pirl-unc / hitlist / 39195446795
89%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 63
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 88.824%. Remained the same
39195446795

push

github

web-flow 
Record the memory attribution and the allele taxonomy (#566, #597) (#608)

* Record the memory attribution and the allele taxonomy (#566, #597)

tasks/todo.md: what the 1,057,558 unmatched rows were (87.6% in curated
studies, so a code defect rather than the curation backlog), what
dominates #566's 15.2 GB peak, the four directions measured and rejected,
and why each remaining issue is left open.

tasks/lessons.md gains five entries, each from a measurement that
contradicted a plausible expectation:

- Live bytes and touched pages are different quantities needing different
  instruments. Narrowing columns to category on arrival cut live bytes
  1.3 GB and cost 337 MB of CI peak RSS -- the #569 astype trap reached
  through a better instrument that is structurally blind to it. Of four
  instruments tried on unchanged code, only CI /usr/bin/time -v answers
  the question.

- Bound a direction before optimizing inside it. The number that closed
  #572 was arithmetic nobody had done: 3.28 GB of pure Arrow offsets,
  independent of content, against 5.4 GB of live content. The peak is the
  frame's shape.

- An extrapolation is worth measuring when you will carry it as a fact.
  The dictionary probe measured -0.07%, 0.11x the job's noise, from a
  local reduction of 0.89 GB -- a 1.2% transfer ratio, below the points
  the estimate came from. Right conclusion, optimistic by two orders.

- Widening a predicate for coverage can destroy the validation that made
  it safe. Dropping the haplotype roster for "any non-numeric field"
  gained 6 rows and made H2-K*x return ("H2-K*x", "exact", 1).

- A fixture that does not reproduce production's dtype makes every test
  built on it decorative. Correcting one immediately exposed
  qc.proteome_coverage() as broken outright on the real corpus.

Claude-Session: https://claude.ai/code/session_01UfmwqWGdHv9QTHWQXXhNdD

* Plan PR 608 review and release

* Clarify historical memory comparison

12383 of 13941 relevant lines covered (88.82%)

0.89 hits per line

Jobs
ID Job ID Ran Files Coverage
1 39195446795.1 63
88.82
GitHub Action Run
Source Files on build 39195446795