perf(#190): the pooled reader's scope, measured on all three rows — plus the save-path attribution and the stride1 scaling curve by yifanxuaaa · Pull Request #206 · Ephemeral-AI-Lab/layerfs

fix(core): the locator query's LIMIT costs 15.5 us a call, and #209's regression is mostly that

yifanxuaaa

committed

63fa15c

docs(#209): the RCA, its pre-registration, and the second writer's measured wait

yifanxuaaa

committed

c2197be

docs(#209): refute the write_pack hypothesis, withdraw L54's 4.8 s, and bound the residual at 1.59 s

yifanxuaaa

committed

09bfbd2

docs(#209): attribute the whole 10.107 s to the millisecond, and hand off the commit path

yifanxuaaa

committed

6e0d526

docs(#209): date the commit-optimization handoff from the commit that landed it

yifanxuaaa

committed

bbc121b

docs(#209): the commit-optimization handoff's next free ledger entry is L57, not L56

yifanxuaaa

committed

10df6ef

fix(core): a step commits the policy state it changed, not the policy state it re-asserted

yifanxuaaa

committed

7045806

docs(#209): the per-append commit multiple is 7.7x, not 4x, and the regression is window-dependent

yifanxuaaa

committed

1dd1cb5

docs(#209): keep the committed addendum identical to what was posted

yifanxuaaa

committed

f3e84c0

docs(#209): the per-step statement cache is withdrawn, and hand off the commit-time RCA

yifanxuaaa

committed

3419266

docs(#209): confirmation window — the kept treatment reproduces on commit_ns

yifanxuaaa

committed

b65d09a

docs(#209): start the handoff from b65d09a81, and carry the confirmation window

yifanxuaaa

committed

6221cad

docs(#209): the step's transaction read with SQLite's own instruments — the commit path is closed

yifanxuaaa

committed

73e0b96

docs(#209): pre-register the pack-format treatment — reserve the directory, pre-allocate the row, write appends incrementally

yifanxuaaa

committed

1bfb0c5

docs(#209): the pack format's price measured — commit_ns -52.8 % — and reverted by owner decision

yifanxuaaa

committed

f876242

docs(#209): retain the closing summary as posted, and close the issue

yifanxuaaa

committed

8be0ae1

Navigation Menu

perf(#190): the pooled reader's scope, measured on all three rows — plus the save-path attribution and the stride1 scaling curve - #206

perf(#190): the pooled reader's scope, measured on all three rows — plus the save-path attribution and the stride1 scaling curve#206
yifanxuaaa merged 16 commits into
mainfrom
codex/190-pooled-scope

Commits on Sep 20, 2026

fix(core): the locator query's LIMIT costs 15.5 us a call, and #209's regression is mostly that

docs(#209): the RCA, its pre-registration, and the second writer's measured wait

docs(#209): refute the write_pack hypothesis, withdraw L54's 4.8 s, and bound the residual at 1.59 s

docs(#209): attribute the whole 10.107 s to the millisecond, and hand off the commit path

docs(#209): date the commit-optimization handoff from the commit that landed it

docs(#209): the commit-optimization handoff's next free ledger entry is L57, not L56

fix(core): a step commits the policy state it changed, not the policy state it re-asserted

docs(#209): the per-append commit multiple is 7.7x, not 4x, and the regression is window-dependent

docs(#209): keep the committed addendum identical to what was posted

docs(#209): the per-step statement cache is withdrawn, and hand off the commit-time RCA

docs(#209): confirmation window — the kept treatment reproduces on commit_ns

docs(#209): start the handoff from b65d09a81, and carry the confirmation window

docs(#209): the step's transaction read with SQLite's own instruments — the commit path is closed

docs(#209): pre-register the pack-format treatment — reserve the directory, pre-allocate the row, write appends incrementally

docs(#209): the pack format's price measured — commit_ns -52.8 % — and reverted by owner decision

docs(#209): retain the closing summary as posted, and close the issue