Status: Strategic positioning (founding reference) Last updated: 2026-06-24 Related: product-scope.md, architecture/build-vs-consume.md, risks/risk-register.md, vision.md
Why Provna over the alternatives. This is a strategic argument - naming competitors and their gaps is fine; the details live in product-scope.md and architecture/build-vs-consume.md.
Provna does not own a single pillar. It owns the intersection of four axes: vertical-FS x S2-compensation x S1-IFC-fusion x S4-signed-anchored-evidence. In a scan of 60+ candidates as of mid-2026, none closes this intersection - each deepens in one or two axes and leaves the rest architecturally empty. And all of them leave S2 empty, which confirms the moat with several independent negative proofs.
The white space has three fusion components: (a) IFC-aware compensation - the S1 preventive side and the transactional side are taint-blind in everyone else, and no one joins them; (b) a ready per-connector inverse library - the "library vs pattern" gap; (c) action-bound signed-anchored FS evidence - not logging.
| Competitor | Pillar(s) | Why it does NOT close the white space |
|---|---|---|
| MGAT (Microsoft, horizontal) | S1 partial + S3 + S4 | Horizontal breadth kills the horizontal dream, but S2 is a stub, IFC is host-dependent, audit is unsigned, and compliance is not EU-focused -> it validates the vertical-depth thesis. A substrate, not a rival. |
| MVAR (solo, Apache-2.0) | S1 + S4 | An S1+S4 concept-twin (the most mature S1 reference) but S2 is absent, S3 is single-axis, there is no anchor, the node is mutable, and it embeds no kid -> a blueprint, not a threat (patent caution). |
| Invariant (Snyk) | S1 | Treats "flow" as chronology, and the source admits its injection classifier is just a heuristic; the verdict enum is only BLOCK/LOG; the trace is unsigned -> a DSL-ergonomics reference plus a secondary marketing threat. |
| ACP (solo) | S3 + S4 | Architecturally excludes S1 and S2 (states there is no rollback of approved requests) -> it brings stateful-admission prior-art to S3 and confirms the moat choice; attenuation / Merkle / anchor are absent. |
| Verity | S2 | Exactly-once-forward only; no compensation, no dry-run, unsigned -> solves "do not double-execute," explicitly does not solve undo. |
| Rubrik (Agent Rewind) | S2 | Snapshot-restore, not per-action inverse - cannot rewind a sent email or a triggered payment; a perception threat with no mechanism. |
| Temporal ($5B) | S2 | Strong durable execution, but compensation is a manual pattern (not a library); no inverse, no dry-run, unsigned -> a substrate to consume. |
| SGNL (acquired by CrowdStrike) (~$740M) | S3 | Very strong per-action authz + CAEP + transaction tokens, but no S1/S2 and weak S4 -> it owns S3; Provna aligns and consumes here. |
The axes measure the two BUILD pillars: X is deterministic IFC (S1) strength, Y is reversible transactional action (S2) strength. [OPINION] - coordinates are reasoned estimates, not measured data.
quadrantChart
title Agentic Action Safety Landscape
x-axis Weak IFC --> Strong deterministic IFC S1
y-axis Prevention only forward --> Reversible transactional action S2
quadrant-1 Provna white space S1 x S2 fusion
quadrant-2 Compensation present IFC blind
quadrant-3 Saturated single axis
quadrant-4 IFC present compensation blind
Provna: [0.82, 0.82]
MGAT: [0.45, 0.40]
MVAR: [0.62, 0.18]
Invariant: [0.18, 0.20]
ACP: [0.34, 0.12]
Verity: [0.20, 0.40]
Temporal: [0.12, 0.45]
Rubrik: [0.10, 0.38]
FIDES CaMeL: [0.78, 0.15]
SGNL: [0.55, 0.08]
Reading: MVAR sits high on X (the most mature S1 reference) but low on Y because its S2 is absent; FIDES/CaMeL is the same - strong IFC, compensation-blind. SGNL and ACP cluster in the low-S1/low-S2 band because this map does not make S3 an axis. Several players hold each axis separately; no one holds the cross-intersection in the top-right.
Thesis: defensible in substance (S1+S2), not in position (S3+S4) - a ~12-24 month clock [OPINION].
