This public roadmap is a communication surface, not the full internal
.gpd execution ledger.
Its purpose is to show where the live repo sits in the scientific journey and what is still open without pretending the lane has already closed green.
| Phase group | Current public reading | Status |
|---|---|---|
| `08` | Comparator expansion vs AV1, VVC, and learned baselines: all three killed the universal-video-codec claim. | complete — killed |
| `09` through `09.2` | Broad and semi-narrow video surfaces: bounded signal, no decisive wedge. | complete — bounded |
| `09.3` through `09.3.2` | Narrow VIRAT facility-crossing work: real portal-local signal, weak consumer stability; retired. | complete — retired |
| `09.4` | Repo public-surface hardening, docs/proofs routing, governance spine. | complete |
| `09.4.1` through `09.4.1.1.1` | Ground-state wedge audit and pod-backed live archive-query benchmark on VIRAT: economics real, specificity not yet commercial. | complete — bounded_signal_only |
| `09.4.1.1.2` | Fair-baseline archive-query falsification: raw struct+zlib strictly dominates on storage AND query latency with identical semantics. Archive-query wedge as ZPE-specific is falsified. | complete — killed |
| `09.4.1.1.2.1` | Three parallel bounded wedge experiments: Candidate A (ROI sidecar) killed; Candidate B (video-LLM object memory) defended on cross-writer hash stability; Candidate C (state-layer enrichment) defended-with-caveat. | complete — B defended |
| `v0.1.0` | Ship the Candidate B perception-receipt wedge as the public API. Full transparency bundle ships alongside. | shipping |
| Next | Harden the perception-receipt surface against larger / messier benchmarks; identify 1-2 buyer partners; extend the receipt envelope only where evidence demands. | always-in-beta |
| Track | Current next move |
|---|---|
| Public repo | Keep docs aligned to shipped code and current transparency artifacts; update on every release. |
| Wedge hardening | Extend the cross-writer hash test to a larger, messier VideoQA surface (LongVideoBench / NExT-QA spatial subsets); identify 1-2 buyer partners (C2PA, video-LLM infra, regulated chain-of-custody). |
| Cadence | Ship minor releases when there is real new utility or a real new verdict to record. No milestone theatre. |
This roadmap does not mean:
- every prior research direction will be revived; retired or killed surfaces stay retired or killed
- the Compass-8 primitive-native substrate is an active product goal for this codec (it is historical research context; the product is the perception receipt)
- "always-in-beta" is an excuse to under-deliver; it means continuous improvement on a live, working package