docs(prds): add Project Foglight BRD for HVE Core integration#2075
docs(prds): add Project Foglight BRD for HVE Core integration#2075namitts82 wants to merge 7 commits into
Conversation
- Project Foglight BRD for HVE Core integration as experimental collection - Defines MVP pack with Foglight Orchestrator, Confidence Dashboard, Decision Ledger - Addresses key gaps for TPMs/Dev leads delivering intelligent systems - Covers inputs, outputs, RAI posture, architecture fit, pilot use cases - Ready for community review per discussion microsoft#1387 📋 Generated by Copilot
docs(prds): add Project Foglight BRD for HVE Core integration
|
David (david.ratcliffe@microsoft.com) |
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #2075 +/- ##
==========================================
- Coverage 81.32% 81.31% -0.01%
==========================================
Files 130 130
Lines 19142 19142
Branches 12 12
==========================================
- Hits 15567 15566 -1
- Misses 3572 3573 +1
Partials 3 3
Flags with carried forward coverage won't be shown. Click here to find out more. 🚀 New features to boost your workflow:
|
|
@namitts82 please read the following Contributor License Agreement(CLA). If you agree with the CLA, please reply with the following information.
Contributor License AgreementContribution License AgreementThis Contribution License Agreement (“Agreement”) is agreed to by the party signing below (“You”),
|
|
@namitts82 - this is really, really well formed. I am going to have to digest this a bit as is does exactly what you intend it to do ... wrapping multiple workflows into a coherent and cohesive agentic experience. The one place I have a hang up on right now, is that per NIST RAI specs, this gets close to the definition of "over functioning" for users. My secondary concern is that while I 100% KNOW this will work, and work really well, it does have risk from burning tokens from poor/weak input framing by the user. I'm going to take this one to the core maintainer team and let them think it through. Also, a massive thanks for building a BRD first as a proposal! I really need to update the contributing docs to suggest this as a proposal path for new AI capabilities ... strong work! |
Thanks Bill, really appreciate the close read and both callouts. On "over-functioning": that's a fair thing to pressure-test. A few of the design choices were aimed at keeping the human in the loop rather than deciding for them: confidence is always tied to a named decision and the orchestrator coaches toward a tool rather than making the call (Stage 6 is co-author, not decide); we deliberately avoid composite scores; and every artifact has to carry its evidence, remaining uncertainty, and a refresh trigger. If it helps, I'm happy to add a short RAI guardrail note making the "surface evidence, don't auto-resolve" stance explicit against NIST AI RMF language. On token burn: also fair. The pilot pack is intentionally the smallest useful one, context gathering is bounded, and the orchestrator is meant to flag insufficient signal and offer to defer rather than force an artifact. I could make that an explicit input-sufficiency gate before any heavy generation, and add token/cost-per-session as a Phase 1 guardrail to actually measure it. Glad to fold either of these in before it goes to the maintainer team. |
- Add Phase 1 success metrics and milestone gate (per jkim323) - Add explicit RAI over-functioning guardrail citing NIST AI RMF (per WilliamBerryiii) - Add input-sufficiency gate to orchestrator to address token-burn concern - Add cost guardrail as Phase 1 pilot metric
|
@namitts82 - sorry for the delay on this ... been dealing with some security hardening and other workflows that are taking priority. I've not forgotten about this ... just still mulling over the boundaries and what lines we want to hold in HVE Core. I absolutely recognize what's possible, and what's practical, and this fits cleanly into both buckets ... I'm still trying to figure out with the other maintainers if we should ... thanks for hanging tight for a bit while we process the ramifications and impact, assess how this would fundamentally alter the projects forward trajectory and the impact on existing workflows ... like do we need a composite workflow builder that can take our primitives in and build abstractions over it like this does? Should we raise the level of abstraction fully towards workflows like this. All hard decisions to reflect on as the industry grapples with the economics of all this rapid change. |
Adds \docs/prds/foglight.md\ — the Project Foglight Business Requirements Document — for community review before any implementation work is proposed.
Description
Adds the Project Foglight BRD addressing the structural gap that standard TPM tooling has when applied to intelligent systems (RAG copilots, agents, model-driven workflows).
MVP pilot pack proposed:
Type of Change
Related Issue
Related to discussion #1387
Checklist