You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
ARB-APR-3: parity receipt for the decide lane's cell on the agent's host (Qwen3.5-9B-Q4_K_M @ lambda-labs) #3574
ARB-APR-001 row 3, aprender-owned.rust-agentic-harness.md §3.4: a lane on a backend that has not passed parity is not a reviewer — so the apr decide lane needs a committed parity receipt for the cell it runs.
Produced 2026-09-20 rather than waiting. An earlier note here held this row until the agent host's declared apr carries CUDA (ARB-APR-2a, blocked on forjar's cargo provider having no features: field). That was the wrong call: the CLI works today — apr run answers 20/20 cells across 0.8B–27B on two GPU hosts — and holding the receipt because a preferred cell is not declared yet ships nothing while the lane runs unreceipted.
Measurement, published 0.68.2 built from crates.io with --features cuda, on the agent host (noah-Lambda-Vector, RTX 4090, sm_89, x86_64):
apr parity ~/models/Qwen3.5-9B-Q4_K_M.gguf --prompt "<68-token corpus>" --json
→ 68 positions, parity: true, passed 68 / failed 0
→ min cosine 0.9987 at position 44 (threshold for qwen3.5-9b: 0.99)
→ judged: scripts/check_model_parity.sh --judge … --model qwen3.5-9b → PASS
What this receipt does and does not certify. It certifies the (model, quant, host) cell: Qwen3.5-9B-Q4_K_M on this host passes CPU-vs-CUDA parity at the shipped threshold. The decide lane's binary on this host is currently the default-features build (default = ["cli"], no CUDA), so the lane runs the CPU arm of a cell whose CPU and CUDA arms are now measured to agree at 0.9987. That is a loud, receipted state — the doctrine allows it; what it does not allow is silence.
When ARB-APR-2a lands and the declared resource carries --features cuda, the same command re-run on the CUDA arm replaces this record rather than supplementing it, and the backend named in the run receipt must match the declared backend (ARB-APR-2's own RED).
Remaining for done_when: the infra decide config must reference this receipt, and decide must refuse apr cell unreceipted when the reference is missing. That half is infra's.
PROMETHEUS (PRM-001) REX-01 cross-ref: the review lane (not the decide lane) puts gx10 CUDA as primary and lambda-labs as CPU-only failover (R-9). The ARB-APR-001 gx10 revision request is on paiml/paiml-implement#411; the executor declarations are paiml/infra#1088. Epic #4354.
P0 to P1 (2026-10-14 P0 triage): P0 now means it blocks the next release or carries the operator's own words on it; this issue does neither, so it stays on the 0.71.0 milestone at P1.
merged into #5317 (master ticket for epic #4381; triage rule 7b, 2026-10-14). It is a checklist row there; reopen this issue when work on it starts or its PR opens.
Activity
ARB-APR-3 receipt: name the comparator — this is self-consistency, no…
noahgift commentedon Sep 23, 2026
milestone → 0.72.0 per docs/roadmaps/milestone-triage-2026-09-23.md (#4024), operator approval 2026-09-23 (relayed by aprender-cf). Reason: parity receipt for the agent decide lane: Agent Ready (#4000)
noahgift commentedon Sep 25, 2026
PROMETHEUS (PRM-001) REX-01 cross-ref: the review lane (not the decide lane) puts gx10 CUDA as primary and lambda-labs as CPU-only failover (R-9). The ARB-APR-001 gx10 revision request is on paiml/paiml-implement#411; the executor declarations are paiml/infra#1088. Epic #4354.
release(0.69): batch — 15 receipted PRs + fleet-pv andon fix, one CI …
PMAT-3685: apr parity REFUSES (exit 12) a model the CUDA capability g…
noahgift commentedon Oct 14, 2026
P0 to P1 (2026-10-14 P0 triage): P0 now means it blocks the next release or carries the operator's own words on it; this issue does neither, so it stays on the 0.71.0 milestone at P1.
noahgift commentedon Oct 14, 2026
merged into #5317 (master ticket for epic #4381; triage rule 7b, 2026-10-14). It is a checklist row there; reopen this issue when work on it starts or its PR opens.