Each row expands the stages above it. Segment widths show their share of the parent time; colours distinguish branches and depth.
Packet work180.16s
Prepare packet0.20s
Resolve current engine0.00s
Fetch latest Simplify main0.00s
Resolve selected commit0.00s
Check / refresh corpus index0.00s
Read corpus identity0.04s
Resolve previous engine0.02s
Fingerprint current grammar and parser0.00s
Fingerprint previous grammar and parser0.00s
Fingerprint visualizer pipeline0.05s
Check static-page renderer0.03s
Choose packet / resume position0.02s
Other preparation0.03s
KPI refresh178.92s
Prepare pinned engine0.01s
Check out required engine files0.00s
Link runtime data0.00s
Other snapshot preparation0.01s
Prepare worker0.12s
Identify engine and runtime0.07s
Retire previous worker0.00s
Remove previous engine snapshot0.00s
Write worker request0.02s
Model initialization0.00s
Other worker preparation0.02s
Compute KPIs and artifacts177.18s
Select sentences0.05s
Measure new patterns3.05s
Check whether these sentences were already counted0.00s
Check indexed sentence records · read 0 bytes · host I/O stall some 0 ms, full 0 ms0.00s
Open acquisition ledger0.00s
Read ledger schema0.00s
Close acquisition ledger0.00s
Parse uncached sentences2.79s
Colouring and hypertrie168.65s
Prepare corpus input0.00s
Load Simplify modules0.00s
Set up sentence replay0.00s
Replay corpus sentences121.13s
Simplify120.29s
Coreference3.03s
Preprocess coreference3.03s
Preprocess and lineage0.10s
Preprocess layout and lineage0.10s
Tree++ and rule processing25.59s
Prime parser and build forests2.64s
A · Expand input0.01s
A · Build document packets0.00s
B · Apply rules / normalize22.80s
C · Join output surfaces0.00s
D · Build annotation cache0.14s
Build and emit output67.19s
Render simplified text0.01s
Build output annotations and Tree++ structure · after first: median 64–128 ms, p95 256–512 ms · slowest sentence #410: 5,390 ms (10 words, 39 characters) · 1001-stories-cc0-2026:L160367.19s
Produce semantic payload · after first: median 64–128 ms, p95 256–512 ms · slowest sentence #428: 480 ms (16 words, 91 characters) · five-minute-mysteries-pd-2026:L314452.91s
Legacy annotation assembly7.16s
Build WordNet profile0.15s
Apply authored-marker overlay0.03s
Build document units5.11s
Run final annotation passes1.84s
Other legacy annotation work0.02s
Other semantic production45.75s
Build typed document0.88s
Post-assembly13.39s
Complete annotation envelope0.01s
Other / not yet broken down24.38s
Convert annotations to JSON0.66s
Visualizer bookkeeping0.04s
Count closed words0.03s
Attribute rule activity0.00s
Merge clauses and relations0.00s
Build output clauses0.01s
Other sentence replay work0.14s
Prepare replay output0.00s
Write annotations0.26s
Prepare Hypertrie input4.78s
Hypertrie41.82s
Parse output clauses1.28s
Build trees and slices9.05s
Finish Rust aggregation10.03s
Compact and write indexes16.46s
Other hypertrie work0.01s
Worker overhead outside measured trie work4.99s
Setup before trie work0.09s
Worker dispatch / wrapper overhead4.90s
Validate Hypertrie output0.00s
Publish graph artifacts0.13s
Check rule ownership0.50s
Unattributed coverage time0.01s
Measure rule activity0.03s
Update shape ledger0.00s
Measure rule-message context0.00s
Load enabled-rule inventory0.00s
Inspect rule-work queues0.00s
Compare full-corpus progress0.00s
Compare ready-to-test rules0.00s
Other KPI work5.40s
Publish latest KPI and manifest1.57s
Check KPI output is complete0.04s
Validate Hypertrie index0.10s
Audit rule ownership0.00s
Confirm engine stayed consistent0.01s
Write run manifest1.29s
Publish verified files · host I/O pressure during publish: some 0.000s, full 0.000s0.11s
First parse call: 0.10s; included in the steady rate.
Hypertrie aggregation: Rust backend · 1 disk shards
Enabled-rule inventory: reused the worker's in-memory inventory
Enabled-rule inventory parser: not used (process cache hit)
SpaCy parser timer: 2.20s (diagnostic; overlaps the engine phases above and is not added again)
Simplify phases are summed across this packet. The listed top-level engine phases are additive; some diagnostic subtimers overlap them and are intentionally omitted. Tree++ relation-only time is not measured separately.