Home Synthesis Fit

Fit

The answer

Does not fit the framework (P1 not met); contested: P2, P5

Salesforce clears the universe and trips no exclusion, but the year-10 durability gate (P1) does not clear with the very-high conviction it demands, and that gate is decisive on its own — nothing offsets it. Confidence is low: a load-bearing criterion was contested (P5, the temporary-versus-permanent diagnosis) and its probability spread of 0.26 exceeded the 0.25 threshold, while the consistency criterion (P2) split across model families. No watchlist-only flag applies; long-dated options exist. The framework's own rule governs: any proper doubt at the gate resolves to does not fit.

Universe and exclusions

The screen is clean, and it is worth saying plainly what "clean" means here: nothing in the universe or exclusion set blocks the name, so the case turns entirely on the pillars below.

Both universe lines are met. Salesforce is a US-domiciled, Delaware-incorporated common stock listed on the NYSE under ticker CRM — not an ADR, not a Chinese issuer [1]. Market capitalization is roughly $163B (956M shares at $170.77, 2026-07-17), about sixteen times the $10B floor, and even the June-2026 trough of $150.12 (~$143B) stayed far above it.

Every hard exclusion was checked and none hit:

  • Auto-OEM (X1): not triggered — this is cloud CRM software at a 77.7% gross margin, not vehicle manufacturing [2].
  • Promotional CEO (X2): not triggered — Benioff's register is promotional, but the checkable promises delivered (the 30% non-GAAP operating-margin target was hit at 30.5% in FY2024 and 33% in FY2025; the FY2026 revenue guide of $41.0–41.3B closed at $41.5B), and he holds ~22.8M shares (~2.8%, ~$3.9B). The counter-fact in the same breath: one large forward promise — a return to double-digit revenue growth in the second half of FY2027 — remains unproven.
  • Structural decline (X3): not triggered — revenue has not declined in any of the last ten years, let alone high-single-digit for three consecutive years; the counter-fact is that growth has decelerated from 28.7% (FY2020) to 9.6% (FY2026).
  • Market darling / extreme multiple-to-sales (X4): not triggered — the stock trades at ~3.9x sales and ~21.9x trailing GAAP earnings after a 59.2% drawdown; the counter-fact is that the sell side still rates it a consensus buy with a mean target ~40% above spot [3].
  • China dependence (S1): immaterial — total Asia Pacific revenue is 10.4% of the total, China is not separately disclosed, and renminbi is not among the currencies Salesforce flags for FX exposure (per the FY2026 segment and market-risk disclosures; the exact China sub-portion is not separately reported).

Pattern match

Of the framework's four recognition setups — cyclicals at the bottom, high dividend plus high FCF yield, healthcare/insurance forecasting errors, and quality tech monopolies/duopolies on a fear dip — Salesforce most resembles the fourth in shape but fails its structural check.

The fear-dip shape is real: a 59.2% peak-to-trough fall on a specific, testable narrative (AI agents cannibalizing the seat-based model), a capitulation volume spike of 3.5x the pre-peak median, and a company buying back its own stock at prices management called "low." What the pattern also requires — a monopoly or duopoly market structure — is absent. The company's own 10-K calls its market "highly competitive, rapidly evolving and fragmented, and subject to changing technology with low barriers to entry" [4]. The lead rests on switching costs and scale, which the framework values below the regulator-gated, capital-heavy structures it prizes. The setup carries the fear the framework hunts but not the structure it requires.

The pillar ledger

Reference lines throughout — where the company sits against the framework's own bars, with the deciding arithmetic. Full treatment sits on the linked evidence tab.

Year-10 durability gate — P1 (not met; the decisive result)

Here is the decisive point. The gate asks whether year-10 revenue and free cash flow will be higher than today with very high conviction, and it is binary by construction: any proper doubt resolves to does not fit. All four jury seats returned not met (trimmed-mean probability 0.635, spread 0.08, agreement across both model families). The base rates favor a higher year-10: revenue rose in every one of the last ten fiscal years ($8,437M FY2017 to $41,525M FY2026) and reported free cash flow rose in every one ($1,698M to $14,402M) [5], the structural-decline disqualifier is not tripped, and Salesforce is the #1 CRM provider globally [6]. The counter-fact sits in the same treatment and is what fails the gate: the conviction sources the framework rewards do not apply. There is no monopoly/duopoly structure, no regulatory entry barrier, and capital intensity is negligible (capex $594M / revenue $41,525M = 1.43%), so the moat is switching costs — execution-adjacent, which the framework explicitly says is not a year-10 moat. And the 10-K itself concedes a live, unresolved threat: "New AI offerings may disrupt our service offerings or transform workforce needs and negatively impact demand for our offerings" [7]. Rising history plus an unresolved doubt is exactly the case the binary gate resolves against. Full treatment: Durability.

