From 9fe4e497a174b80e21b47c169e08c730dbf09086 Mon Sep 17 00:00:00 2001 From: Daniele21 Date: Thu, 3 Sep 2026 17:42:40 +0200 Subject: [PATCH 1/3] docs: refresh value frontier after VALUE-02 integration --- docs/current-state.md | 43 ++++++++++++++++++++++++++++++------------- 1 file changed, 30 insertions(+), 13 deletions(-) diff --git a/docs/current-state.md b/docs/current-state.md index 7646e13f..3a2c3286 100644 --- a/docs/current-state.md +++ b/docs/current-state.md @@ -4,7 +4,7 @@ Status: active Document type: current-state Owner: repository Canonical scope: state.repository -Last reviewed: 2026-09-02 +Last reviewed: 2026-09-03 Short operational ledger only. Durable behavior belongs in architecture/ADR/design docs; active detail belongs in workstreams; Git history owns implementation history. @@ -16,10 +16,12 @@ Primary product question: > For this use case on this device, which available model + quantization + configuration gives me the best evidence-backed trade-off, and why? -Work now advances through vertical VALUE slices that unlock end-to-end outcomes. M1-M9 remain the capability/maturity coverage map rather than the delivery sequence. +Work advances through vertical VALUE slices that unlock end-to-end outcomes. M1-M9 remain the capability/maturity coverage map rather than the delivery sequence. ## Integrated baseline +Current integration head: `dev@0b9c376f2dbcbb167f8df8488374c1b0c5d0ace2`. + `dev` contains Overview; Find best setup `Goal -> Models -> Optimization -> Review -> Campaign -> Results`; Test a model; Live Run recovery; Runs/Run Detail; Compare; Library/Settings; Benchmark Detail; Run -> Samples -> Sample Evidence; same-case comparison; browser-local Light/Dark/System appearance; and built/package lifecycle support. Campaigns revalidate frozen plan digests, persist reconnectable lifecycle separately from immutable Runs and apply compatibility before explicit decision policy. Quality, Performance and Resources remain separate; missing, incompatible, unavailable and not-retained evidence remain explicit. @@ -30,43 +32,58 @@ VALUE-01 software/readiness lanes are integrated: - **VALUE-01B / #118 DONE** — retained evidence completeness/portability verifier; - **VALUE-01C / #119 DONE** — exact-head PRE_REAL-gated real-run entry point and retained run manifest. -Their branch validation passed Repository Validation, Browser Acceptance, Built Product/PRE_REAL and Repository Health before integration. The merge commits moved `dev`, so the representative run still requires a **fresh exact-head PRE_REAL PASS** for the final `dev` revision. +VALUE-02 software lanes are also integrated through PR #141: -## Current value frontier +- **VALUE-02A / #129 software DONE** — configured-target multi-model discovery and per-candidate Local LLM Server identity/telemetry attribution; +- **VALUE-02B / #130 software DONE** — bounded real multi-model Campaign/browser harness; +- **VALUE-02C / #131 software DONE** — retained multi-model decision evidence verifier. -Current executable slice: **VALUE-01 — Real single-model evidence loop** in [`workstreams/incremental-value-delivery.md`](workstreams/incremental-value-delivery.md). +This software integration does not satisfy the representative real-device acceptance owned by VALUE-01D / #120 and VALUE-02D / #132. + +## Current value frontier -Target loop: +The next convergence sequence is: ```text -real target/device -> discover one real model -> Test a model -> real inference --> Run Detail -> sample evidence -> retained .plab.zip +fresh exact-head PRE_REAL on current dev +-> VALUE-01D / #120 single-model REAL_ENVIRONMENT PASS +-> VALUE-02D / #132 multi-model REAL_ENVIRONMENT PASS +-> integrate the prepared VALUE-03 / VALUE-04 / VALUE-08 lanes ``` -VALUE-01 remains **ACTIVE**. Its only remaining acceptance gate is **VALUE-01D / #120**, now READY for one retained `REAL_ENVIRONMENT` execution after a fresh exact-head PRE_REAL/Built Product pass. +VALUE-01 remains **ACTIVE** until #120 passes. VALUE-02 software is integrated, but VALUE-02 acceptance remains **BLOCKED** by #120 and then #132. + +The previous VALUE-01 PRE_REAL handoffs are historical only because PR #141 moved `dev`; a new exact-head PRE_REAL/Built Product PASS is required for the current `dev` revision before representative execution. ## Active work | Workstream | State | Next gate | | --- | --- | --- | -| [Incremental value