V-Score Card

TICKER:         BBAI (BigBear.ai Holdings, Inc.)
V-SCORE:        1.88
VERDICT:        COLLAPSED
κ (conviction): 0.00  (V < 3.0 → excluded from basket)

GATES
  G₁ (E > 1):           PASS  (E = 3)
  G₂ (A > 1 ∨ Σ ≥ 12):  PASS  (A = 2 > 1)

FAST SCREEN (Bustamante):  1/3
  Proprietary data?      NO   — processes government's data, doesn't own it
  Regulatory mandate?    YES  — FedRAMP + IL5/IL6/TS authorization
  Transaction-embedded?  NO   — analytics/tooling, not the transaction rail

DIMENSIONS
  C (w=0.25):  2   GenAI wrapper on third-party LLMs, siloed acquisitions
  E (w=0.22):  3   FedRAMP/IL5/IL6/TS — real but replicable in 12-18mo
  U (w=0.18):  2   5 products from 4 acquisitions, no cross-module flows
  A (w=0.12):  2   100K DoD users (walled garden), zero commercial API
  M (w=0.15):  2   Customer A vanished ($21.3M→$0), no patents, no network effects
  F (w=−0.06): 3   61% T&M, integration required, classified access overhead

ARITHMETIC
  V = 0.25(2) + 0.22(3) + 0.18(2) + 0.12(2) + 0.15(2) − 0.06(3)
    = 0.50 + 0.66 + 0.36 + 0.24 + 0.30 − 0.18
    = 1.88

  G₁ · G₂ = 1 · 1 = 1
  V = 1.88 · 1 = 1.88

TIER: COLLAPSED (V < 2.0)

Regime Context

Window: 15 weeks (2025-12-26 → 2026-04-07, 69 trading days)

FACTOR REGRESSION: r_BBAI = α + β·r_SPY + γ·r_ITA + ε
  α̂ (ann):     -157.6%    (t = -1.26, p = 0.213)
  β (SPY):       1.64      (t = 2.23)
  γ (ITA):       1.38      (t = 3.42)
  R²:            0.389
  σ_idio (ann):  64.2%
  %Idio Var:     61.1%     (below 75% target)

IR_i = α̂ / σ_idio = -157.6% / 64.2% = -2.453
δ_i  = V_i − V_market = 1.88 − 3.0 = -1.12

ρ_intra (defense sector, avg pairwise): 0.406
ρ_resid (after SPY+ITA removal):        0.339

Period returns: SPY -4.3%, ITA +1.5%, BBAI -42.8%
BBAI excess vs sector: -44.3%

PEER IR (15wk):
  BAH   +0.230  (98% idio — clean)
  SAIC  -0.089  (94% idio — flat)
  LDOS  -1.477  (91% idio — idio-specific pain)
  PLTR  -1.458  (76% idio — beta drag)
  BBAI  -2.453  (61% idio — worst in class)

Regime interpretation: ρ_intra = 0.41 — the idiosyncratic channel is wide open. This is a discriminating market, not an indiscriminate selloff. ITA is up 1.5%. BAH and SAIC are positive. BBAI's -42.8% drawdown is company-specific, not regime-driven. IR = -2.45 measures real negative alpha, not noise from correlated selling.

V(s) is orthogonal to regime — scored against structural properties, not price. IR does not gate the verdict. But when the regime is discriminating and the market agrees with the structural score, that's confirmation, not coincidence.

Dimension Analysis

C — Compound Cognition: 2/5

The growth engine is a GenAI wrapper. Ask Sage "leverages cutting-edge third-party large language models" — the intelligence is rented, not crystallized (10-K FY2025). The orchestration layer is re-derivable in weeks.

The company's own auditors valued Ask Sage technology at $65.2M with a 7-year life and the FedRAMP certification at only $6.0M. Five products from four acquisitions (Ask Sage, Pangiam, ProModel, CargoSphere) share no architectural integration — "solutions can be sold individually or combined" via paid integration services, not platform-native data flows (10-K FY2025). $155.6M in goodwill impairments across two years confirms the acquisitions destroy value faster than they crystallize knowledge.

