V-Score Card

DIMENSION   WT     SCORE   WEIGHTED    EVIDENCE ANCHOR
──────────────────────────────────────────────────────────────────────
C (0.25)    0.25     3      0.75       16yr dev, 3 unified solutions on shared store.
                                       Fork replicates 2021 engine; full stack 1-2yr
                                       re-derive. 10-K: "limited technological barriers."
                                       Decaying → 2 within 24mo as frontier models
                                       compress re-derivation timelines.

E (0.22)    0.22     2      0.44       OpenSearch fork maintained by AWS (1,400+
                                       contributors). 10-K line 2618: "limited
                                       technological barriers to entry." GRR undisclosed
                                       (NRR 112% but GRR est. 90-95%). 51% rev
                                       self-managed = already portable. CISA = 1 customer
                                       ($26M), not regulatory mandate. Median switching
                                       timeline: 4-8 months. c_ℓ finite, not ∞.

U (0.18)    0.18     3      0.54       Search + Observability + Security on shared
                                       Elasticsearch store. ≈12 workflows across 5-7
                                       technical depts (DevOps, IT, SOC, SRE, Product).
                                       Cross-module data flows are real. But concentrated
                                       in technical functions — doesn't reach HR, Finance,
                                       Sales, Legal.

A (0.12)    0.12     3      0.36       MCP server shipped (Microsoft Foundry catalog),
                                       A2A live. 5.5B+ downloads, 3,000+ AI customers.
                                       But zero AI-specific revenue disclosed. Pinecone,
                                       pgvector, ChromaDB compete for RAG; Datadog, CRWD
                                       for obs/sec. Well-known, not default.

M (0.15)    0.15     3      0.45       ≈21,500 customers, 1,660+ >$100K ACV, $1.65B RPO
                                       (+22% YoY). Embedded dashboards and SIEM rules =
                                       real switching friction. But fork = escape route.
                                       No counterparty network effects. Patent count
                                       undisclosed.

F (−0.06)  −0.06     3     −0.18       3 license regimes (Apache/SSPL/AGPL), consumption
                                       pricing opacity (class action trigger), self-managed
                                       cluster complexity. Good APIs + self-serve partially
                                       offset.
──────────────────────────────────────────────────────────────────────

Arithmetic

V = 0.25(3) + 0.22(2) + 0.18(3) + 0.12(3) + 0.15(3) − 0.06(3)
  = 0.75 + 0.44 + 0.54 + 0.36 + 0.45 − 0.18
  = 2.36

Gates

G₁ = 𝟙[E > 1] = 𝟙[2 > 1] = 1         ← PASS
G₂ = 𝟙[A > 1 ∨ (C+E+U) ≥ 12]
   = 𝟙[3 > 1 ∨ 8 ≥ 12]
   = 𝟙[TRUE ∨ FALSE] = 1               ← PASS (A binding)

V = 2.36 × 1 × 1 = 2.36

Result

Dimension Analysis

C = 3 — Compound Cognition: Real but Melting

Elastic has 16 years of continuous development layered into three unified solutions (Search, Observability, Security) sharing a single Elasticsearch data store. Cross-module data flows — security events generate observability alerts, ML models trained on observability data power security detections — represent genuine architectural complexity.

The fork cleaves cognition into two layers. Layer A (the engine): inverted indices, BM25, distributed sharding — replicated by OpenSearch in months. This is open-source Lucene IP, not Elastic's. Layer B (the stack): BBQ vector quantization, Agent Builder, ESQL, Logsdb, 5 years of post-fork innovation. OpenSearch has been trying for 4+ years and still lags materially.

But the half-life is shortening. BBQ is a published quantization technique. ESQL is a SQL dialect. Agent Builder is RAG + tool orchestration — every vector DB vendor ships one. The only genuinely hard-to-re-derive component is the cross-module tuning accumulated across thousands of deployments: how Search, Observability, and Security interact on a shared store at enterprise scale.

Forward curve: C = 3 today → ≈2.5 in 12 months → 2 in 24 months as frontier models compress re-derivation and OpenSearch closes the gap.

10-K FY2025, line 2618: "Limited technological barriers to entry into the markets in which we compete." The defendant confessed. Securities counsel doesn't add this language for companies with irreproducible domain knowledge.

E = 2 — Irreducible Infrastructure: The Fork Is the Ceiling

This is the dominant discriminator (w = 0.22, calibrated gap = 3.8 between dead and alive averages in the V-Score framework). E = 2 is correct and possibly generous.

The fork is dispositive. AWS maintains OpenSearch as a first-party managed service with 1,400+ contributors. When the world's largest cloud provider forks your core product and offers it for free, your infrastructure is not irreducible. No company with a maintained fork by its largest channel partner gets E ≥ 3.

