Tempus AI, Inc. [TEM] — Trade Construction & Risk Management (Task 5)

Valuation date 2026-07-27 · Spot $42.91 · Framework investment-memo v1.4.2, applied unmodified Prerequisites Task 1 (TEM_Research.md) and Task 3 (TEM_Valuation.md) complete; model TEM_Model.xlsx verified and tied to filed statements.

0. Portfolio context (read the book first)

portfolio_book.json read at 2026-07-27. 0 open positions; 13 watchlist names (MU, ISRG, NET, SMR, NBIS, SNDK, GH, MSFT, GOOGL, NTRA, CIEN, TXG, TWST). Config: max single-name weight 5%, max sector concentration 25%, max pairwise correlation without disclosure 0.60, cash hurdle 4.7% annual.

Marginal correlation (trailing 1-year daily log returns, Alpaca, n=252):

No correlation limit is breached. With zero open positions, the marginal-correlation-vs-holdings test is vacuous; the figures above are recorded for the watchlist cluster. Note that a healthcare/diagnostics cluster is forming on the watchlist — GH, NTRA, TXG, TWST and now TEM — and if more than one converts, the 25% sector-concentration cap becomes live. Recorded now so it is not discovered later.

Capital-competition test. The book is empty, so no existing position would be displaced. The binding comparison is therefore against cash at 4.7%. TEM's scenario-weighted expected return net of costs is +2.1% (§Gate 4). Cash wins on the central weighting. An empty book does not waive this step and it has been performed.

1. The four conclusions (carried forward from Task 3 §7, restated for the decision)

1. Fundamental — genuine, evidenced Diagnostics gross-margin expansion (45.3% → 61.3% in five quarters) and Adjusted EBITDA inflection, running alongside a widening GAAP operating loss, SBC compounding at 129% vs 36% revenue growth, $589m of net debt, ~$73m/quarter of cash burn, and growth that was only 33.5% organic in FY2025.
2. Expectations — consensus FY2026 revenue is guidance ($1,592.9m vs $1.59–1.60bn, 18-analyst range spanning 1.2%). House +1.1%. EPS estimates revised up on both FY2026 and FY2027. Street: Buy, $66.57 average.
3. Valuation — methods span $8 to $145. DCF $12.80 at 13.6% WACC with 72% terminal value; the market price requires ~11% WACC and a 20–24% terminal EBIT margin. Probability-weighted target $43.93 (+2.4%).
4. Portfolio — below.

2. Hard gates

Gate 1 — Causal mechanism · PASS (long side), qualified

The contemplated direction is long, since that is where the peer discount, the Street target, the positive revision trend and the 58% drawdown all point.

Named mechanism: operating leverage in the Diagnostics line, evidenced in disclosed data rather than asserted. Diagnostics gross margin 45.3% (9M 2024) → 58.9% (9M 2025) → 61.3% (Q1 2026) — ~1,600bp in five quarters. Consolidated gross margin 54.9% → 62.7% → 63.8%. Adjusted EBITDA −$104.7m (FY24) → −$7.4m (FY25) → +$65m guided (FY26). This is a metric moving in filed statements, not a risk listed in a bullet.

Required quantitative corroboration (Factor & Anomaly Scorecard) — and it SPLITS. The long-side standard is strong profitability + low accruals + high F-score:

The disagreement is stated explicitly rather than ignored, as the gate requires. Two of the four corroborating measures fail, one of them (asset growth, Cooper/Gulen/Schill 2008) at an extreme reading. The mechanism is real and evidenced; the quantitative corroboration is partial. Gate 1 passes on the strength of the disclosed margin series, at reduced weight.

Counter-mechanism, recorded because it is equally evidenced. A capital-allocation mechanism points the other way and would support a short: TEM is issuing equity at ~5.2x forward revenue to buy Personalis at ~16.7x forward revenue, at a 13.7% gross margin versus its own 63.8% — approximately 19% dilution for ~5.6% of revenue. SBC growing 129% against 36% revenue growth compounds the same per-share dilution. Both mechanisms are live simultaneously; the gate does not adjudicate between them, and Gate 2 is where the decision actually turns.

