BHI 3.1·Bulletin
Validation: preliminary
Today's bulletin →
Highest B
Full paper →Rankings →A4 · single-page brief · v3.1 ·
B

Black Hole Index — Quarterly Brief

Structural lock-in · 184 platforms · 18 sectors
Release
V 3.1

The release scores 184 platforms on the eleven-parameter CES rubric with V3.1 introducing the org-depth multiplier, momentum amplifier, and a formalised ε = 0.1 floor. The headline finding: infrastructure platforms (EDA, TSMC, ASML) exhibit lock-in an order of magnitude higher than attention-based platforms (Netflix, Dogecoin), despite the latter commanding larger user bases or market caps.

Universe
184
Platforms · 18 sectors
Mean B (universe)
4.41
Median ≈ 3.6 · skewed by infrastructure tail
Above B = 1
172 / 184
Only 12 platforms below the capture = escape threshold

Highest-B platforms

PlatformSectorB (V3.1)ZoneStructural driver
WeChatSocial17.03Black HoleSuper-app: identity, payments, government — all through one system
Synopsys/Cadence (EDA)Critical Infra13.50Black HoleNo chip design without EDA — zero substitutability (s=1)
TSMCBig Tech12.28Black HoleNo advanced chips without TSMC — monopoly foundry
PalantirSaaS9.90Black HoleOrg-depth 10/10 — removal requires structural reorganisation
ASMLBig Tech9.62Black HoleSole EUV lithography supplier globally

The structural read

Infrastructure compounds; attention decays.

The top 5 are all infrastructure — not consumer apps. Netflix (325M subs, B = 1.00) proves massive scale with near-zero switching costs produces fragile revenue.

AI lock-in is real but stratified.

Copilot (B = 5.23) outscores ChatGPT (B = 2.25) by 2.3× despite lower user preference — organisational embedding, not quality, drives lock-in. The lowest-scoring named assistant is Perplexity at B = 0.95; Claude sits at 1.66.

Crypto: market cap ≠ lock-in.

Dogecoin ($15B, B = 0.17) — near-zero structural lock-in. Solana (B = 4.83) — developer ecosystem creates real capture. Bitcoin (B = 1.82) — network effects without deep process integration.

Association, not validation.

Spearman ρ = 0.79 (p < 0.001, n = 14) on an author-selected, concurrent cohort of retention-type metrics, as published. The paper corrigendum (rows 11–16) records that four of the fourteen values are not measured annual retention rates and that the association does not distinguish the formula from its escape half. Single-evaluator scoring — inter-rater reliability study pending. No prospective external-criterion validation has yet been completed; that stage (V2) is pending by design.