Skip to content

docs: require a work-log record per experiment, and record the region floor - #10

Merged
shinaoka merged 1 commit into
mainfrom
docs/worklog
Oct 14, 2026
Merged

shinaoka merged 1 commit into
mainfrom
docs/worklog

Conversation

@shinaoka

Copy link
Copy Markdown
Member

Makes the fork's records followable, and puts the format where the others already live.

What

  • AGENTS.md gains a Work Logs And Review Intent section: follow the shared Work Logs And Design Records policy and the local format in docs/worklogs/README.md; one record per change theme for a nontrivial multi-phase change, a non-obvious design choice or a performance experiment. PATCHSET.md stays the ledger of what the fork carries; the record explains why it carries it — the isolation, the candidates that were tried and dropped with the measurement that dropped them, and what is linked to the result. Unmeasured leads go under "what remains unverified" in the record, not into the ledger.
  • AGENTS.md also states that upstream feedback is deferred at this stage: measure an upstream defect, record it with the upstream revision that was measured, and leave the decision to the maintainer. Nothing is prepared or opened upstream.
  • docs/worklogs/README.md: the local format, adapted from tenferro-rs's docs/worklogs/README.md (same policy, same ## Decisions / ## Verification conclusions and constraints skeleton), plus one duty specific to this fork: a performance record says which upstream revision was measured, or that none was.
  • docs/worklogs/2026-10-14-compact-factor-region-floor.md: the first record, for perf(householder): no parallel region below the compact factor's own floor #9. Why the floor is asked once, the two rejected candidates with the numbers that killed them (the area-based gate: 4096x256 f64 6.36 → 6.82 ms; forcing the n x n x n head product to Par::Seq: 512x512 f64 2.49 → 3.43 ms), the upstream comparison (upstream 0.24.4 is 4.46x slower threaded at 128x128 f64, against the fork's 4.57 before the change), the limits of the evidence (a scan, not a paired comparison; one unattributable pooled cell; one pre-existing unrelated test failure), and the three leads nobody has measured yet (bidiag.rs discards par, evd/tridiag.rs never threads, lu scales 1.85x at 512 and needs a phase profile first).
  • The ledger row for the region floor now links its record.

Documentation only: no source change, no PATCHSET.md status change.

The fork's whole purpose is a measurable delta against upstream, so the reason for
each ledger row and the course of an optimisation have to survive the session that
made it. AGENTS.md now points at the shared Work Logs And Design Records policy and
the local format in docs/worklogs/, states that upstream feedback is deferred at
this stage, and says where an unmeasured lead belongs.

The first record covers #9: the floor the compact factor now asks
once, the two candidates that lost with the measurements that killed them, how the
fork and upstream compare, and the leads this session did not measure.
@shinaoka
shinaoka merged commit 9de95f7 into main Oct 14, 2026
7 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant