Case studies
The framework, walked.
Case studies show the framework working in real situations. Two kinds, both useful for different reasons.
- External cases walk publicly-reported failures through the framework and ask: would the framework have caught this? Teaching material, citing public reporting.
- Internal cases are real audits of products in the Startvest portfolio. They show the framework operating as intended and, occasionally, what happens when reality finds a gap in the framework itself.
External — public reporting · 2026-04-25
Delve, walked through the framework
First public case study under v1.0. Walks the publicly-reported allegations against Delve, an AI compliance startup separated from Y Combinator in April 2026, through the five failure modes and the six-row vendor scorecard. Score: 0 / 6.
Read the case study →
Internal — portfolio audit · 2026-04-28
FieldLedger, the first portfolio audit
First internal portfolio audit. 14 PASS / 5 PARTIAL or NEEDS UPDATE / 1 DEFERRED / 0 FAIL across 20 dimensions. Surfaced two false negatives in the framework's own rule set, revised in base manifest v1.3.0 and v1.4.0.
Read the case study →
Internal — portfolio audit · 2026-04-28
ClarityLift, the documentation-drift audit
Second internal portfolio audit. 13 PASS / 6 PARTIAL or NEEDS UPDATE / 1 BUYER-DRIVEN / 0 FAIL across 22 dimensions. Two of three failures were claims in INTEGRITY.md that referenced code never actually shipped — three days after the entry was written. Drove HIGH-SV-INTEGRITY-MD-CLAIMS-VERIFIABLE in base manifest v1.5.0.
Read the case study →
Internal — portfolio audit · 2026-04-29
IdeaLift, retroactive TIF on a non-compliance product
Third internal portfolio audit. The reference for what TIF looks like when applied to a product not designed against it. Most dimensions are honestly N/A. Drove v1.6.0 (monorepo glob widening) and v1.7.0 (CLAIMS-VERIFIABLE extended to scan Outstanding Risks for reverse drift).
Read the case study →
Internal — portfolio audit · 2026-04-29
Hireposture, the candidate-data calibration
Fourth internal portfolio audit. 16 PASS / 3 PARTIAL or NEEDS UPDATE / 1 OUT-OF-SEGMENT / 0 FAIL across 20 dimensions. Surfaced the input-data-provenance vs output-attestation-provenance distinction. Drove v1.8.0 — accepting snake_case marker variants like signed_by_user_id and customer_attested.
Read the case study →
Internal — portfolio audit · 2026-04-29
marketing-agent, the constraint-file pattern
Sixth internal portfolio audit. Internal Startvest tooling defended against marketing-automation failure modes by an explicit 29-item constraint file with kill-switch discipline. A different shape than the framework's code-pattern CI rules — being watched for in future audits, not yet codified.
Read the case study →
What comes next
The first portfolio audit cycle ran across six product types over two days (2026-04-28 / 2026-04-29). PRAPI is the seventh and is scheduled for v0.1 ship — its codebase doesn't exist yet. Quarterly cross-product re-audits begin ~July 2026. See the audit log for the v1.2.0 → v1.10.0 base-manifest revisions and which audit triggered each.
External case suggestions welcome at integrity@startvest.ai. We work from public reporting only. See the case-study template for the expected structure.