ProModel carries decades of simulation expertise (discrete-event modeling for defense readiness, supply chain, hospital workflows) — genuine domain knowledge that is not trivially replicated. But it's the legacy, not the growth thesis. ConductorOS validated in Talisman Sabre military exercises has real edge orchestration capability. Neither compounds with Ask Sage.

The CTO departed 9 weeks after the Ask Sage acquisition (Q4 2025 earnings call). Institutional knowledge walking out the door on the flagship product.

Frontier model test (2-year horizon): ≈65-70% of BBAI's task domain is re-derivable. Ask Sage orchestration (weeks), VeriScan biometrics (commoditized), CargoSphere CV (commoditizing). The 30-35% that resists (ProModel calibration, ConductorOS military integration) is real but legacy.

E — Irreducible Infrastructure: 3/5

FedRAMP authorization at IL5/IL6/Top Secret is the single real barrier. For classified government tasks, the local cost c_ℓ(τ) is effectively prohibitive — you cannot run unapproved models on classified networks. Ask Sage is "the first platform of its kind to have been FedRAMP authorized" for generative AI (10-K FY2025). 579 security-cleared employees create a workforce moat with 6-18 month replication timelines per person.

But the barrier is finite, not infinite. The auditors valued the certification at $6.0M — 2.2% of the $271.6M Ask Sage purchase price. If FedRAMP were truly irreducible, it would be the most valuable intangible in the deal. Instead, it's the smallest.

Customer A proves switching works: $21.3M in FY2024, $0 in FY2025 — the largest customer vanished entirely (10-K concentration table). Backlog collapsed 41% ($418M to $248M), with priced unexercised options dropping from $283M to $131M. Most contracts contain termination-for-convenience provisions (10-K risk factors). The government explicitly structured the exit door.

FedRAMP protects the class of authorized providers, not BBAI specifically. Palantir is already FedRAMP authorized. AWS GovCloud and Azure Government are expanding AI services. Booz Allen, Leidos, and SAIC all have cleared workforces and will achieve GenAI-specific authorization within 12-18 months.

E = 3 is the ceiling. The IL6/Top Secret tier is genuinely rare today, buying time. But the clock is ticking.

U — Ecosystem Breadth: 2/5

Eight to ten identifiable use cases across national security and travel/trade — but they don't compound. Ask Sage (GenAI), ConductorOS (edge AI), ProModel (simulation), VeriScan (biometrics), and CargoSphere (cargo screening) are architecturally siloed acquisitions. No evidence of cross-module data flows in any filing or transcript.

Revenue concentration confirms narrow breadth: 90% from a single customer type (U.S. government), top 5 customers account for 59% of revenue. Revenue recognition is 61% T&M (custom work per engagement), not platform subscription. Cross-selling Ask Sage into the existing customer base is aspirational — management acknowledged this in the Q4 2025 call without citing progress.

Switching cost scales linearly with products adopted, not superlinearly. The "ecosystem" exists on slide decks, not in architecture.

A — Distribution: 2/5

100,000 DoD users across 16,000 government teams is real distribution — inside a walled garden. Ask Sage is the distribution channel for frontier AI models within classified environments.

Outside that garden: nothing. Zero commercial API ecosystem. Zero developer integrations. Zero presence in commercial AI agent training data or routing infrastructure. The probability of a commercial AI agent encountering and routing through BBAI approaches zero.

Government procurement is slow, lumpy, and re-competed. No compounding flywheel (usage does not generate training data that creates agent preference). Model-agnostic architecture means BBAI is a passthrough, not a destination — when model providers achieve their own FedRAMP authorization, agents route directly.

