Content Engine runs every version through OPDP, ISI, fair-balance, and brand-voice rules before it reaches your reviewer. Flags caught upstream, scored, surfaced with auto-rewrite suggestions. The enforcement layer of the OVYN architecture.
Architecture · Layer 02 · Enforcement
This is Content Engine, the rule engine. Pulls from Intelligence for citation context, hands rejected versions back to Activation for auto-rewrite.
Three axes run on every execution before it reaches MLR. Compliance catches RULE violations. Self-check catches the tonal drift regex misses. Brand-fit catches IDENTITY drift, executions that respect every rule but still read as generic. Each axis answers a different question and surfaces independently in the audit drawer.
Brand-fit is independent from compliance. An execution can pass compliance but fail brand-fit (off-brand voice), or pass brand-fit but fail compliance (recognizably on-brand but a forbidden phrase slipped in). Hard floor at brand-fit ≥ 60 (the BRAND_FIT_THRESHOLD constant). Audit-drawer tiers: ≥80 strong, 60-79 mid, <60 off-brand.
Axis 1
Compliance rules
Deterministic phrase scan against the brand's forbidden list plus OPDP, fair-balance, ISI, off-label, outcome-guarantee, and regulatory boilerplate. Regex plus per-indication thresholds. Hard-fails block the execution from leaving Activation.
Axis 2
Model self-check
The model re-reads its own output against the same rule set and flags borderline framings with a structured rationale. Catches tonal drift the regex misses. Three-axis provenance (rules · self-check · brand-fit) shown per execution.
Axis 3
Brand-fit identity scoring
Independent score 0-100 measuring how closely the execution matches brand essence and corpus voice. Hard floor at 60. Below that = off-brand. Surfaced in the audit drawer with a one-line rationale: which signature tell the execution reached for and where it drifted.
An execution can be compliance-clean but brand-fit-low (generic-but-legal), or brand-fit-strong and compliance-flagged (recognizably on-brand but a rule slipped in). Both signals surface independently. Brand-fit lives on /pillars#architecture · A3.
Open queue
4 flags awaiting your review
2 high2 med0 low
DEM-2026-04-216highOPDP-FB-01 · Fair balance
Compound DT-204 cost & access · static card
Benefit claim on 'most pay $10 or less' appears 4 lines before ISI block. Recommend tightening to within 2 lines or moving ISI above.
Auto-rewrite available, moves co-pay claim adjacent to ISI block.
DEM-2026-04-211medOPDP-COMP-05 · Comparative
Caregiver carousel frame 5 · 'beyond steroids'
Phrase 'beyond steroids' implies superiority without head-to-head citation. Soften to 'an alternative to steroid-only management.'
Auto-rewrite available, replaces with non-comparative phrasing.
OPDP guidance, fair-balance, ISI requirements, comparative-claim rules, pediatric language, encoded once, enforced on every output. Hit count + failure rate per rule lets you tune.
Rule ID
Category
Standing instruction
Checks
Fails / mo
OPDP-FB-01
Fair balance
Risk information must accompany every benefit claim within 50% proximity.
412
7
OPDP-DT-02
Disease/treatment
Disease-state content cannot pivot to branded claim without ISI.
