The Big Picture

The models are, collectively, in a sour mood. Across 23 companies and 5 trading days, the aggregate picture is one of mild but broadening skepticism: four of five models now sit in negative average upside territory, and the one exception — GPT at +0.3% — barely clears zero. This is not a crash signal; it is something subtler and arguably more interesting. The models are telling us that current market prices have run slightly ahead of what discounted cash flow discipline can justify, and they are saying it with unusual consistency.

The average upside across all models ranges from -4.4% (DeepSeek) to +0.3% (GPT). That 4.7 percentage point spread between the most bearish and most bullish model is itself informative — it reflects genuine architectural differences in how these systems weight near-term earnings momentum versus long-run terminal assumptions, not random noise.

Trends

Two names generated multi-day directional conviction this week, and they could not be more different in character.

Wärtsilä (WRT1V) is the more striking case. Four consecutive rising consensus days, zero down-days, a 6.57% range across those sessions. For a Finnish marine and energy technology company trading at €39.63 against a consensus target of €31.78 — implying -20% downside — this is a paradox worth unpacking. The models are simultaneously bearish on absolute valuation and consistently revising targets upward on a daily basis. What this likely reflects is models responding to incremental data inputs (order flow signals, energy transition narratives) that push near-term estimates higher, even while the DCF anchor remains well below spot. Momentum and value are pulling in opposite directions, and the models are faithfully registering both.

Alphabet (GOOGL) shows three rising days out of four, but the 0.06% range is almost comically tight. This is model consensus behaving like a metronome — small, regular, directionally consistent adjustments. At a spot of $338.89 and a consensus target of $417.06, GOOGL carries one of the more constructive setups in the universe this week.

Sector Signals

The most important rotation story this week is what *didn't* happen in technology. The sector remains the most bullish in the universe at +4.5% average upside across 8 companies, but that figure fell from +7.2% last week — a -2.7 point shift, the sharpest deterioration of any sector. When the models' favorite sector starts losing altitude, it warrants attention even if the absolute reading stays positive.

Energy staged the week's most meaningful recovery in model sentiment, improving +3.6 points to -18.2% — still deeply underwater, but the direction of travel matters. With only two companies (XOM and NESTE), this is a thin sample, but both names saw target price increases this week (NESTE led all movers at +7.6%). Healthcare also gained +2.6 points, reaching +16.6% average upside — the highest absolute reading in the universe, though again a two-company sector warrants caution about representativeness.

Industrials and consumer both slipped modestly. The industrials deterioration is worth noting given Wärtsilä's trending behavior — the sector-level signal and the individual stock signal are pointing in opposite directions, which is precisely the kind of tension that makes sector aggregates misleading.

Note that materials (UPM alone) and telecom (Elisa alone) are single-company sectors. Treat their readings as individual stock signals wearing a sector label.

What the Models Reveal About Themselves

The divergence in bias_shift this week is the most behaviorally revealing data in the entire dataset. Gemini moved the most bearishly, shifting -1.1 points to -2.4%. Claude and Grok both shifted -0.6 points. DeepSeek barely moved at -0.2 points — consistent with its reputation for anchored, low-variance outputs (terminal growth standard deviation of just 0.10, versus Claude's 0.43).

Then there is GPT, which moved +1.0 point — the only model to become more bullish while everyone else turned more cautious. This is not necessarily a contrarian signal to act on, but it is a structural tell. GPT's CAGR standard deviation of 3.9% is the tightest in the group, and its terminal growth is pinned at exactly 2.0% with zero variance. GPT is the most mechanically consistent model here, and its bullish drift may reflect less sensitivity to macro sentiment shifts and more fidelity to a stable internal valuation framework. Whether that is a feature or a limitation depends entirely on your view of the current environment.

DeepSeek's cost profile continues to demand acknowledgment: $0.2629 total for the week's full run, $2.29 per thousand valuations, against Claude's $40.99. The performance gap does not justify an 18x price premium in any obvious way — DeepSeek achieved 100% validity, tighter terminal growth assumptions, and reasonable upside dispersion.

Where the Framework Breaks

The upside_pct figures in the company table are expressed as fractions of 1 rather than percentages — ORNBV shows 0.4 meaning +40% upside, not 0.4%. This is a display normalization artifact, but it creates a genuine interpretive trap. A reader scanning the table sees TSLA at -0.2 and META at +0.3 and might conclude the models are nearly indifferent between them. In reality, Tesla sits -15% below consensus and Meta sits +27% above spot. The compression of a wide valuation distribution into a normalized scale erases the very dispersion that makes cross-stock comparison meaningful. The framework works well for ranking; it obscures magnitude.

The Model Scorecard