VALUE-04 — Device-aware decision

[daniele21]

Goal

Make the recommendation meaningfully device-aware: comparable real performance/resource evidence from the selected device may affect the trade-off while unavailable or incomparable sensors remain explicitly unavailable.

Delivery / REAL_ENVIRONMENT policy

Software slices may be integrated incrementally into dev once their repository-owned deterministic gates pass. REAL_ENVIRONMENT acceptance is intentionally deferred until the planned software modifications are integrated and dev is the consolidated candidate. VALUE-04D / EVID-003 remains required before final device-aware claims, but it no longer blocks integrating VALUE-04A/B/C software.

Current gap on dev

• execution fingerprints already retain HardwareIdentity and runtime identity;
• black-box request performance and LLS /status telemetry are retained with scope/unit/provenance;
• the current LLS status collector primarily captures activity/chunk/concurrency observations, not a complete comparable resource envelope;
• EVID-003 is still PLANNED and explicitly requires supported sensors/metrics to preserve scope/unit/provenance while unavailable data remains typed;
• current policy must not treat missing resource evidence as zero or compare unlike sensors/devices.

Vertical slices

VALUE-04A — Policy-eligible device evidence contract

Define which retained performance/resource measurements can participate in a device-aware decision and the exact comparability requirements (metric name, unit, scope, provenance/protocol and device/runtime context). Do not add guessed measurements.

VALUE-04B — Resource evidence projection

Project eligible real measurements into Campaign Results / same-case views as a separate Resources dimension with explicit available/unavailable/not-comparable states.

VALUE-04C — Decision policy extension

Version the backend decision policy so device/resource evidence affects a recommendation only when explicitly policy-relevant and comparable. Preserve quality-first/compatibility-first semantics and no hidden universal score.

VALUE-04D — EVID-003 representative validation

After software convergence on dev, run on representative hardware, retain supported resource/performance metrics with exact provenance, exercise at least one unavailable/not-comparable case, and verify browser/backend agreement.

Acceptance

• every resource metric has explicit unit, scope, provenance/protocol and device/runtime context;
• unsupported sensors remain unavailable rather than zero;
• cross-device or cross-protocol evidence is not silently compared;
• decision rationale states exactly which resource evidence mattered, if any;
• Quality / Performance / Resources remain separate;
• final claims require REAL_ENVIRONMENT EVID-003 evidence.

Non-goals

• repeated/thermal confidence (VALUE-05);
• configuration search (VALUE-03);
• universal efficiency score.

[daniele21]

Software convergence update: VALUE-04A/B are integrated on dev via PR #153 -> 4c66541e2057f21afafc9e99828c6d7ed8518ce8.

Exact integration validation run 33957313692 PASS, including Python 3.12/3.13, Built Product, PRE_REAL J0-J9 and packaged J0/J1/J8/J9. Campaign Results and same-case views now distinguish policy-eligible Resources from contextual host/runtime telemetry with explicit available/unavailable/not-comparable states.

Remaining software item is VALUE-04C decision policy extension. VALUE-04D / EVID-003 remains REAL_ENVIRONMENT and is intentionally deferred until final software convergence on dev.