ~386m. 393.6m was used anyway, because it produces the higher EV ($28,545m vs $27,996m, +2.0%) and therefore the higher required CAGR — the conservative direction. Effect on the required CAGR: ~+0.35pp. Had the lower count been used, required CAGR would be ~14.9% and the margin +2.1pp rather than +1.77pp.

3. AV OVERVIEW is one quarter stale and OperatingMarginTTM is not a TTM figure. OperatingMarginTTM = 0.214 = exactly DXCM's Q1 2026 single-quarter GAAP margin (21.42%); the true TTM is 21.45% (close here, by coincidence). Confirmed systematic across all three names in this run — BSX's 0.206 = its Q1 2026 quarter, ATRC's 0.0629 = its Q2 2026 quarter. 3 of 3. Field not used.

4. AV has no Q2 2026 transcript for DXCM (transcript_2026Q2.json empty). The most recent transcript is Q1 2026. Unlike BSX, AV also has no Q2 2026 income statement for DXCM, so statements and transcripts are at least consistent with each other here.

5. The prepared/Q&A splitter defect described in BSX_Financial_Model_Notes.md §7 hit DXCM harder — 7 of 13 quarters returned n_prepared = 0 under the naive operator-phrase split, because DXCM's operator opens with "Later, we will conduct a question-and-answer session." Fixed by splitting at the first title containing "Analyst". The Stelo and rebate mention-frequency series in the research document are computed on the corrected split.

6. AV depreciationAndAmortization sign. av_vs_edgar.py raised no DNA_SIGN flag for DXCM (unlike BSX −$351m and ATRC −$5m). EBITDA was nevertheless computed as EBIT + |D&A| rather than from AV's ebitda field, per the brief's blanket instruction.

Consistency checks

• m_EBIT,T 27.0% ≤ m_gross,T 66.0%: passes, 39pp headroom.
• Terminal margin 27.0% vs trailing actual 21.45%: above, +5.55pp. Not the §9.8 defect.
• Opex bridge: 66.0 − 11.0 − 27.0 − 1.0 = 27.0. Exact.
• Terminal gross margin 66.0% vs trailing 61.84%: above trailing but below DXCM's own historical maximum of 68.63% (FY2021). Adopted-and-flagged per the brief's "unprecedented-and-verified" instruction.
• TTM built by summing four filed quarterly periods; DXCM files a discrete Q4, so the §9.2 "TTM that skips an untagged Q4" failure does not arise. Verified: FY2025 = 1,036.0 + 1,157.1 + 1,209.3 + 1,259.6 = $4,662.0m, which ties to the annual report.
• No Excel workbook built, so §1–§3 do not apply.