- Defensible (conditional). The S2 compensation library is the hardest moat candidate because compensation is semantic (per-connector, API-version-bound, needs observe-probe + a round-trip harness). No horizontal / durable-execution / security vendor builds it (they treat it as a "developer problem"). But moat-hood is explicitly conditional: if compensation content genuinely requires multi-year accumulation, then "buy < build" holds and the flywheel turns; if not, the flywheel weakens. This is the single most critical assumption and is sold here as an assumption, not a certainty - it is the first item to validate with design partners (see risks/risk-register.md). The S1 fusion (P/Q isolation) is an architectural differentiator absent in all three torn-down competitors.
- Not defensible: S3 is saturated (CrowdStrike / Cisco / Microsoft); S4 mechanism is commodity (OTel / transparency log / RFC3161). Provna's difference there is assembly + FS mapping, not a breakthrough -> consume it.
- First-mover advantage is in content accumulation, not architecture (the saga mechanism is commodity).
| Absorber | How | Counter-move |
|---|---|---|
| Snyk (moves down) | S1 via Invariant; if it expands into S2/S4 | Nail the S2 catalog to vertical-FS connectors (NetSuite/Stripe/SWIFT/ledger); integrate Article 12/14 regulator evidence - domain depth a horizontal S1 firm cannot reach. |
| Temporal (moves up) | Adds a compensation library with a $5B war chest | Pool value where Temporal does not / cannot: IFC-aware compensation + signed/anchored regulator-grade compensation evidence + vertical-FS connector content. EU-FS forensic evidence + IFC fusion is not its category. |
| Rubrik (perception) | "Snapshot is enough" narrative | Counter-position: "a snapshot cannot rewind SaaS side effects"; demo the sent-email / triggered-payment case. |
| Microsoft (organic) | MGAT horizontal governance for free | Do not race horizontally; narrow the build to S2 + S1-runtime-IFC + Article 12/14; consume the ACS PDP as a substrate; lead with DORA/MiFID II depth. |
MGAT (Microsoft, horizontal, multi-workflow CI, multi-language) touched all four pillars horizontally and shallowly: S2 stub, IFC stateless/host-dependent ("host owns propagation," uninstrumented paths out of guarantee), unsigned audit, compliance not EU-focused. Two consequences: (a) Provna's horizontal positioning is dead - Microsoft is burying horizontal governance as free open source; (b) the same weaknesses validate Provna's vertical-depth thesis. MGAT is therefore not a rival but a PDP/audit-pattern substrate and a narrow-the-vertical signal. Not a generic "agent governance control plane" - a deep, thin layer for EU-FS back-office. This is consistent with the category framing in vision.md: Provna falls under the broad umbrella only on an analyst map; the sub-class it owns is narrow.
Three different "ACS/ACP" must be kept distinct in all Provna materials, or Provna gets mistaken for an "MGAT-ACS clone."
| Abbreviation | Full name | Effect on Provna |
|---|---|---|
| ACS = Agent Control Standard (businesswire, mid-2026) | an external "standard" | Zero corpus footprint -> standard adoption UNVERIFIED. |
| ACS = Agent Control Specification (Microsoft MGAT Rust PDP) | shipped, multi-language bindings | A PDP substrate candidate but no delegation/attenuation; brand-collision risk. |
| ACP = Agent Control Protocol (solo, Go) | stateful-admission prior-art | S3 prior-art; excludes S1/S2 -> patent / prior-art caution. |
[OPINION] In marketing material, always spell out the Standard vs Specification distinction. Also: the AuthZEN opportunity is real - MGAT does not implement AuthZEN, so Provna's AuthZEN 1.0 alignment is a genuine differentiator.
The first market is the EU-exposed financial-services back/middle office. The wedge: payment / supplier-payment approval, OR reconciliation-break correction email. The reasons: (R1) errors are irreversible and pre-budgeted - a SOX / four-eyes budget already exists; (R2) the most date-stamped regulation - EU AI Act Article 12/14 + DORA + MiFID II; (R3) live business pull - banks are deploying recon/AP/close agents but have no write layer; (R4) integrations are idempotency-native - payment rails / ERP already offer void/reversal; (R5) highest founder-fit (saga + compliance + eval). The others are strong #2s, not the beachhead: IT/cloud-ops (weak forcing function), healthcare RCM (hostile integration), insurance (slow).
Deadline realism [OPINION]. A specific regulatory date is not the only forcing function and Provna is pre-build, so it will not be a product that meets that single date. The real, continuous sales engine is DORA's ongoing operational-resilience obligations + the evidence demand that recurs every audit cycle + the permanent risk appetite around irreversible money movement. To a sophisticated buyer Provna does not say "we make the deadline"; it says "after the deadline passes you must still produce the same evidence at every audit, and we solve that permanent obligation."