Gate 2 — Variant vs. consensus · FAIL — this is the binding gate

2A — Estimate variant: FAILS, and the reason is structural.

Both are immaterial, and immaterial in opposite directions. Consensus FY2026 revenue is not an independent forecast — it is company guidance ($1.59–1.60bn) transcribed, with an 18-analyst range spanning 1.2%. No estimate variant is available to anyone until guidance itself moves. On the short side the position is worse still: EPS estimates have been revised up monotonically over 90 days (FY2026 −$0.446 → −$0.359; FY2027 −$0.090 → −$0.079), and Q1 2026 beat on both revenue and non-GAAP EPS. There is no evidence for a downgrade cycle.

2B — Duration / optionality variant (MRD): FAILS — 2 of 4 legs.

Fewer than all four legs → 2B fails. The reference is explicit: "A duration variant supported by fewer than all four is narrative, not evidence, and fails the gate."

One further candidate variant, considered and rejected on its merits. Consensus FY2027 EPS of −$0.079 is computed on a pre-deal share count of ~180m; the post-deal count is ~215m, so per-share consensus is arguably ~16% too favourable. This is real arithmetic — but it is a publicly announced, seven-day-old fact that the Street will incorporate on close, not a differentiated view about why consensus is wrong. Gate 2 asks the latter. Recorded, not credited.

Gate 2 fails in both directions. This is the binding gate and it is not a close call.

Gate 3 — Catalyst resolving this specific disagreement · FAIL

A dated catalyst unquestionably exists: Q2 2026 earnings, Thursday 30 July 2026, 4:30pm ET (company-confirmed, announced 9 July 2026) — three days from the valuation date. Further out: the Personalis shareholder vote and close (late 2026/early 2027).

But the gate does not ask whether an event exists; it asks whether an event resolves the Gate 2 disagreement. The reference states the disqualifying case directly: "A catalyst that tests execution when the disagreement is actually about a multiple assumption does not satisfy this gate."

That is exactly the situation. Task 3 §4.5 decomposes the house-vs-Street gap on a shared revenue base and shows it is entirely the exit multiple: 4.59x vs 7.02x on FY2027E revenue, with house and Street within 1.1% on the numbers. A quarterly print tests execution against a guide that consensus has already transcribed. It cannot resolve a disagreement about the required rate of return and the terminal margin. With no qualifying Gate 2 variant, there is no disagreement for any catalyst to adjudicate.

Gate 3 fails by dependency and on its own terms.

Gate 4 — Scenario-weighted expected return · FAIL — but the hurdle sits INSIDE the range

Reported as a range with a flip point, per v1.4.2 item C4. Scenarios built on FY2027E revenue of $2,091m (post-deal), net debt $395m, shares 214.7m.

E[R] RANGE: −9.0% to +15.7%. Central: +2.1%. Cash hurdle: +4.7%.

EXPLICIT VERDICT CATEGORY: the hurdle sits INSIDE the E[R] range. Per v1.4.2, the honest statement is that Gate 4 is not determined by the analysis — it is determined by a judgement input. The central weighting fails by 2.6pp; a modestly more optimistic weighting passes comfortably.

Flip points (bull held at 25%, base = 75% − bear): - E[R] clears the 4.7% cash hurdle at a bear weight of 28.4% (versus the 35% used) — a 6.6pp swing in an unverifiable judgement input. - E[R] reaches zero at a bear weight of 40.3%.

Price flip points (targets held fixed): E[R] = 4.7% at $41.84; = 10% at $39.83; = 15% at $38.10.

Gate 4 is therefore scored FAIL on the central weighting, with the failure explicitly declared indeterminate. It is not the binding gate, and this matters: CALIBRATION_WATCH.md item B13 observes that Gate 4 carries zero marginal information conditional on a Gate 2 failure (0 of 16 names have ever failed Gate 2 and passed Gate 4). TEM does not break that pattern, and the decision does not rest here.

Gate 5 — Implementation feasibility · PASS for a long; FAIL for a short

Long side: feasible. Liquidity is ample and the 26.45% short float is a squeeze tailwind rather than a constraint. But options are systematically expensive here (an 11–27 volatility-point premium to realised, widest in short-dated puts), so any expression must be equity, not purchased premium — see §6.