delivery](workstreams/incremental-value-delivery.md) | VALUE-01 ACTIVE; A/B/C DONE; D READY | refresh exact-head PRE_REAL, then execute #120 on representative hardware | +| [Incremental value delivery](workstreams/incremental-value-delivery.md) | VALUE-01 ACTIVE; VALUE-02 software integrated / acceptance blocked | refresh exact-head PRE_REAL -> execute #120 -> execute #132 | | [Product UX/UI convergence](workstreams/product-ux-ui-convergence.md) | automated product/visual acceptance PASS | representative human accessibility/usability acceptance; then finalize/delete | | [Representative device evidence](workstreams/representative-device-evidence.md) | EVID-001 READY | #120 retained real run supplies the first representative artifact set | | [Local LLM Server migration](workstreams/local-llm-migration.md) | MIG-001 DONE / MIG-002 evidence-blocked / MIG-003 blocked | EV-3 + real PL replacement evidence; consumed by VALUE-07 | +Prepared off `dev`, but intentionally not integrated until VALUE-02 real acceptance: + +- **VALUE-03A / PR #149** — consume runtime/model-declared generation parameter domains; +- **VALUE-04A / PR #148** — classify decision-grade device/resource evidence separately from contextual telemetry; +- **VALUE-08A / PR #147** — artifact-owned launcher that removes repository-development setup from normal launch. + ## Delivery model - `dev` is the integration line; ordinary feature branches start from current green `dev` and target `dev`. - `main` is stable/release-oriented and is promoted deliberately with `FULL` validation. -- VALUE slices are vertical user outcomes. After VALUE-02, configuration optimization, device-aware evidence and distribution may proceed in parallel when ownership does not conflict. +- VALUE slices are vertical user outcomes. After VALUE-02 real acceptance, configuration optimization, device-aware evidence and distribution may proceed in parallel when ownership does not conflict. - Parallel work uses explicit non-conflicting owners and one convergence gate; it must not create competing implementations of the same contract. - Do not build broad subsystems in anticipation of later slices. ## Evidence still required -- fresh exact-head PRE_REAL/Built Product readiness for the final integrated `dev` revision; +- fresh exact-head PRE_REAL/Built Product readiness for current `dev`; - VALUE-01D retained real target/model/device execution with verifier PASS; -- later VALUE-02/04/05/06 real decision, telemetry/repeatability and regression evidence; +- VALUE-02D retained 2+ real-model decision with verifier PASS; +- later VALUE-04/05/06 real telemetry/repeatability and regression evidence; - representative human accessibility/usability acceptance before a reference-grade human UX claim; - LLS EV-3 + real PL replacement run + post-disable cross-repository smoke before VALUE-07 cutover; - branch protection/admin work remains tracked separately in GitHub issue #61. From 061b24bf22b4c70a210cf786e60941c207af6c73 Mon Sep 17 00:00:00 2001 From: Daniele21 Date: Thu, 3 Sep 2026 17:43:04 +0200 Subject: [PATCH 2/3] docs: align incremental delivery with integrated VALUE-02 software --- .../workstreams/incremental-value-delivery.md | 55 ++++++++++++------- 1 file changed, 35 insertions(+), 20 deletions(-) diff --git a/docs/workstreams/incremental-value-delivery.md b/docs/workstreams/incremental-value-delivery.md index 5af0ae25..94245e43 100644 --- a/docs/workstreams/incremental-value-delivery.md +++ b/docs/workstreams/incremental-value-delivery.md @@ -29,50 +29,65 @@ M1-M9 in `docs/roadmap.md` remain the capability/maturity map. This workstream o | ID | User value unlocked | Depends on | Parallel | State | | --- | --- | --- | --- | --- | | VALUE-01 | **Real single-model evidence loop** — discover, test, inspect and export one real target/model | fresh exact-head PRE_REAL + VALUE-01D | internal lanes complete | ACTIVE | -| VALUE-02 | **Real model decision** — compare 2+ real candidates and return an explainable recommendation/no-rank | VALUE-01 | no | BLOCKED | -| VALUE-03 | **Configuration decision** — choose a supported configuration, not only a model | VALUE-02 | VALUE-04/08 | BLOCKED | -| VALUE-04 | **Device-aware decision** — real performance/resource evidence affects the trade-off | VALUE-02 | VALUE-03/08 | BLOCKED | +| VALUE-02 | **Real model decision** — compare 2+ real candidates and return an explainable recommendation/no-rank | VALUE-01D + fresh PRE_REAL + VALUE-02D | software A/B/C integrated | BLOCKED | +| VALUE-03 | **Configuration decision** — choose a supported configuration, not only a model | VALUE-02 real acceptance | VALUE-04/08 | BLOCKED | +| VALUE-04 | **Device-aware decision** — real performance/resource evidence affects the trade-off | VALUE-02 real acceptance | VALUE-03/08 | BLOCKED | | VALUE-05 | **Confidence / repeatability** — controlled variability supports the recommendation | VALUE-04 | late VALUE-03 | BLOCKED | | VALUE-06 | **Regression workflow** — real baseline vs candidate produces a policy outcome | VALUE-02, VALUE-05 | no | BLOCKED | | VALUE-07 | **LLS evaluation cutover** — PL owns new evaluation; LLS stays serving/runtime owner | VALUE-03 + LLS gates | no | BLOCKED | -| VALUE-08 | **Low-friction distribution** — launch/connect/evaluate without repo-development setup | VALUE-02 | VALUE-03..07 | BLOCKED | +| VALUE-08 | **Low-friction distribution** — launch/connect/evaluate without repo-development setup | VALUE-02 real acceptance | VALUE-03..07 | BLOCKED | -Allowed states: `READY`, `ACTIVE`, `BLOCKED`, `DONE`. After VALUE-02, VALUE-03/04/08 may proceed in parallel where ownership is independent. +Allowed states: `READY`, `ACTIVE`, `BLOCKED`, `DONE`. After VALUE-02 real acceptance, VALUE-03/04/08 may proceed in parallel where ownership is independent. ## VALUE-01 execution graph -The software/readiness lanes were developed independently and converged on `dev`. The representative-device run is now the only remaining acceptance gate. +The software/readiness lanes are integrated. The representative-device run remains the acceptance gate. | ID | Work | Owns/writes | State | | --- | --- | --- | --- | | VALUE-01A / #117 | Real built-browser loop against Local LLM Server | target-environment Playwright + bounded browser launcher | DONE | | VALUE-01B / #118 | Evidence completeness + portability verifier | real-runtime verifier + deterministic tests | DONE | | VALUE-01C / #119 | Exact-head real-run operator entry point | real-runtime smoke/runbook + readiness tests | DONE | -| VALUE-01D / #120 | Retained representative device execution | RUNTIME-1 artifact set + state transition | READY | +| VALUE-01D / #120 | Retained representative device execution | RUNTIME-1 artifact set + state transition | READY after fresh PRE_REAL | -A/B/C passed their required repository gates before integration. Because integration changes the commit SHA, #120 must start only after a fresh PRE_REAL/Built Product PASS whose source revision matches the final integrated `dev` HEAD. +PR #141 intentionally moved `dev` to integrate the completed VALUE-02 software for end-to-end testing. Therefore older PRE_REAL handoffs are historical evidence only; #120 must use a fresh PRE_REAL/Built Product PASS for the current `dev` revision. -## Current acceptance — VALUE-01 +## VALUE-02 execution graph -User outcome: +The deterministic/software portions are already integrated on `dev`; only representative acceptance remains. -> Connect a real Local LLM Server target on a representative device, run one real model, inspect trustworthy Run/Sample Evidence, and retain a portable evidence bundle. +| ID | Work | State | +| --- | --- | --- | +| VALUE-02A / #129 | Configured-target multi-model discovery + candidate evidence attribution | SOFTWARE DONE | +| VALUE-02B / #130 | Real multi-model Campaign/browser harness | SOFTWARE DONE | +| VALUE-02C / #131 | Retained multi-model decision evidence verifier | SOFTWARE DONE | +| VALUE-02D / #132 | Representative 2+ real-model decision | BLOCKED by #120 + fresh PRE_REAL | -Acceptance: +VALUE-02 is not `DONE` merely because A/B/C are integrated. Acceptance still requires #120 PASS first and then one retained REAL_ENVIRONMENT #132 run whose browser/backend/verifier evidence agrees. -- fresh exact-head PRE_REAL/Built Product readiness passes before the real run; -- real `/v1/models` discovery and inference complete through **Test a model**; -- first-party runtime identity and `/status` telemetry are retained when supplied, with explicit provenance; -- Run Detail and at least one retained Sample Evidence view are inspectable; -- canonical store + `.plab.zip` pass the VALUE-01B verifier; missing/not-retained evidence stays typed; -- one retained `real-runtime-device` / `RUNTIME-1` run is sufficient for a reviewer to reproduce or bound the claim. +## Current