Switching timelines are finite and quantifiable:

Weighted average c_ℓ ≈ 4-8 months for the median customer. Not ∞. Not even particularly high by enterprise SaaS standards.

GRR silence is evidence. NRR at 112% is table stakes (CrowdStrike 115%+, Datadog 115-120%). Elastic discloses NRR but not GRR. CrowdStrike discloses GRR ≈97% because it's strong. Estimated GRR: 90-95% (5-10% annual logo churn). If GRR were fortress-grade, they'd tell you.

The cliff risk in E: If OpenSearch reaches functional parity, E drops from 2 to 1. At E = 1, Gate 1 fails (G₁ = 𝟙[E > 1] = 0) and V → 0 regardless of every other dimension. P(E → 1 within 24mo) ≈ 30%. Expected forward V including gate risk ≈ 1.48. The gate is a cliff, not a slope.

U = 3 — Ecosystem Breadth: Technical Monoculture

Three solutions across ≈12 workflows in 5-7 technical departments. The shared data store and resource-based pricing (not per-solution) create natural cross-solution expansion. CEO highlights customers starting with search and expanding to observability and security.

But breadth is concentrated: DevOps, IT, SOC, SRE, Product, AI/ML Engineering. Elastic doesn't touch HR, Finance, Sales, Marketing, Legal, Procurement. Compare ServiceNow (U = 4-5, IT + HR + customer service + procurement) or SAP (U = 5, every department).

Calibration: similar to Atlassian (developer-focused, U = 3) and Palo Alto Networks (security-focused, U = 3).

A = 3 — Distribution: Well-Known, Not Default

Elastic shipped MCP and A2A early. It's in the Microsoft Foundry Tool Catalog. 5.5 billion cumulative downloads. 3,000+ AI customers. 470+ AI customers above $100K ACV. The surface area is impressive.

But no AI-specific revenue is disclosed. The company brands itself "Elastic, the Search AI Company" while reporting zero AI revenue attribution in MD&A. 3,000 AI customers is an engagement count, not a revenue line. Marketing narrative vs. financial reality.

For agents specifically: Elastic is discoverable (massive training data presence, functional APIs) but not the default route. Vector search has Pinecone, pgvector, ChromaDB, Weaviate. Observability has Datadog. Security has CrowdStrike Charlotte AI. Agents have many options. A = 4 requires "agents prefer you" — not yet proven.

M = 3 — Ecosystem Gravity: Meaningful but Portable

$1.65B RPO growing 22% YoY. >$1M ACV commitments growing 30%+. Customers are signing forward commitments. Embedded Kibana dashboards, SIEM detection rules, ML jobs, and integration workflows create real switching labor.

But the fork provides an escape route for basic workloads. No counterparty network effects (each deployment is independent). Patent portfolio is thin ("a number of active patents" — no count disclosed). One channel partner accounts for 10-12% of revenue (concentration risk).

Gravity comes from data weight and tooling investment, not structural lock-in or network externalities.

F = 3 — Friction: License Complexity Tax

Three license regimes (Apache 2.0, SSPL/Elastic License, AGPL) depending on version creates genuine enterprise confusion. Consumption-based pricing opacity triggered the August 2024 guidance reset and securities class action. Self-managed cluster management requires meaningful operational expertise.

Partially offset by strong developer experience ("data to dashboard in minutes"), self-serve paths (free trial, AGPL download, marketplace), and the new Cloud Serverless offering.

Sensitivity

                   C  E  U  A  M  F  →  V_raw  G₁  V_final  Tier
Base               3  2  3  3  3  3    2.36    1   2.36     AT_RISK
C decay (24mo)     2  2  3  3  3  3    2.11    1   2.11     AT_RISK
E upgrade (SIEM)   3  3  3  3  3  3    2.58    1   2.58     AT_RISK
Bull (C4 E3 A4)    4  3  3  4  3  3    2.95    1   2.95     AT_RISK
E decay (fork)     3  1  3  3  3  3    2.14    0   0.00     DEAD
Full decay         2  1  3  3  3  3    1.89    0   0.00     DEAD

ESTC stays AT_RISK under every plausible parameterization. Even the maximum bull case (C = 4, E = 3, A = 4) yields V = 2.95 — five basis points short of EMBEDDED. The fork anchors E, and E controls the gate.

Thermodynamic Summary

Elastic sits in a high-entropy equilibrium. The energy barriers to exit (switching costs) are real but finite — measured in months of migration labor, not years of irreplaceable infrastructure. The fork acts as a permanent entropy source: it ensures the system can never reach a low-entropy locked state because an alternative crystallization path always exists.

