Performance
These numbers were measured on 2026-11-03 or earlier on a reference workstation. They describe the index and the local model, not a general guarantee.
Index
| Repository | Tokens | Chunks | Index time |
|---|---|---|---|
| Auwai | 282,315 | 281 | about 2.2 s |
A repeated ledger sync reuses unchanged labels. The measured repeated sync on Auwai was about 0.035 s.
Full-read plan
| Repository | 64K read segments | 128K | 256K |
|---|---|---|---|
| Auwai | 9 | 4 | 2 |
The plan adds one final synthesis segment. The read budgets are 35,344 tokens at 64K, 92,688 at 128K, and 208,144 at 256K.
Model latency
Model throughput depends on the endpoint, the model build, quantization, and the context window. Measure it against the endpoint you use. The repository does not publish a reference machine or a hardware-specific number. See calibration for the formula.
Kernel-scale ledger
The Linux kernel index was used on 2026-11-03:
| Measure | Result |
|---|---|
| Chunks | 177,872 |
| Included chunks | 177,864 |
| Included tokens | 364,909,391 |
| First ledger sync | 154.8 s |
| Unchanged resync | 14.8 s |
| Changed-path ledger sync | 1.1 to 2.0 s |
| Watcher end to end | 16.9 s |
| Ledger size | about 492 MB |
The first sync is a one-time cost after an index build. The unchanged resync is a full reconciliation. A changed-path sync updates only the affected chunks. The ledger slice took 1.1 to 2.0 seconds. The watcher also refreshes the JSONL index and the derived map before the ledger update; one kernel change took 16.9 seconds end to end. That is down from a full ledger reconciliation of about 109 seconds. The partition table is marked dirty; the survey and planner build the tree from the current chunks.
The first-sync and unchanged-resync times were measured before 23 .log
fixture files were excluded from the index.
The current index is 23 chunks smaller; the timings are unchanged at this
scale.
What is not measured
The time to read a segment on a model is not measured end to end for every endpoint. One 64K full-window segment was validated in the stage 9 evidence. The current scale benchmark is Auwai on itself and the Linux kernel.