Short side: fails. 26.45% of float short at a record high with 5.03 days to cover; borrow will be expensive and is not verifiable from available sources (stated rather than assumed away); and a pending all-stock acquisition creates mechanical merger-arb short flow in TEM that unwinds on close — pre-loaded squeeze fuel with a known expiry. This is CALIBRATION_WATCH.md item S1 in live form.

Gate 6 — Don't fight the tape without naming it · FAIL

The conflict is named explicitly, as the gate requires: a long here fights strong, accelerating negative momentum and is the textbook "falling knife" the gate specifically identifies as systematically expensive (Jegadeesh & Titman 1993; Asness/Moskowitz/Pedersen 2013).

And the tape is idiosyncratic, which makes it more informative, not less. Over twelve months: TEM −33.5% while GH +227.5%, TXG +225.4%, EXAS +136.9%, NTRA +85.8%, PSNL +84.6%. TEM is not falling with a sector — it is the only faller in a violently rallying peer group, underperforming by 120–260 percentage points. This forecloses the most common exculpatory reading ("the whole group de-rated") and points at something company-specific. The stock also fell −7.7% on the Personalis announcement and −18.6% over the following four sessions, so the most recent leg is a direct market verdict on management's capital allocation.

The gate can pass only with (a) explicit naming — done — and (b) either a catalyst strong enough to break the trend within the horizon, or a stated mitigation. Gate 3 has already failed, so there is no qualifying trend-breaking catalyst to invoke; and the reference's own remedy for the remaining case is prescriptive: "the decision reverts to Watchlist with the entry rule stated as the upgrade trigger." That entry rule is specified in §4.

Gate 6 fails.

3. Classification and the tech/growth trap check

Against the reference's classification table, TEM maps to: fundamentals mixed-to-improving, valuation ambiguous, expectations gap absent → No position.

The tech/growth trap, checked in the direction it actually applies here. The reference warns repeatedly about shorting a rich multiple in tech/growth. The mirror error is the live risk on this name: buying a collapsed multiple on the assumption that it is unearned. Applying the same discipline in reverse:

• "Do not treat richer than the peer group's historical multiple as evidence of anything" — inverted, do not treat cheaper as evidence either. TEM at 6.16x against NTRA 14.4x / GH 18.4x looks like a 60% discount until the only completed control transaction in the set is added: Abbott paid 6.65x for Exact Sciences in March 2026. Against that anchor TEM is roughly fairly valued.
• "Require an actual data point, not a general tendency" — the improving-margin data point is genuine. But so is +145.6% asset growth, a 4/9 F-score, SBC at 129% growth, and a widening GAAP operating loss.
• The bull case is named honestly as a live possibility, not a straw man. It is that TEM is a genuine multimodal-data platform whose Insights business (126% NRR, +44% growth) and MRD position compound into a business the market is currently valuing as a commodity lab. This memo has no evidence against that case. What it has is the observation that the case is a terminal-value case — 72% of DCF value sits past FY2035 — and that there is no near-term expectations gap through which to express it. "I find the bull story plausible but unproven" is not Gate 2 evidence and is not dressed up as such.

Valuation cannot upgrade a position from Watchlist. Even a clean Gate 4 would not overcome the failure of Gates 2, 3 and 6 — and Gate 4 is not clean.

4. Investment decision

INVESTMENT DECISION: WATCHLIST  (long bias, not actionable)

GATES: 1(Mechanism): PASS (qualified — corroboration splits; asset growth +145.6% contradicts)
       2(Variant vs. Consensus): FAIL  <-- BINDING
       3(Catalyst): FAIL (event exists 30-Jul-2026, but tests execution, not the multiple disagreement)
       4(Expected Return): FAIL on central weighting — HURDLE SITS INSIDE THE E[R] RANGE
       5(Feasibility): PASS (long) / FAIL (short — 26.45% of float short, merger-arb squeeze fuel)
       6(Momentum/Tape): FAIL — AGAINST. Named: -58.4% from high, -29.1% vs 200-DMA,
         at 52-week lows, while every peer is +85% to +228%. No qualifying mitigation available
         because Gate 3 already failed.

ENTRY: no position at $42.91  |  TARGET: $43.93 (probability-weighted)  |  INVALIDATION: $27.38 (bear case)
TIME HORIZON: next assessable point Q2 2026 earnings, Thursday 30 July 2026 (3 days);
              full re-underwrite on Personalis close, late 2026 / early 2027

SCENARIO-WEIGHTED E[R]: +2.1% net of costs (central 35/40/25)
       RANGE: -9.0% to +15.7% across plausible weightings
       CASH HURDLE 4.7% SITS INSIDE THE RANGE -> Gate 4 is a judgement input, not a result
       FLIP POINT: clears the hurdle at a 28.4% bear weight (vs 35% used); E[R]=0 at 40.3%
       Simple: risking ~36% to make ~55%

SIZING: Conviction: LOW (composite 0.380)  |  Volatility: HIGH (realised 65.7% 1yr, beta 2.50)
        Sizing grid, Low x High = "below minimum size - round down to Watchlist"
        Resulting size: 0.0% of book

VEHICLE: None initiated. On conversion: OUTRIGHT EQUITY, not options — implied vol 77-93% against
        realised 66-72% is an 11-27 point variance-risk premium, widest in the short-dated puts.
        Buying premium here is negative expected value even with a correct directional view.

INVALIDATION TRIGGERS (move to AVOID):
- FY2026 revenue guidance cut below $1.55bn, or FY2026 Adjusted EBITDA guidance cut below $30m
- Any equity raise priced below $40.00, or a Personalis cash election that pushes pro-forma
  net debt above $1.2bn
- Two consecutive quarters in which SBC growth again exceeds revenue growth by more than 50pp
  (Q1 2026: +129% vs +36%)
- Diagnostics gross margin declines sequentially in any two consecutive quarters
  (the Gate 1 mechanism dies)
- Personalis deal terminated, or re-cut at a higher price

Conviction composite (per the required weighting):

Consensus positioning — stated, not skipped. A Watchlist-with-long-bias is with a bullish Street (Buy; $66.57 average; 7 Strong Buy / 1 Buy / 7 Hold / 1 Strong Sell) and with an improving revision trend — but against a market that has taken the stock down 58% and against a record 26.45% short float. Street targets have been drifting down (~$72 → ~$66-67) while ratings have not changed, which is the classic pattern of a sell side marking to market rather than changing its mind. Going with a consensus that the tape is actively repudiating is its own risk, distinct from the contrarian case, and it is named here rather than absorbed into the volatility tier.

5. Conversion triggers — what would make this a position

Per the reference, a Watchlist entry must state the specific evidence that would satisfy the failed gates. Because Gates 2, 3 and 6 all fail, a price move alone is not sufficient and no price-only route is offered. Both routes below require a gate to be fixed on its own terms.

Route A — fix Gate 2 (the binding gate)

Any one of:

1. Guidance-led numbers variant. FY2026 revenue guidance raised above $1.65bn (i.e. management breaks the guide that consensus is transcribing), creating a real gap between the house forecast and a stale Street. First checkable: 30 July 2026.
2. MRD becomes a modellable line. TEM discloses MRD revenue and ASP (not just volume) for two consecutive quarters, and sell-side models begin carrying an explicit MRD line — observable as an FY2028 consensus revenue revision of >3% not attributable to the arithmetic of the Personalis close. This would satisfy 2B legs 2 and 4 on their own terms.
3. Consensus and guidance decouple downward — 18-analyst FY2027 revenue dispersion widens beyond ±5% with the mean falling, creating a short-side variant. (Would flip the bias, not confirm it.)

Route B — fix Gate 6 (required in addition, not instead)

Either: - A weekly close back above the 200-day moving average (currently ~$60.50 and falling — the level will migrate down, so it must be re-read at the time, not fixed at $60.50); or - A confirmed higher-high / higher-low sequence off the low with the 50-day MA (~$51.15) reclaimed and held for two weeks.

Both routes must fire. Fixing Gate 6 alone converts a value view into a momentum view with no variant behind it; fixing Gate 2 alone means buying into an accelerating downtrend, which the gate exists to prevent.

