This page answers one question: are the same measurements improving over time? Runs with a different corpus or definition are kept separate.
Headline KPIs over time
One point per engine run. These 71 points compare engine revisions at their largest completed checkpoint. Packet-by-packet progress for each engine is shown on Runs.
Corpus coloured
81.5%
Share of observed clauses carrying a rule.
60 engine runs · 73.6% → 81.5% (+7.9)
Corpus closed
47.6%
Share of clauses a rule claims need no more processing; this is not proof.
60 engine runs · 46.6% → 47.6% (+1.0)
Corpus words closed (recurse:leaf)
17.2%
Share of emitted words in final recipe segments explicitly marked recurse: leaf; this is a rule claim, not proof.
60 engine runs · 17.0% → 17.2% (+0.2)
Shapes with rules
59.9%
Share of observed powerset shapes covered by rules.
60 engine runs · 52.0% → 59.9% (+7.9)
Sentences until a new pattern
8.8
Expected corpus cost of finding another unseen pattern; higher means greater saturation.
60 engine runs · 8.8 → 8.8 (+0.0)
Corpus processed
100.0%
Share of the frozen corpus processed in this generation.
60 engine runs · 100.0% → 100.0% (+0.0)
Pipeline sentences / second
1.6
End-to-end diagnostic throughput, not Simplify alone.
60 engine runs · 3.3 → 1.6 (-1.7)
Rules
1,359
Rules registered in the measured Simplify engine.
60 engine runs · 1,010 → 1,359 (+349.0)
⚠ REPLAY REQUIRED — coherent frozen KPI/trie/queue run at 2ca1c0abe8b29c6bf6bc96375e26014656801799; live Simplify has uncommitted source or grammar changes at 414961792. (It was 41496179210205d706f85dd6a87e31a0138ebc73 when this run finished.) The census remains auditable, but do not treat it as a current-engine result.
Engine runs
One row per engine version. The 500-sentence checkpoints, restarts, and old per-file measurements are collapsed here, so they cannot push earlier engines out of view. The row reports the largest completed census retained for that engine.
engine run
sentences processed
corpus coloured
clauses closed
words closed (recurse:leaf)
shapes with rules
sentences / new pattern
rules
engine 2ca1c0ab · 78 measurements
38,698
81.5%
47.6%
17.2%
59.9%
8.8
1,359
engine 3af73dc1 · 78 measurements
38,698
82.7%
47.2%
17.3%
60.5%
8.8
1,351
engine 342f66d1 · 78 measurements
38,698
83.4%
48.5%
17.2%
60.3%
8.8
1,342
engine 3bf9aa34 · 78 measurements
38,698
83.7%
48.3%
17.4%
61.4%
8.8
1,341
engine 8bc3758e · 78 measurements
38,698
83.7%
48.3%
17.0%
61.4%
8.8
1,316
engine 723ea2b3 · 78 measurements
38,698
83.4%
48.4%
16.6%
60.4%
8.8
1,308
engine eb07227b · 78 measurements
38,698
83.3%
48.2%
18.3%
60.5%
8.8
1,305
engine e97a9f7a · 78 measurements
38,698
72.0%
48.0%
18.3%
54.5%
8.8
1,303
engine 38edc419 · 78 measurements
38,698
72.0%
48.0%
18.3%
54.5%
8.8
1,301
engine ee4884cf · 78 measurements
38,698
72.0%
48.0%
18.3%
54.5%
8.8
1,301
engine 5ad8cded · 78 measurements
38,698
73.4%
48.2%
18.7%
55.0%
8.8
1,300
engine d0915c0d · 78 measurements
38,698
73.3%
46.8%
17.1%
57.3%
8.8
1,276
engine 0f41d461 · 78 measurements
38,698
73.3%
46.8%
17.1%
57.3%
8.8
1,269
engine 10c906c4 · 78 measurements
38,698
72.9%
46.8%
17.1%
57.3%
8.8
1,269
engine 6f31fc7a · 78 measurements
38,698
73.0%
48.2%
18.6%
55.1%
8.8
1,269
engine b7976c70 · 78 measurements
38,698
73.3%
48.2%
18.6%
55.2%
8.8
1,268
engine 4f710e3d · 78 measurements
38,698
73.2%
48.2%
18.6%
55.2%
8.8
1,264
engine 80df4143 · 78 measurements
38,698
73.2%
48.0%
18.7%
55.1%
8.8
1,264
engine e303a926 · 78 measurements
38,698
73.1%
47.8%
19.0%
55.2%
8.8
1,262
engine 287cc01b · 78 measurements
38,698
73.1%
47.8%
19.0%
55.2%
8.8
1,261
Older measurement cohorts remain in this archive, but are not joined into the current trend charts because their corpus or metric definition differs. “Closed” is a rule's claim that no more processing is needed, not proof that the result is correct.
Archive only. These snapshots retain their original figures and provenance, but do not support a current trend claim. The current corpus-trie cohort begins separately and will gain a trend only after comparable full-census snapshots accumulate.
Archived legacy snapshots
Each row includes the frozen engine and corpus fingerprint so the old evidence remains auditable without being mixed into the current census.
frozen
hardest miss
proven
open
sentences
engine 70716aee · audit v18 · corpus 1ea5474
45.9%
49
27
10,685
engine 6ed2ff6a · audit v18 · corpus 1ea5474
45.9%
49
27
10,527
engine 93deddbf · audit v18 · corpus 1ea5474
46.8%
49
27
3,664
engine d9733d93 · audit v18 · corpus 1ea5474
46.7%
49
27
3,663
Snapshots are keyed by engine build and audit version, so re-running the writer when nothing has changed is a no-op. Frozen by bin/snapshot-report; the record lives in reports/ and is versioned with the code.