Everything you need to understand our deterministic set-theoretic analysis engine, data handling pipeline, commercial policies, and cross-platform conflict detection.
The 6 collision classifications, Venn intersection models, and how to resolve audience contamination.
Read Guide →Ephemeral anonymous memory guarantees, registered durable storage boundaries, and download samples.
View Architecture →Detailed capability matrix against native vendor dashboards, manual spreadsheets, and proxy metrics.
Compare Capabilities →Answers on offline security, supported targeting operators, multi-project workflows, and pricing.
View FAQ →Six classifications, what each means, and what to inspect next.
Unknown stays unknown. A rule the parser cannot normalize is flagged for manual review, never converted into safe.
Targeting rules across both platforms contain conditions that deterministically overlap (e.g. matching equality on device, geo, or plan).
Next: Inspect shared dimensions in the detail panel to evaluate interaction effects.
Targeting information indicates intersection is probable, but incomplete attribute distributions prevent deterministic confirmation.
Next: Check user attribute distributions in production logs.
Targeting conditions contain mutually exclusive constraints (e.g. device == "mobile" vs device == "desktop") within the evaluated rules.
Note: Not a guarantee of zero collision outside these evaluated rules.
Rules evaluate disparate dimension sets with zero shared attributes and no mutual constraints.
Next: Confirm whether the two squads target intersecting user bases.
Essential targeting metadata or clause values were absent from the uploaded configuration.
Next: Re-export the complete flag or audience definition JSON.
Rule uses a custom clause or operator the parser cannot yet normalize. Unknown is never converted to safe.
Next: Inspect the custom clause manually.
Architected for strict data minimization. Code-verified boundaries, zero marketing inflation.
We make no absolute claims of "never logged anywhere across the globe". We describe our actual server architecture and verified log inspections truthfully.
Strictly ephemeral in-memory processing. Configuration payloads are evaluated in volatile process memory and discarded immediately upon request completion. Zero database persistence for anonymous diagnostics; no database writes without registration.
Customers explicitly ask FlagConflict to persist workspace projects, configuration baselines, and audit runs to enable change tracking, history reviews, and CSV exports.
Founding Plan history window caps at 24 months. 30-day and 7-day warnings notify of scheduled deletions, purging the oldest 6-month block down to ~18 months. Active projects and current baselines are never deleted.
Targeting payload contents, credentials, and sensitive configuration values are not intentionally written to application logs. Limited operational metadata may be processed where necessary for security and diagnostics.
FlagConflict operates purely offline on exported JSON. It maintains zero write permissions or webhook channels into your LaunchDarkly or Optimizely accounts.
Strict request payload ceiling prevents memory exhaustion and oversized uploads.
Users may delete their account and associated project data through available self-service controls, subject to active subscription safeguards and any limited information retained for security, billing, fraud prevention, audit or legal obligations.
FlagConflict is not an experimentation platform. It adds a cross-platform collision analysis layer to the tools you already run.
| Capability / Dimension | LaunchDarkly | Optimizely | Statsig | FlagConflict Sentinel |
|---|---|---|---|---|
| Feature Flag Evaluation & Delivery | Core Capability | Core Capability | Core Capability | Not Offered (Complements) |
| A/B Testing & Variant Assignment | Core Capability | Core Capability | Core Capability | Not Offered (Complements) |
| Experiment Analytics & Metrics | Core Capability | Core Capability | Core Capability | Not Offered (Complements) |
| Single-Platform Audience Targeting | Core Capability | Core Capability | Core Capability | Not Applicable |
| Cross-Platform Collision Detection | Not identified in reviewed public documentation | Not identified in reviewed public documentation | Not identified in reviewed public documentation | Core Purpose |
| LD + Optimizely Logic Normalization | Not identified in reviewed public documentation | Not identified in reviewed public documentation | Not identified in reviewed public documentation | Supported (Deterministic) |
| Epistemic Overlap Explanations | Platform-specific only | Platform-specific only | Platform-specific only | 6-State Matrix |
| Offline Read-Only Analysis (Zero SDKs) | SDK Required | SDK Required | SDK Required | Zero SDKs / JSON in |
| Replaces Existing Platform? | Primary Platform | Primary Platform | Primary Platform | No · Additive Layer |
core/competitor_evidence.json.
FlagConflict offers three clearly bounded entitlement tiers designed to prove value before commitment.
Full pairwise collision matrix, detailed WHY explanation, and sample audits. Zero account or payment required.
1 saved project, preserved configuration baseline, last 3 audit runs, and change tracking (NEW / RESOLVED / PERSISTENT).
Founding Plan
Up to 10 active projects, recurring change tracking, CSV/JSON history exports, and 24-month rolling history window.
Capabilities, boundaries and security. Still unclear on something? Tell us what you need.
Two rules on different platforms whose conditions can both be satisfied by the same user. A US enterprise customer on mobile matching a LaunchDarkly flag and an Optimizely experiment at the same time is a collision.
No. It means two treatments could have reached the same people, which is worth knowing before you interpret a result. Whether that changed the outcome is a judgement your team makes with production data.
LaunchDarkly flags and Optimizely audiences and experiments, via JSON export. Other platforms are not supported today.
No. Analysis runs on offline JSON you paste or upload. There are no tokens to issue and no permissions to grant.
It is parsed and compared in memory, then discarded. It is not written to a database and its values are not written to application logs.
No. FlagConflict is strictly read-only and has no write path back to LaunchDarkly or Optimizely.
You can, but the value is limited. The engine exists to compare across platforms; within a single platform your own tool already has that view.
That one rule used a construct the parser cannot yet normalize. It is reported as needing manual inspection. It is never silently counted as clean.
Statsig is the most requested. Nothing is committed — the order follows what teams actually ask for.
The anonymous diagnostic is completely free ($0) with no account required. A Registered Free account ($0) allows you to save 1 project, preserve your configuration baseline, and track changes across the last 3 audit runs. The Founding Plan ($29 / mo ($29/month)nth) supports up to 10 active projects, full audit history, recurring change tracking, CSV/JSON exports, and 24-month rolling history.
FlagConflict Sentinel is operated under the controlled public experiment umbrella of ProbeHaven.
Direct technical and billing support is provided directly by the product research lead.
We aim to respond to support inquiries within 1–2 business days. For urgent billing reviews regarding Paddle subscriptions, please include your workspace email.
Summary of key customer commitments, boundaries, and subscription management terms.
All subscriptions and payments are securely processed by Paddle.com Market Limited as Merchant of Record. FlagConflict Sentinel never receives or stores payment card credentials.
FlagConflict does not provide an additional discretionary money-back guarantee. Purchases, cancellations and refunds are processed by Paddle, our Merchant of Record, under Paddle's applicable Buyer Terms and Refund Policy. Paddle may provide refunds where required by law or at its discretion. Nothing in this policy limits mandatory consumer rights.