M — Ecosystem Gravity: 2/5

Customer A's complete departure ($21.3M to $0) is the single most important data point. Migration cost was not prohibitive for a top customer. The company explicitly disclaims patent dependence: "not dependent on any particular patent or application for the operation of our business" (10-K FY2025).

No network effects — each deployment is independent. Customer relationship intangibles tell the story: Ask Sage relationships valued at $12.8M with a 3-year life (rapid decay), Pangiam relationships at $21.7M with a 20-year life (legacy government stickiness). The growth product has weak gravity; the legacy product has moderate gravity.

$128M revenue in a $100B+ defense IT market produces zero pricing power and zero scale-based gravity. Goodwill "limited headroom" — the company's own DCF barely justifies what they paid for their acquisitions (10-K FY2025).

F — Ecosystem Friction: 3/5 (penalty)

Standard government/enterprise complexity. Multi-year procurement cycles, continuing resolutions, budget uncertainty. Integration services required for deployment — not self-serve. 61% T&M revenue model is inherently high-friction. Classified access requirements (clearances, facility access, SCIF) add overhead.

Ask Sage trends toward lower friction (100K users suggests some self-serve adoption), but the base business remains consultant-dependent.

Thermodynamic Summary

One thing prevents intelligence from flowing around BBAI: a FedRAMP certification on a GenAI wrapper.

The underlying technology is explicitly built on "cutting-edge third-party large language models" — the intelligence is rented, not owned. The orchestration layer is shallow (C = 2). There is no physical infrastructure — $1.6M in PP&E (E stops at 3). The products don't compound (U = 2). Commercial AI agents are completely unaware of BBAI (A = 2).

As model providers pursue their own FedRAMP authorizations and defense system integrators build competing certified platforms, the certification moat degrades monotonically. Kill cycle estimate: 2-3 years before competitors match certification plus technology. When E degrades from 3 to 2, V drops to 1.66 — still Collapsed, trajectory down.

The balance sheet confirms the thermodynamic diagnosis: $155.6M in goodwill impairments across two years, $637M in equity dilution (shares nearly doubled from 250M to 476M), adjusted EBITDA of -$35.1M (from -$2.4M prior year), backlog down 41%, revenue down 19.3%, and "limited headroom" on remaining goodwill. The market at $1.6B on $128M revenue (12.5x, unprofitable) is pricing the defense AI narrative, not the balance sheet.

This is UiPath with a security clearance: a wrapper on external intelligence, temporarily protected by a certification barrier that degrades as competitors replicate it.

Conviction and Basket Verdict

V = 1.88          COLLAPSED
κ = (V − 3.0)⁺ = 0.00
w_BBAI = 0.00     EXCLUDED from basket

δ = V − V_market = 1.88 − 3.0 = -1.12  (below investable threshold)
IR = -2.453        (regime is discriminating; market agrees with V)

Durable revenue:  ≈15%  (FedRAMP GenAI first-mover, 12-24mo window)
Exposed revenue:  ≈85%  (T&M analytics, biometrics, simulation, commercial)

BBAI fails on structural grounds. κ = 0 produces zero weight regardless of basket normalization. The regime analysis provides no refuge — ρ_intra = 0.41 means the market is discriminating, not selling indiscriminately, and it's singling BBAI out for -42.8% while the defense sector rises. IR = -2.45 is not noise from a correlated selloff; it's the market pricing structural weakness in real time.

No position. No watchlist. Move on.

Evidence Table

LR < 1 = evidence against survival. LR > 1 = evidence for survival.

Sources

All dimension scores derived from primary SEC filings and earnings transcripts:

1. BBAI 10-K filed 2026-03-02 (FY2025 annual report)
2. BBAI Q4 2025 earnings transcript dated 2026-03-02
3. BBAI Q2 2025 earnings transcript dated 2025-08-11
4. Yahoo Finance market data as of 2026-04-08
5. Factor regression: BBAI vs SPY + ITA, 15-week window (69 trading days)