Pre-authorised entry mechanism (documented so conversion is executable, NOT to be used before the trigger fires)

Should both routes fire at a lower price, the documented mechanism is a cash-secured put rather than a market buy:

Selling a put IS long exposure. This must not be initiated before the conversion triggers fire. Recorded here only so the conversion does not require re-deriving the mechanism under time pressure.

6. Vehicle selection (for the conversion case only — nothing is initiated now)

No catalyst-specific tactical recommendation is made for the 30 July print. The reference makes a tactical contract additive when Task 6 identifies a dated catalyst — a dated catalyst does exist, but with no Gate 2 variant there is no differentiated view for it to express, and the options market is charging an 11–27 volatility-point premium to take the other side. Recommending a contract here would be manufacturing activity, not edge.

IV-crush disclosure, stated even though no options trade is recommended: the 14-Aug $43 straddle carries 87–93% implied volatility three days before a scheduled print. That implied volatility will collapse immediately after the event regardless of direction, and a correct directional call could still lose money if the move is smaller than ±15.5%.

7. The questions every decision must answer

What is the market wrong about? On the evidence assembled — nothing demonstrable. Consensus revenue equals guidance; house is within 1.1%; estimates are being revised up. The market and the Street disagree violently with each other about the multiple (4.59x vs 7.02x on FY2027E), but this memo has no evidence establishing which is right, and its own DCF (1.38x implied at $12.80) sits below both.

What evidence supports a differentiated view? Insufficient. The strongest genuinely differentiated finding is the SBC/dilution arithmetic — SBC +129% vs revenue +36%, ~19% dilution for ~5.6% of revenue, equity issued at 5.2x sales to buy 16.7x-sales revenue at a 13.7% gross margin. That is a real, quantified, generatively-sourced finding and it points bearish. But it does not translate into a revenue or EPS forecast below consensus, and the Street models non-GAAP where it does not appear.

Why would the disagreement resolve now? It would not. It is a terminal-value and cost-of-capital disagreement; 72% of DCF value sits beyond FY2035. Q2 earnings on 30 July tests execution against a guide, not the multiple.

What prevents the company growing into its valuation? Nothing structural — this is the honest answer, and it is why no short is recommended. At 25% growth with expanding Diagnostics margins, TEM can plausibly grow into 4.59x within three years without the multiple moving at all.

What would invalidate the view? Listed in §4.

What is the adverse scenario? For the long bias: the bear case at $27.38 (−36.2%), reached via decelerating growth, a Personalis integration that dilutes without adding margin, and a forced raise. Given 72.4% three-month realised volatility, a worse outcome is well within one standard deviation.

Why is taking the position superior to not owning it? It is not. Net expected return of +2.1% against a 4.7% cash hurdle, at 65.7% realised volatility and a 2.50 beta, with the binding gate failing on a structural fact rather than a judgement. Cash is the better position.

8. Book write-back

The following is written to portfolio_book.json (every decision writes back, including Watchlist entries with conversion triggers):