FCF consistency — P2 (contested)

Reported free cash flow rose every year for a decade with zero negative episodes, which meets the consistency test on a reported basis — two claude seats voted met. The two codex seats returned cannot-determine, because the framework's actual metric is adjusted-FCF stability, and that series is not computable: stock-based compensation was absent from the structured cash-flow feed for all of FY2017–FY2026, so the rolling five-year adjusted series could not be built [8]. The split is 2 met / 2 cannot-determine, across families — the aggregate is contested. Full treatment: Durability, Yield.

Dislocation and yield — P3a, P3b, P3c, P3d

Entry trigger (P3a, not met). The 59.2% drawdown is genuine, but it was not driven by the framework's canonical short-term earnings cut. Every recent quarter beat, forward consensus rose, and the fall is a multiple de-rating on decelerating growth and AI-disruption fear. Two seats scored not met; the aggregate is not met. Full treatment: Dislocation.

Fear gauge (P3b, met). Volume spiked to 3.5x the pre-peak median in the peak-to-trough leg and clustered at the bottom (a 55.4M-share session on 2026-06-18, four days before the low) — emotion-driven capitulation, confirmed by all four seats. The counter-fact: the spike concentrated in the final leg, so it validates the trough, not the whole slide. Full treatment: Dislocation.

Yield versus the bar (P3c, not met). On the framework's adjusted basis — free cash flow less stock-based compensation less the five-year average of acquisition spend — the yield computes to 3.3% (FY2026 adjusted FCF $5,413M on a $163,256M cap), roughly 470 bps short of the 8–9% fortress bar. Even the charitable stock-comp-only version is 6.7%, still below 8%. The counter-fact: reported FCF yield is 8.8% and would appear to clear the bar; the gap is entirely the SBC-and-M&A adjustment for a serial acquirer [9]. Full treatment: Yield.

Forward path (P3d, not met). Reversion above the 8% adjusted bar within one to three years is roughly 20–25% likely (trimmed-mean probability 0.22, spread 0.01, both families agree). Consensus reported FCF yield reaches ~10.9% by FY2029, but net of ~$4B SBC and the ~$5.5B five-year M&A charge the adjusted forward yield is only ~5.1% (8.4% if acquisitions cease) — clearing the bar depends on the serial-acquisition run-rate collapsing. Full treatment: Yield.

Balance sheet and self-help — P4a, P4b, P4c

Can it outlast the problem (P4a, met). Net debt is a fraction of one year's free cash flow — roughly $4.8B (including marketable securities) against $14.4B of FY2026 FCF, or 0.34x — with a single trivial near-term maturity (the $4.0B Informatica 364-day facility) and a $5.0B undrawn revolver [10]. The counter-fact in the same breath: in FY2026 the company drew $6.0B of new debt to run an $8.1B acquisition and a $12.6B buyback at once, so the constraint on repurchases is acquisition appetite, not solvency [11]. Full treatment: Self-Help.

The repurchase engine (P4b, not met). The buyback is executed and large but not yet a powerful share-retirement engine: SBC of ~$3.5B a year reissues roughly two-thirds of what is bought, so across FY2024–FY2026 gross repurchases of 116M shares netted to only a 41M-share reduction (997M to 956M), about 1.4% a year, and fit_features still classes the multi-year trend as rising (700M in FY2017 to 956M in FY2026). At a 10% adjusted yield the framework's flywheel would retire ~10% a year; the delivered rate is ~1.4% [12]. The counter-fact: the count has now fallen three years running and a $50B authorization plus a $25B accelerated repurchase at the depressed price show intent to accelerate. The absurdity check does not trigger — at today's price it takes ~11 years of reported FCF (or ~15 net of SBC) to retire the float, not the ~3 years that flags an impossible price. Full treatment: Self-Help.

Dividend (P4c, not applicable). The dividend is immaterial to the case — a ~1.0% yield covered ~9x by free cash flow, initiated only in FY2025, with no downturn record to test.

Diagnosis — P5 (contested)

The temporary-versus-permanent question is the trial's, not this tab's, and the blind panel returned it contested. The probability the impairment is temporary is 0.62, with seats at 0.62, 0.42 and 0.68 (spread 0.26) and order-stability holding (temporary-first mean 0.62 versus permanent-first 0.55, gap 0.07). The temporary reading: the price destroyed ~$196B (-54.6%) of market cap while the near-term cash hit was near zero — FCF grew ~16% and forward consensus held flat-to-up — so a reverse-DCF at a 9% discount rate implies the current price embeds only ~0% perpetual FCF growth, versus ~5.5% at the 2024 peak. The permanent reading, carried in the same breath: that ~0% embedded growth is arithmetically supportable for a business whose growth has structurally reset to high single digits and whose recent re-acceleration was "bolstered by the acquisition of Informatica" plus ~2 points of FX [13], with Agentforce Apps up only 9% [14]. The gap is nearly the whole drawdown under one reading and closes under the other; the panel did not resolve it. Full treatment: Damage Math.

