Future Research Bets (Tier B) — Index

Status: Planned

This document’s own status header marks it as planned or aspirational — what it describes may not be implemented yet. See the roadmap for what has actually shipped.

Document ID: FUT-INDEX-001 File Path: docs/18-roadmap/future/index.md Version: 1.4.0 Status: Exploratory Owner: Product Team Last Updated: 2026-07-05

---

1. Purpose

This folder holds one exploration doc per Tier B research bet — the “beyond-1.0” directions from future.md §2, turned into requirements-level detail by PRD-002 §6.

Each doc states the problem, what real code it would build on, a non-committal design sketch, the invariants it must preserve, the risks, and — most importantly — the graduation gate: the evidence required before the bet becomes an ADR and takes a slot in a concrete release.

These are exploratory and not committed. They exist to make the bets plannable and prunable, not to promise them.

---

2. Register

Doc	Direction	Origin	Status
FUT-001	Autonomous Multi-Agent Systems	future §2.1	Exploratory — prototype slice for direction (b) in examples/workflows/research-team.yaml + wovyr-server, pre-ADR
FUT-002	Self-Optimizing Platform	future §2.2	Exploratory
FUT-003	Advanced Memory	future §2.3	Exploratory
FUT-004	Execution Frontiers	future §2.4	Exploratory
FUT-005	Ecosystem & Interoperability	future §2.5	Exploratory
FUT-006	Trust & Evaluation	future §2.6	Exploratory — prototype spike in crates/wovyr-eval, pointed at FUT-001’s workflow via a compare module, and at a real mistralrs model (one run so far: a tie), pre-ADR

---

3. Dependency Ordering

The bets are not independent. FUT-006 Trust & Evaluation is upstream of most of the others — FUT-001, FUT-002, and FUT-003 all state graduation gates that require a working evaluation harness to substantiate (“outperforms a single agent,” “measured improvement / no regression,” “retrieval-quality change”). So the natural first bet to graduate is FUT-006.

FUT-006 (evaluation harness)
   ├─ enables → FUT-001 (multi-agent: "outperforms?")
   ├─ enables → FUT-002 (self-optimizing: "improved, no regression?")
   └─ enables → FUT-003 (advanced memory: "better retrieval?")
FUT-004 (execution frontiers)  — gated by the sandbox-escape battery, independent
FUT-005 (ecosystem/interop)    — federation gated by a cross-org threat model, independent

---

4. Graduation Process

Shared with future.md §3:

future (idea) → PRD requirements → ADR (decision) → release roadmap (v1.x/v2) → docs + build

These docs are the requirements elaboration. A bet becomes committed only via an ADR and a concrete release slot; bets that never graduate are pruned, not carried indefinitely.

---

5. Related Documents

• 18-roadmap/future.md — the research-bet catalogue (RM-005)
• 01-product/prd-future.md — the requirements PRD (PRD-002)
• 17-adr/index.md — where graduation decisions land

---

6. Revision History

Version	Date	Description
1.4.0	2026-07-05	FUT-006’s wovyr-eval gained a real mistralrs model run of the comparison harness — one run so far, a tie — updated FUT-006’s register row; see B6-trust-evaluation.md §8.2
1.3.0	2026-07-05	FUT-006’s wovyr-eval gained a compare module pointed at FUT-001’s research-team.yaml — updated FUT-006’s register row; see B6-trust-evaluation.md §8.1 for what it proves and doesn’t
1.2.0	2026-07-05	FUT-001 gained a code prototype for direction (b) (examples/workflows/research-team.yaml + wovyr-server) — updated its register row; see B1-multi-agent-systems.md §8 for what it proves and doesn’t
1.1.0	2026-07-05	FUT-006 gained a code prototype (crates/wovyr-eval) — updated its register row; see B6-trust-evaluation.md §8 for what it proves and doesn’t
1.0.0	2026-07-05	Initial Tier B research-bet index (FUT-001…FUT-006) with dependency ordering