{
  "ticker": "TEM",
  "company": "Tempus AI, Inc.",
  "bias": "long",
  "status": "watchlist",
  "as_of": "2026-07-27",
  "memo_version": "1.4.2 (Tasks 1/2/3/5 only)",
  "price_at_review": 42.91,
  "probability_weighted_target": 43.93,
  "scenario_targets": {"bear": 27.38, "base": 44.42, "bull": 66.33},
  "scenario_probabilities": {"bear": 0.35, "base": 0.40, "bull": 0.25},
  "dcf_base": 12.80,
  "net_expected_return_at_spot": 0.021,
  "expected_return_range": [-0.090, 0.157],
  "hurdle_inside_range": true,
  "cash_hurdle": 0.047,
  "gate4_flip_bear_weight_to_hurdle": 0.284,
  "gate4_flip_bear_weight_to_zero": 0.403,
  "gates_passed": ["1_causal_mechanism_qualified", "5_feasibility_long_only"],
  "gates_failed": ["2_variant_vs_consensus", "3_catalyst", "4_expected_return", "6_momentum_tape"],
  "binding_gate": "2_variant_vs_consensus",
  "conviction_composite": 0.380,
  "conviction": "LOW",
  "volatility_tier": "HIGH",
  "realised_vol_1y": 0.657,
  "beta_1y_vs_spy": 2.50,
  "size_if_converted_pct_of_book": 0.0,
  "conversion_requires_both": {
    "gate2_route": "FY2026 revenue guidance raised above $1.65bn, OR MRD revenue+ASP disclosed two consecutive quarters AND FY2028 consensus revised >3% not attributable to Personalis close arithmetic",
    "gate6_route": "weekly close above the 200-DMA (~$60.50 and falling, re-read at the time), OR higher-high/higher-low sequence with the 50-DMA (~$51.15) reclaimed and held two weeks"
  },
  "pre_authorised_entry_mechanism": {
    "contract": "TEM261120P00035000",
    "description": "20-Nov-2026 $35 cash-secured put",
    "bid": 3.32, "iv": 0.811, "delta": -0.239,
    "effective_basis_if_assigned": 31.68,
    "note": "NOT to be used before both conversion triggers fire - selling a put IS long exposure"
  },
  "invalidation_to_avoid": [
    "FY2026 revenue guidance cut below $1.55bn or Adjusted EBITDA guidance below $30m",
    "equity raise priced below $40.00, or Personalis cash election pushing pro-forma net debt above $1.2bn",
    "two consecutive quarters with SBC growth exceeding revenue growth by more than 50pp",
    "Diagnostics gross margin declines sequentially in two consecutive quarters",
    "Personalis deal terminated or re-cut higher"
  ],
  "next_review": "2026-07-30 (Q2 2026 earnings), then Personalis close",
  "cluster": "healthcare/precision-oncology - with GH, NTRA, TXG, TWST on the watchlist; 25% sector cap becomes live if more than one converts",
  "max_pairwise_correlation": {"PSNL": 0.560, "TXG": 0.472, "NTRA": 0.461, "TWST": 0.460, "GH": 0.401},
  "no_options_recommendation": "IV 77-93% vs realised 66-72%; house expected move does not exceed the 15.5% implied move for the 30-Jul print"
}

(Task 9 ledger append is out of scope for this run by instruction, and no specific options recommendation was made, so there is nothing to log.)

9. Calibration note (standing obligation per SKILL.md)

Two entries are proposed for CALIBRATION_WATCH.md. Neither loosens any gate — logged for the user to rule on.

1. TEM is the first name in coverage that actually tests item B1/B11. That item records that the "does Gate 6 wrongly block beaten-down names" question "requires a deliberately-selected sample of names actually trading BELOW their 200-day — it cannot be tested opportunistically," and notes the GH/TWST/TXG experiment failed to test it because all three turned out to be top-decile momentum names. TEM trades 29.1% BELOW its 200-day, at a 52-week low, −58.4% from its high. Gate 6 is a genuine blocker here rather than a tailwind. However, TEM is a weak test of the charitable case for B1, because Gate 2 binds independently and Gate 6 is not decisive — deleting Gate 6 entirely would not change the verdict. Recorded as the first data point of the right type, with that caveat stated.

2. Item B9's exculpatory reading gains an out-of-sample confirmation. B9 records that 16 of 16 house targets sat below spot at a median −46.1%, and offers the exculpatory split that "handed a name that has not run, the house prices it at fair value, not at a discount." TEM's house probability-weighted target of $43.93 sits +2.4% ABOVE spot — the first above-spot target in coverage, on the most beaten-down name yet underwritten. That is consistent with H2 (funnel) and inconsistent with H3 (systematic valuation-process bias). Recorded as counter-evidence against the framework being too strict, which is the side of this file that most needs feeding.

A third observation is recorded as a structural question, not a complaint: on a name where 18 analysts transcribe company guidance into a 1.2%-wide consensus band, Gate 2A is unsatisfiable by construction for the guided year — nobody can hold a material variant against a number that is definitionally the guide. That is arguably correct behaviour (there genuinely is no variant), but it means Gate 2 will systematically bind on tightly-guided names regardless of their merit, and it interacts with existing item B12 on the unwritten materiality threshold. Flagged; not acted on, per the rule that gates are never loosened unilaterally.

Framework: ~/.claude/skills/investment-memo/references/trade-construction.md v1.4.2, applied as written. No amendment was adopted, anticipated or applied in this run.