convergence sequence -`representative-device-evidence.md` owns the detailed real-device protocol/artifact rules. +```text +current dev +-> fresh exact-head PRE_REAL/Built Product PASS +-> #120 single-model REAL_ENVIRONMENT PASS +-> #132 multi-model REAL_ENVIRONMENT PASS +-> VALUE-02 DONE +-> integrate prepared VALUE-03A / VALUE-04A / VALUE-08A +-> continue VALUE-03 / VALUE-04 / VALUE-08 in parallel +``` + +Prepared isolated lanes already exist and remain intentionally unmerged until VALUE-02 acceptance: + +- VALUE-03A / PR #149 — runtime/model-declared bounded generation domains; +- VALUE-04A / PR #148 — policy-eligible device evidence classification; +- VALUE-08A / PR #147 — artifact-owned launcher. -## Later slice acceptance +## Acceptance | Slice | Minimum acceptance | | --- | --- | +| VALUE-01 | one real target/model/device; Test a model -> Run/Sample Evidence; canonical store + `.plab.zip`; VALUE-01 verifier PASS. | | VALUE-02 | 2+ real candidates, same use case/device; compatibility and explicit policy precede backend-owned recommendation/no-rank; differentiating evidence remains drillable. | | VALUE-03 | 2+ supported real configurations using runtime-declared mutable ranges; quantization stays candidate identity; unsupported parameters are not fake controls. | | VALUE-04 | real latency/throughput/resource measurements retain scope/unit/provenance; unavailable sensors stay unavailable; only comparable policy-relevant evidence affects decisions. | From 937c411ae02e2b664356526e77bb4dc938b858f9 Mon Sep 17 00:00:00 2001 From: Daniele21 Date: Thu, 3 Sep 2026 17:43:44 +0200 Subject: [PATCH 3/3] docs: preserve VALUE-01 acceptance detail --- .../workstreams/incremental-value-delivery.md | 20 +++++++++++++++++-- 1 file changed, 18 insertions(+), 2 deletions(-) diff --git a/docs/workstreams/incremental-value-delivery.md b/docs/workstreams/incremental-value-delivery.md index 94245e43..fdf14faf 100644 --- a/docs/workstreams/incremental-value-delivery.md +++ b/docs/workstreams/incremental-value-delivery.md @@ -52,6 +52,23 @@ The software/readiness lanes are integrated. The representative-device run remai PR #141 intentionally moved `dev` to integrate the completed VALUE-02 software for end-to-end testing. Therefore older PRE_REAL handoffs are historical evidence only; #120 must use a fresh PRE_REAL/Built Product PASS for the current `dev` revision. +## Current acceptance — VALUE-01 + +User outcome: + +> Connect a real Local LLM Server target on a representative device, run one real model, inspect trustworthy Run/Sample Evidence, and retain a portable evidence bundle. + +Acceptance: + +- fresh exact-head PRE_REAL/Built Product readiness passes before the real run; +- real `/v1/models` discovery and inference complete through **Test a model**; +- first-party runtime identity and `/status` telemetry are retained when supplied, with explicit provenance; +- Run Detail and at least one retained Sample Evidence view are inspectable; +- canonical store + `.plab.zip` pass the VALUE-01B verifier; missing/not-retained evidence stays typed; +- one retained `real-runtime-device` / `RUNTIME-1` run is sufficient for a reviewer to reproduce or bound the claim. + +`representative-device-evidence.md` owns the detailed real-device protocol/artifact rules. + ## VALUE-02 execution graph The deterministic/software portions are already integrated on `dev`; only representative acceptance remains. @@ -83,11 +100,10 @@ Prepared isolated lanes already exist and remain intentionally unmerged until VA - VALUE-04A / PR #148 — policy-eligible device evidence classification; - VALUE-08A / PR #147 — artifact-owned launcher. -## Acceptance +## Later slice acceptance | Slice | Minimum acceptance | | --- | --- | -| VALUE-01 | one real target/model/device; Test a model -> Run/Sample Evidence; canonical store + `.plab.zip`; VALUE-01 verifier PASS. | | VALUE-02 | 2+ real candidates, same use case/device; compatibility and explicit policy precede backend-owned recommendation/no-rank; differentiating evidence remains drillable. | | VALUE-03 | 2+ supported real configurations using runtime-declared mutable ranges; quantization stays candidate identity; unsupported parameters are not fake controls. | | VALUE-04 | real latency/throughput/resource measurements retain scope/unit/provenance; unavailable sensors stay unavailable; only comparable policy-relevant evidence affects decisions. |