# BHI Research Note RN-001 — Portability Is Not Exit Capacity

**Series:** BHI Research Notes — short, dated, reproducible research statements. Not marketing.
**Status:** **PUBLISHED v1.0.1** — published **22 August 2026** (went live ≈22:15 Kyiv).
v1.0's metadata initially stated 23 August per the publication instruction; the owner caught
the future-dated metadata the same night and ordered a visible correction: *v1.0.1 —
publication metadata corrected from 23 Aug to 22 Aug 2026; research content unchanged.*
Permanent URL:
blackholeindex.com/research/rn-001-portability-is-not-exit-capacity. This file is the source
text; the published page is content-equivalent (formatting and cross-references adapted:
the page adds "forthcoming" to the RN-002 pointer and a Cite-as/Revisions block; no factual
content differs).
**Author:** Ivan Savich (ORCID 0009-0003-9542-7022) · independent researcher
**Dataset/version:** BHI Q3-2026 release (pinned: `bhi-scores-q3-2026.csv`, 184 platforms) ·
Cloud Core frozen snapshot S1, SHA-256
`6565778b84b38a9420c41fec37219303548cf6b7fb52f88237c6677b949712fe`
(hash publicly fixed 22.08.2026 at blackholeindex.com/research-snapshots.txt; two independent
RFC 3161 timestamps of the digest, same day).

---

## Claim

The capability most often cited as the answer to cloud lock-in — "your data is exportable" — does
not measure the ability to leave. Exit capacity is the reconstruction of a working system
elsewhere within a bounded time; export is one input to it. The two come apart in provider
documentation itself, and an instrument that scores "portability" on export alone will
systematically overstate escape.

## The distinction, operationalised

The BHI Cloud Core measurement fixed an exit criterion before scoring: a completed exit is one in
which every workload runs elsewhere, the data **and configuration state** required to operate
them has been reconstructed there, no business function has been withdrawn, and the transition
completed within twelve months using existing engineering staff plus contracted assistance not
exceeding 25% of that headcount. Three
consequences were fixed with it: an export capability is not an exit; twelve months is the
horizon; partial exit is not exit.

Under that criterion, the portability parameter (`x`) turned on facts that raw-export scoring
never sees. Three of them, from the frozen determination, all first-party-documented:

1. **A portable configuration language is not a portable resource graph.** Two of four measured
   providers use a cross-platform format for their first-party infrastructure-as-code surface;
   two use proprietary formats. Six years of operational encoding either transfers as working
   material or is re-authored on exit. But the caveat travels both ways: a resource block written
   for one platform does not deploy on another merely because the language is shared — the win is
   in tooling, state handling and skills, not in the definitions themselves.
2. **Outbound transfer tooling is asymmetric.** One provider of four documents first-party
   tooling that writes into competitors' storage; the other three document inbound or intra-cloud
   paths only — verified symmetrically on all four, not inferred from the one.
3. **Derived and learned state is not in the export.** Three of the four measured providers
   document models trained on the customer's own operational history, with no export path
   documented anywhere; the fourth documents exactly two autoscaling modes, neither learned. What
   the platform has learned about the workload leaves with nothing.

## Why this matters for remedies

A remedy that improves export (formats, registers of exportable structures, fee removal) touches
the first input of exit capacity and none of the rest. The frozen prediction register
accompanying the Cloud Core determination pre-registers the consequence: an egress-charges remedy
that changes nothing else moves the structural index by exactly zero. Whether that is the right
property of an instrument is a fair question — it is pre-registered precisely so a regulator's
post-review facts can grade it.

## Limitations

All underlying scores are by a single evaluator; inter-rater reliability is not established (the
study is drawn, frozen and paused). The Cloud Core determination reached an identification
boundary: three of eleven parameters are Not Scorable from provider evidence at this boundary,
so no composite index exists for the measured providers (see RN-002). Findings are statements
about documentation read in a declared window against a stipulated Reference Customer, not about
any real organisation.

## Reproducibility

The full determination — boundary, evidence rules, four ledgers with quotes/URLs/retrieval dates,
construct audit, comparative profile — is frozen as snapshot S1; its SHA-256 above was publicly
fixed on the freeze date and independently timestamped (RFC 3161) the same day. Whatever is later
released under that hash is the corpus that existed on 22.08.2026. Release-level data:
blackholeindex.com/bhi-scores-q3-2026.csv (pinned) — scores recompute from the 11 published raw
parameters per platform.