CLAIMS salience register

Generated by tools/claims_salience_register.py — the first wired consumer of the dos salience verdict (dos-kernel docs/391 (salience)). It routes every claim in CLAIMS.md through dos.salience.partition: [SHIPPED] classifies LIVE, while [SIMULATED]/[STUB] — true at their stated scope but off the production hotpath — classify PARKED and land in the recoverable register below. See the usefulness audit.

The contract: PARKED ≠ dropped. Every parked claim is RETAINED here with its typed reason and a re-activation line — never silently lost on a cleanup. The no-loss invariant is asserted on every run: 106 claims in → 106 classified, 0 dropped.

PARKED · SIMULATED (6)

Reactivation: promote to [SHIPPED] when the real path is exercised on live hardware / a live engine and the illustrative numbers become measured — the seam is already real

claim section
continuous-batching / token-per-watt are read-only SIMULATED telemetry (no watt source on the box). NOTE: “no GPU dependency” is no longer s… What fak is NOT
headroom (Rust) page-out codec: the v0.1 default is pure-Go content-addressed page-out; the headroom backend is an optional labeled seam, … Context-MMU (write-time result admission)
metrics-service scrape adapter / KV-residency / token-per-watt: labeled SIMULATED telemetry — there is no watt source on the build box (… Engine
Real-world vDSO hit-rate: the demo trace tau2-smoke is deliberately cache-favorable (~50% hits). The EXPERIMENTS measured addressable puri… Tool vDSO (3-tier local fast path)
The live WALL-CLOCK validation of the floor’s ratios at >100k (the floor↔wall-clock loop sessionbench closes at small scale via -validate)… Ultra-long-context work floor (longctxbench — per-agent context > 100k tokens)
The token-multiplier, prefix-cache tax_clawed_back (~62% at the N≈256 plateau on the default cost model), critical-path-vs-total-work late… Fan-out benchmark (fanbench — one master goal → N sub-agents, N=1…1024)

PARKED · STUB (7)

Reactivation: promote to [SHIPPED] when the deferred behavior is built behind the present plumbing — the labeled no-op is the placeholder, not a missing seam

claim section
Decode-time logit-mask (grammar-constrained generation) — the strongest form (never-emit a malformed call) — requires owning the decode loop… Pre-flight ladder + grammar rung
Dream-time durability consolidation (principled sleep-time promotion) over the rung-1 class signal. Tracked under the S7 epic #82; live child… S7 write-time durability gate (context is not memory)
No FINE-TUNED syscall/adjudication LLM and no AsyncLM interrupt behavior — the harvest-corpus consumer edge CLOSED (#580) by a SMALL class… What fak is NOT
No LIVE transport attaching a real external serving engine’s KV region — vLLM/SGLang owns the KV in Python/CUDA, so importing its pinned p… What fak is NOT
Rung-2 bitemporal: recall.Page validity interval (ValidFrom/ValidTo) + an as-of read gate (ErrExpired) that makes bounded the firs… S7 write-time durability gate (context is not memory)
Rung-2 dry-run probe and rung-3 sandbox probe — the offline/sandbox escalation rungs above rung-1 — are not built in v0.1; only rung-0 stati… Pre-flight ladder + grammar rung
Rung-3 engine-integrated TTL: kvmmu.Segment TTL + Context.Expire over the bit-exact model.KVCache.Evict, so a turn/session span is for… S7 write-time durability gate (context is not memory)

Cross-checked against the ledger: LIVE == 93×[SHIPPED], PARKED == 6×[SIMULATED] + 7×[STUB]. Regenerate with python tools/claims_salience_register.py --markdown docs/claims-salience-register.md.