semantic activate rejected on fresh isolated daemon: train fallback bytes changed; runtime degraded invalid_runtime_status

[ScriptedAlchemy]

Shipped semantic activation journey is rejected on a fresh isolated daemon

Exact operator path, binary built from PR #882 head c6130cb23 (the same code is on codex/tracedecay-total-redesign-plan-reopened): scripts/with-isolated-tracedecay-daemon.sh --bin target/debug/tracedecay --ready-timeout 60 -- <probe> against a 2-file git project, semantic models linked from ~/.tracedecay/semantic-models.

Observed, in order:

1. Before activation, tracedecay tool tracedecay_runtime reports
   semantic_runtime.state = {"state":"degraded","reason":"invalid_runtime_status","active_generation":"sha256:20a6b83f…"} while the code index is serving=ready, freshness current.
2. tracedecay semantic activate --json (after index readiness) exits 1:

   semantic activate: evaluating profile 'hybrid-conservative' natively for <project> (current+10x workload; typically minutes)
   Error: config error: semantic activation rejected: semantic activation input was rejected: semantic profile qualification failed: native semantic evaluation failed: native activation direct evaluation report failed: hybrid-conservative:train fallback bytes changed
   

   (crates/tracedecay-query/src/search_quality/report.rs:470). An earlier run in the same setup instead failed with semantic_activation.model_not_installed (state: indexing) while the runtime was still acquiring, so the CLI does not wait for or report acquisition either.
3. After the rejection, CLI tracedecay_search and MCP tools/call tracedecay_search both report the exact code generation with freshness=fresh, but semantic = {"status":"unavailable","reason":"calibration_unavailable","mode":"fallback_allowed"}.

So the review requirement "shipped operator semantic-activation journey verified on the exact runtime" (#753, #882, #707 acceptance) currently fails. Either the direct-evaluation determinism check is wrong for a freshly indexed project (fallback results legitimately change while the index settles, so the comparison must bind a generation), or the runtime really serves different fallback bytes across two identical queries, which is a retrieval determinism defect. invalid_runtime_status on a fresh profile also needs a root cause: it is reported before any activation exists.

Repro artifacts: probe script and outputs under the session scratchpad (probe.sh, out3/), isolated-daemon wrapper in scripts/.

Acceptance: on a fresh isolated daemon the shipped semantic activate succeeds (or reports a typed, actionable acquisition/loading state and then succeeds), tracedecay_runtime never reports invalid_runtime_status without a cause, and CLI + MCP tracedecay_search report semantic.status = available bound to the exact runtime generation. Regression test through the production composition, not a unit fake.

[ScriptedAlchemy]

Fixed on codex/tracedecay-total-redesign-plan-reopened (PR #707): c5fe03731 reports a never-activated profile as semantic_not_activated instead of invalid_runtime_status; 4b338d393 makes the direct-evaluation diagnostic name the real cause (pin drift, with expected/observed digests) instead of the tautological 'fallback bytes changed'; 373fc9c52 indexes the attested symbol path so qualified-symbol queries (VectorWatermark::merge_max, TimeInterval::validate) return candidates again; 2abda27c8 re-pins the workload digests and gates the pin in CI; 4e6ecf029 regenerates the packaged native qualification through the production path. Probe on the exact operator path (isolated daemon, shipped binary): semantic activate rc=0 state=ready with a receipt on the exact generation; CLI and MCP tracedecay_search both report semantic status=complete, coverage.semantic=complete, recall=full on the same code_generation. Evidence: search-eval lib 39 passed, query lib 300 + search_quality_suite 68, usecases semantic_runtime 107, dashboard-api 219, workspace clippy clean.