Record the spread of the reads and the turns beside the tokens' in the task-bench decision
CI / tests (pull_request) Has been cancelled
CI / tests (pull_request) Has been cancelled
This commit is contained in:
+11
-1
@@ -825,7 +825,7 @@ Valid while Claude Code loads `~/.claude/rules/*.md` in every project, a path-sc
|
||||
|
||||
**Defaults stay 5 bugs and 5 runs.** A sixth run per side cuts the spread of the mean of the bugs' cache-read changes from 16% to 13% for 10 more runs, about 35 minutes.
|
||||
|
||||
**A chunk-sized change doesn't rise above the noise.** PR #490 moved the medians by 5% to 9%. The spread at 5 × 6, the largest count measured, is 11% to 13%. It falls with the square root of the runs, so seeing a 9% change takes about 13 runs per side and bug: 130 runs, 8 hours and 70M cache-read tokens. No chunk waits that long. The bench shows a change of more than about 15% in tokens, if you work out the mean of the bugs' own changes from its bug rows. The All row it prints needs more than 31% in output tokens and 50% in cache-read tokens. What to try is filed as items 514 (print the mean of the bugs' changes, which has a third to half of the All row's spread in tokens), 515 (plant bugs Haiku fails at the base, so the pass rate moves) and 516 (measure the spread of the reads and the turns, which are counts, not token sums).
|
||||
**A chunk-sized change doesn't rise above the noise.** PR #490 moved the medians by 5% to 9%. The spread at 5 × 6, the largest count measured, is 11% to 13%. It falls with the square root of the runs, so seeing a 9% change takes about 13 runs per side and bug: 130 runs, 8 hours and 70M cache-read tokens. No chunk waits that long. The bench shows a change of more than about 15% in tokens, if you work out the mean of the bugs' own changes from its bug rows. The All row it prints needs more than 31% in output tokens and 50% in cache-read tokens. What to try is filed as items 514 (print the mean of the bugs' changes, which has a third to half of the All row's spread in tokens) and 515 (plant bugs Haiku fails at the base, so the pass rate moves). Reads and turns, which are counts, not token sums, spread as wide as the tokens: see "Reads and turns" below.
|
||||
|
||||
**Pilot first.** A new setup, or a project the bench hasn't run on, runs 2 bugs × 2 runs before a full bench (lilleman, 2026-11-21). A pilot takes about 40 minutes. The first pilot of this calibration passed no run, because the API refused a token that held a line break. It showed that within 15 minutes, where a full bench would have planted for 25 minutes first and printed the same.
|
||||
|
||||
@@ -843,6 +843,16 @@ Measured on `fejkdata` (Go, 5 bugs planted in the files `larv-review places` nam
|
||||
| 3 × 6 | 28 / 14 | 40 / 17 | 16 / 12 |
|
||||
|
||||
The All row pools every run, so its median lands in whichever bug's runs sit in the middle of the pooled costs, and a bug's cost differs from another's by up to 8×. In tokens, the mean of the bugs' changes has a third to half of the All row's spread. In wall time it gains only from 5 × 5. The 86 at 5 × 3 wasn't traced. For the mean of the bugs' token changes, 5 × 4 and 3 × 6 give the same spread (14 and 14, 18 and 17) for 40 and 36 runs.
|
||||
- **Reads and turns** (item 516, 2026-11-22) spread as wide as the tokens. `fejkdata` at `095d9bb3` against itself, 5 bugs × 5 runs after a 2 × 2 pilot: all 58 runs passed. Each bug's 10 runs were split at random into two sides 4,000 times. Each cell is the 95th percentile of the change's size, for the All row's change as the bench prints it, then for the mean of the bugs' own changes:
|
||||
|
||||
| Figure | All change | Mean of the bugs' changes |
|
||||
|---|---|---|
|
||||
| Output tokens | 25% | 12% |
|
||||
| Cache-read tokens | 41% | 17% |
|
||||
| Turns | 31%, or 3.0 turns | 14%, or 1.8 turns |
|
||||
| Reads before the first open of the bug's file | 4.0 reads | 0.6 reads |
|
||||
|
||||
Neither count sees a change the tokens can't. The bugs' median turns average 14.5 and their median reads 3.9, so the mean's spread is 12% of the turns and 15% of the reads, in line with the tokens' 12% and 17%. The All row's 4.0 reads is more than the reads themselves. Reads are small whole numbers, from 0 to 9, and 9 of the 50 runs never opened the bug's file, so a read is given as a difference, never a share. The tokens' own figures moved by up to a fifth against the first calibration's above (25 and 41 against 31 and 50 at the All row), so each 95th percentile is good to about that much.
|
||||
- **One finished chunk**: `fejkdata`'s PR #490, base `33b7fa80`, head `5b9c2a14`, which the comprehension panel ruled better on Navigation, Locality and Shape and the same on Self-sufficiency. The bench planted 5 bugs and ran 38 of 50 runs before the machine restarted. The first 3 runs per bug and side, 30 runs, are the ones counted, and all passed. The mean of the bugs' changes was +4.6% in output tokens, +8.9% in cache-read tokens and +3.8% in turns, with the bugs' own changes in cache-read tokens from -41% to +47%. The two token figures sit inside the spread at 5 × 3, 17% and 22%; the spread of the turns wasn't measured. The mean of the wall times' changes, +105%, lies above the 86% of that row, from one bug whose head runs took 6.8 times as long at the median. Nobody read its streams to find out why. The head cost slightly more tokens where the panel finds it easier to read, and the wall time can't be judged, so the bench neither confirmed nor contradicted the panel.
|
||||
- **Cost**: 5 bugs × 6 runs took 3.9 hours and 34M cache-read, 1.2M cache-write and 0.3M output tokens; a run takes about 3.5 minutes, and planting 16 to 25 minutes. A 5 × 5 bench is about 3.3 hours and 28M cache-read tokens. Gremlins ran out of 8 GB at the base of PR #490, so the help's example takes 16 GB.
|
||||
|
||||
|
||||
@@ -3384,3 +3384,20 @@ Read most (opens and searches):
|
||||
- `bin/reading_cost/measure.py` 2
|
||||
- `bin/review_rules/review_folder.py` 2
|
||||
- `bin/task_bench/figures.py` 2
|
||||
|
||||
## Reading cost, chunk 666aabdd77a8, finished 2026-11-22T15:50:09Z
|
||||
|
||||
283046 new input tokens before the first commit.
|
||||
|
||||
Read most (opens and searches):
|
||||
|
||||
- `docs/decisions.md` 14
|
||||
- `bin/task_bench/main.py` 9
|
||||
- `bin/task_bench/figures.py` 4
|
||||
- `bin/task_bench/plant.py` 4
|
||||
- `bin/task_bench/test_figures.py` 4
|
||||
- `bin/task_bench/runs.py` 3
|
||||
- `CHANGELOG.md` 2
|
||||
- `README.md` 2
|
||||
- `docs/comprehension-history.md` 2
|
||||
- `docs/score-anchors.md` 2
|
||||
|
||||
Reference in New Issue
Block a user