Free energy decomposition:

The binding energy (what holds customers) is operational state: dashboards, detection rules, ML models, team expertise. This is labor lock-in, not physics lock-in. It degrades as tooling improves — migration tools get better, AI assists re-creation of detection rules, and OpenSearch's migration compatibility improves with each release.

The activation energy to leave (c_ℓ ≈ 4-8 months) is above the thermal noise of routine vendor evaluation but below the threshold that prevents rational switching when better alternatives emerge. For SIEM-heavy deployments, c_ℓ extends to 12-18 months — genuine friction, but concentrated in one of three solutions.

Phase transition risk: E = 1 is a phase boundary. If crossed (OpenSearch functional parity), Gate 1 fails and the system collapses to V = 0 discontinuously. P(phase transition, 24mo) ≈ 30%. This is not priced in the continuous V-Score but is captured in the expected forward V ≈ 1.48.

Entropy production rate: C decays at approximately 0.5 points per year as frontier models compress re-derivation timelines and competitors close feature gaps. The 5-year post-fork innovation advantage is a melting ice cube, not a frozen moat.

Regime Context (T = 15 weeks)

These metrics measure the regime, not the stock. They do not gate the verdict.

METRIC                  VALUE       INTERPRETATION
─────────────────────────────────────────────────────────────────
IR_i (vs IGV)           +0.577      α not significant (p=0.763). Noise.
IR_i (vs WCLD)          −0.211      Sign flips on benchmark choice. No signal.
ρ_intra                  0.618      Elevated. Sector-dominated regime.
ρ_intra (peak, 20d)      0.916      Near-indiscriminate at selloff worst.
%Idio variance           46.4%      Below 75% target. Factor name, not stock name.
β_sector (IGV)           1.393      ESTC = 1.4x leveraged software bet.
β_market (SPY)          −0.176      Zero after sector control.
Excess return (15w)
  vs IGV                −6.1%       Idiosyncratic component is negative.
  vs WCLD               −9.3%

Why IR is uninformative: With ρ_intra = 0.62 and %Idio = 46%, over half the variance is sector. The regression residuals are white noise (Ljung-Box all p > 0.43) with a fat left tail (skew = −2.36, kurtosis = 13.26) — one or two large idiosyncratic shocks in an otherwise factor-dominated window. IR measures the regime, not the alpha.

V(s) ⊥ r_sector(t): The V-Score is scored against structural properties (fork existence, entrenchment depth, cognition durability) that don't change with the sector's 15-week return. A software sector selloff does not make the fork disappear or GRR materialize.

δ — The Story

V          = 2.36    (structural assessment)
V_market   ≈ 2.0-2.3 (implied by 30-35% excess drawdown beyond sector)
δ          = V − V_market ≈ +0.1 to +0.3

δ is near zero. The market is approximately correct about ESTC's structural position.

The 49% YoY drawdown vs the sector's 14-20% captures three ESTC-specific risks: (1) the fork caps entrenchment, (2) consumption pricing opacity triggered a class action and guidance reset, (3) "Search AI Company" branding without AI revenue disclosure. These map directly to E = 2, the undisclosed GRR, and the A-score's marketing-vs-reality gap.

There is no mispricing to exploit. The market isn't applying a uniform sector discount that creates a structural gap — it's pricing ESTC's specific vulnerabilities through a different channel (price) than we do (V-Score), and arriving at approximately the same conclusion.

Conviction Weight and Basket Verdict

κ = (V − 3.0)⁺ = (2.36 − 3.0)⁺ = 0.00

w_i = W_S · κ_i / Σ_j κ_j = W_S · 0 / Σ_j κ_j = 0

Verdict: FILTER. Zero conviction weight. Not because the regime obscures signal — because the structural score doesn't clear the threshold. κ is regime-invariant by construction: even in a calm market with %Idio at 85% and statistically significant alpha, V = 2.36 < 3.0 → κ = 0.

The fork is the binding constraint. Every other dimension scores 3. E = 2 is the outlier, and E controls the gate. If E were 3, V would be 2.58 — still AT_RISK, still κ = 0. ESTC would need E = 3 AND C = 4 AND A = 4 just to reach V = 2.95. The structural profile cannot support basket inclusion under any reasonable parameterization.

What would change this: E ≥ 3 requires the fork to become non-viable (AWS abandons OpenSearch) or Elastic to build infrastructure so differentiated that the fork becomes irrelevant (regulatory mandate, unique data gravity, or a self-reinforcing network effect that doesn't currently exist). None of these are on a visible trajectory.

Evidence Table

All primary claims cite SEC filings (10-K FY2025 filed 2025-06-10, 10-Q Q3 FY2026 filed 2026-02-27) or earnings transcripts (Q3 FY2026, 2026-02-27).