Phase Space AI

Financial Model Notes

KLA Corp [KLAC]

KLA Corp [KLAC] — Financial Model Notes & Data-Integrity Log

As of 2026-07-29 · framework v1.5.1

1. Source hierarchy actually used

2. Screen-input validation

KLAC is the only name of the three that the screen scored. Its record is wrong in six places, and the PASS verdict is an artifact. Source: reports/scan_v2/KLAC_analysis.json.

Root cause 1 — the split-basis mismatch (the dominant error)

KLA executed a 10-for-1 stock split effective 2026-06-12 (AV SPLITS, confirmed). The screen paired a pre-split share count (130.6m) with a post-split price ($190.865, which is exactly the 2026-07-28 close on the post-split basis). This is the CRWD failure the brief names: the net income ÷ shares ≈ filed EPS check does not catch it, because both sides of that check can be on the same wrong basis.

The screen also carried its own internal contradiction and did not act on it: it recorded entity_public_float = $83.7bn alongside market_cap = $24.9bn. A public float cannot exceed market capitalisation. That is an arithmetic impossibility — a REJECT-class signal on the brief's own test — sitting unexamined in the record. A float > market-cap assertion should be a hard invariant in the scanner. It would have caught this instantly.

Root cause 2 — the answer to the task's specific question

The task asked whether KLAC's negative required CAGR (−0.8%) is real or a terminal-margin artifact. It is neither real nor primarily a terminal-margin artifact — it is a share-count artifact. Understating market cap by 10x makes almost any price trivially achievable. The 14.4% terminal margin pushed the required CAGR the other way (a lower margin needs more revenue), so the two errors partially cancelled and produced a plausible-looking −0.8%. Corrected, the price discounts 36.2% growth, not decline. This is the "confident wrong number, not an error" pattern: nothing in the record looks anomalous.

Root cause 3 — the terminal margin, for the sixth time

terminal_margin_basis reads "max(own, sector peer median)" — one of the exact methods criteria.md names as already-failed, because it "can only flatter". Here it did the opposite: it returned 14.4% against a company earning 41.7%, i.e. 27.3pp below the company's own trailing actual, because the own input was the defective 6.4% operating margin and the peer median won the max(). This is the same class as AVGO −24.3pp and MSFT −23.7pp. Note also that 14.4% is the identical figure valuation.md cites in its warning about "a terminal margin of 14.4% against an 11.9% gross margin" — the constant appears to be reachable from more than one path.

LRCX and AMAT

Not scored by the screen. Derived here from first principles; no screen inputs to validate.

3. Alpha Vantage defects found on this name

Defect 1 — signed-negative D&A, and AV contradicting itself. For the quarter ending 2026-06-30:

• INCOME_STATEMENT.depreciationAndAmortization = −$196,766,000 (negative)
• CASH_FLOW.depreciationDepletionAndAmortization = +$98,606,000 (positive)

These are the same quantity for the same period from the same vendor, and they disagree by $295.4m as well as in sign. The consequence propagates: AV's own ebitda for that quarter is $1,425.9m against an ebit of $1,622.7m — EBITDA below EBIT with positive D&A, which the brief classifies as arithmetic impossibility and therefore REJECT-class, not merely surprising. Independently flagged by av_vs_edgar.py as DNA_SIGN.

Resolution adopted: the cash-flow figure, absolute value. It is stable at $96–99m across the six prior quarters and the −$196.766m is a lone outlier. TTM D&A = $394.0m; EBITDA computed as operating income + |D&A| = $6,054.6m. AV's ebitda field was not used anywhere.

Defect 2 — EARNINGS_ESTIMATES returns an empty array. {"symbol":"KLAC","estimates":[]}. No consensus EPS, no revenue estimate, no analyst count, and no 7/30/60/90-day revision history — for a $222bn-market-cap company with dozens of covering analysts. The endpoint returns HTTP 200 with a well-formed body, so nothing errors and nothing warns; a pipeline that reads estimates[0] gets an IndexError at best and silently skips the name at worst. LRCX and AMAT both return full data from the same endpoint on the same key in the same minute, so this is name-specific, not a quota or auth problem. Consequence for this memo: the 12-month target's NTM EPS is house-derived and flagged as such, and the Consensus Criteria is BLANK (which per criteria.md blocks nothing).

Defect 3 — OVERVIEW is stale, and its P/E is therefore wrong. OVERVIEW.EPS = $3.32 and PERatio = 57.47. But TTM net income $4,831.0m ÷ 1,306.275m = $3.70 — OVERVIEW excludes the June-2026 quarter that INCOME_STATEMENT (same vendor, same key) already carries. Real TTM P/E at $170.26 is 46.0x, not 57.47x — a 25% overstatement. MarketCapitalization is likewise one day stale ($249,237m implies $190.80/share, the prior close). AV is internally inconsistent across its own endpoints, which is a sharper version of the brief's warning that AV's consistency checks pass while being wrong — here they do not even pass.

Defect 4 — deferredRevenue is present-but-NULL. The field exists in every one of the 81 quarterly balance-sheet records and is None in all of them. The schema key is there, so a if 'deferredRevenue' in r guard passes and the value is then silently treated as zero or skipped. This is calibration item D1, "existence is not validity". Deferred revenue and RPO — the central quantities for this cluster's whole thesis — had to come from EDGAR companyfacts instead.

Defect 5 — longTermDebt is 0.0 in the most recent quarter only. At 2026-06-30 longTermDebt = 0.0 while shortLongTermDebtTotal = $5,887.4m; the five prior quarters carry ~$5,884–5,887m in longTermDebt. Reading longTermDebt would have produced net cash of +$4,902m instead of net debt of −$985m, a $5,887m error in the enterprise value. shortLongTermDebtTotal was used.

Defect 6 — sellingGeneralAndAdministrative is 0.0 for FY2026. Zero for the newest fiscal year while FY2025 reports 8.47% of revenue. The amount reappears in the unexplained residual, so the operating margin survives, but the opex bridge — which valuation.md calls "the real test" for a terminal margin — is unbuildable from AV's expense lines on the current year. The FY2025 bridge was used instead.

4. Derived figures, shown so they can be checked

AV's ebitda field was NOT used, per the brief. It is wrong on this cluster: KLAC Q4 FY2026 shows ebitda $1,425.9m BELOW ebit $1,622.7m — arithmetically impossible with positive D&A, and a REJECT-class figure on the brief’s own test.

AV's ebit field was also NOT used — it is not operating income. On this name the ebit field also diverges from operatingIncome (it includes non-operating items); operatingIncome is used throughout.

5. Reproduction

Raw pulls are in work/wfe/ at the repository root: av_KLAC_INCOME_STATEMENT.json, av_KLAC_BALANCE_SHEET.json, av_KLAC_CASH_FLOW.json, av_KLAC_OVERVIEW.json, av_KLAC_SPLITS.json, est_KLAC.json, ef_KLAC.json (EDGAR companyfacts), KLAC_R*.htm (10-K R-files), snap_KLAC.json.