Locked posture (July 2026)
Reconciled open / proprietary boundary — same story in thesis, README, Lab, and this room:
| Public (trust signal) | Private (product / moat) |
|---|---|
| Methodology descriptions | Customer / partner artifacts submitted for evaluation |
| Detector definitions and published thresholds | Engagement workflows and partner-run orchestration |
| Reproducible evidence (Lab metrics, failure cases, limited demos) | Branded partner reports and continuous-scoring integrations |
| Apache-2.0 licensed benchmark / checker (public claim surface when released) | Accumulated comparative evaluation data across engagements |
Product is the evaluation engagement, not secrecy of the public claim surface. Patenting a testing method is a poor fit at this stage; the compounding moat is reputation + private comparative data.
Repo status ≠ IP posture. haga-core may remain a private GitHub repo until a deliberate public release; the license and claim surface still treat methodology, detector definitions, and the harness as the trust signal — not trade-secret tooling.
What is not the posture
- Not “trade secret over patent for the entire harness” (superseded by this lock)
- Not “methodology public, thresholds secret” — published thresholds are part of the trust signal
- Not open access to partner submissions, private run logs, or cross-customer comparative datasets
Moat that compounds with this posture
From the thesis competitive section:
- Accumulated comparative evaluation data — private growing dataset of how policies / world models behave under adversarial conditions
- Open claim surface, private comparative data — Apache methodology + harness builds trust; the moat is engagement data
- Reputation as neutral third party — compounds with track record
- Integration switching costs (Phase 5) — re-baselining cost once continuous scoring is in a release pipeline
- Dual-artifact coverage — policy-only and world-model-only competitors must build the other pillar from scratch
Founder hygiene (Gate A)
Before external investor invites or Atlas:
- Sole-founder equity locked — 90% founder / 10% option pool (within 85–90 / 10–15 band)
- Vesting locked — 4yr / 1yr cliff; start = incorporation date
- Prior invention schedule complete — no carve-outs — schedule
- IP assignment / side letter drafted — side letter
- Side letter executed; signed PDFs stored outside git — 2026-07-17
Full pack: Founder equity & IP · Status: Ready (Gate A met).
Entity & assignment vehicle
- Target entity: Delaware C-Corp via Stripe Atlas when budget or a committed check appears
- Not: Pakistan registration as the fundraising vehicle
- Pre-Atlas: assign into the company-to-be via Gate A side letter; post-Atlas: confirmatory assignments into the Delaware entity
- External dataroom invites stay locked until incorporation gates unlock — Investor sharing
Formation: Formation checklist.
Patents later (optional)
Revisit patent strategy only if a narrow, novel apparatus or detector claim emerges that is (a) hard to design around and (b) worth disclosure cost. Default remains: public methodology + Apache harness for trust; private comparative data for moat.