Instrument context — I1 (not verifiable)

Long-dated listed options exist (January 2027 and January 2028 LEAPS, the latter listed 2025-09-15, ~18 months out as of July 2026) with deep single-name liquidity, and 30-day implied volatility reads ~45–50% — the upper edge of the framework's acceptable band, not the elevated 60–70 zone. Verdict: not verifiable, because that evidence is web-sourced and cannot be pinned to a corpus page (local data confirms only 37.9% 30-day realized volatility). The framework's watchlist-only case — no qualifying long-dated options — does not apply; options exist. Stated as a framework fact, never as advice. Full treatment: Clock.

What a three-fold return would require

The framework's target test prices the equity at its adjusted-FCF bar on normalized adjusted free cash flow, then measures the distance consensus would have to travel. That arithmetic is unavailable in this run: the tally records the re-rating math as null because the applicable bar or normalized adjusted FCF is missing — stock-based compensation was absent from the structured feed, so the deterministic adjusted-FCF series (and the yield baseline it anchors) could not be computed. The re-rating price target therefore cannot be stated without improvising the figure, which the framework forbids.

What can be stated is the base-rate context from this name's own history, per the Clock tab. Salesforce's one directly comparable episode — a company-specific 55%+ multiple de-rate (2021–22, -58.6% over 403 days) — took roughly 28 months to round-trip. The deeper 2008 drawdown (-70.5%) round-tripped in ~18 months but was a market-wide crash. The name has never round-tripped a 55%+ company-specific de-rate inside 18 months, so a full recovery on the framework's 18-month instrument horizon runs against its own base rate; partial re-recognition is the more supportable expectation.

Contested and undetermined

Two criteria are contested; none were left cannot-determine at the aggregate level.

  • P2 (FCF consistency) — contested. Reported FCF rose 10/10 years with no negative episode (met), but the framework's adjusted-FCF stability series is not computable because SBC was absent from the feed (cannot-determine). Vote split: 2 met (claude) / 2 cannot-determine (codex); cross-family agreement absent.
  • P5 (diagnosis) — contested. The panel split 0.62 / 0.42 / 0.68 on the probability the impairment is temporary; the spread of 0.26 drives the low confidence tier and leaves the temporary-versus-permanent question formally unresolved.

Provenance

Item Result
Jury composition 4 seats — 2 Model Family A (masked seat drawn from this family), 2 Model Family B
Families / agreement Two families; P1 gate agreed across both (not met, spread 0.08); P2 and P5 split across families
Trial order-stability temporary-first mean 0.62 vs permanent-first 0.55; gap 0.07 (stable)
Name-mask probe Every gate result reproduced under the name mask; max probability gap 0.0; no prior-driven-risk flag
Skeptic counts 16 claims fully checked — 10 survived, 6 weakened, 0 refuted; plus 2 unverifiable and 21 triaged-only

The verdict was pressed hard. The decisive gate (P1) held its not-met result across two independent model families and again when the company's name was masked, so it is not an artifact of prior beliefs about Salesforce. The two contested criteria are contested honestly — the panel and the families disagreed, and that disagreement is reported rather than rounded into a clean answer.

Falsifier ledger

These are the standing conditions that would change the read — the framework's own what-would-change-this seeds and the name-specific thresholds, with direction and window where defined. They are reproduced verbatim from the tally:

Data gaps

What this run could not resolve, from the tally's list:

  • Adjusted FCF and adjusted-FCF yield are not computable in fit_features: stock-based compensation was absent from the structured cash-flow feed (FY2017–FY2026) and the acquisition term defaulted to zero, so the deterministic adjusted series, the yield baseline, and float-retirement years could not be derived. The Yield tab reconstructs adjusted FCF directly from the filed 10-K cash-flow statements (SBC and business-combinations lines); this tab carries those reconstructed figures, and the re-rating math above is left unavailable rather than improvised.
  • Balance-sheet class returned "unknown" (FY2026 EBITDA missing from the feature inputs); leverage is characterized from filed debt, cash and FCF instead, and the applicable yield bar is carried as an 8–10% range (net debt/EBITDA 0.41x including securities, 0.59x cash-only).
  • IDC precise CRM share percentages and the full ranked competitor table are from web research, not the filing corpus — the corpus verifies the #1-provider statement but not the exact 20% share or the ~4x gap to the #2.
  • Implied volatility and LEAPS existence rest on web sources (AlphaQuery, options-listing calendar), not a corpus PDF; IV is reported as found, never estimated.
  • No corpus transcript for Q1 FY2027 (reported 2026-05-27); the freshest capital-return and Agentforce commentary is one quarter stale, and the $25B accelerated repurchase is cited to the Calls curation rather than a filing page.
  • Reported short interest and days-to-cover are unavailable in the position feeds for this run, so the short-selling level and any forced/structural sellers cannot be quantified.
  • China-specific revenue is not separately disclosed, so the S1 sensitivity is quantified only at the Asia Pacific level (10.4%).