Validra

AI Performance Report

How our AI tools performed

A simple look at how fast and reliably Validra's AI tools respond when many people use them at the same time

What this report shows

This is a health-check of Validra's AI tools. We simulated several people using each tool at the same time (just like a busy day) and measured two simple things: did it work, and how long did people wait.

Each section below covers one tool. Green numbers are good. A short note explains what the tool does and which numbers matter most. New to the terms? See “How to read the numbers” at the bottom of the page.

At a glance

All tools compared across every measure in one graph. Each group of bars is one measure; each colour is a tool. Bar heights are scaled within each measure (units differ) and the exact value is shown on top.

Report MasterAI EditorAI Chunking

1m 29s

18s

10s

Response time

1m 37s

2m 42s

29s

Total wait

42.5k

1.9k

16.0k

Tokens used

100%

Reliability

435

152

587

AI speed

Report Master — Automatic Report Writing

What it does: turns your study data into a complete, written report — automatically. What to look for: Success Rate shows how often a report was produced without errors, and the Avg times show the typical wait to generate one.

Report Master mixed | 1 request(s) | 2026-06-02T06:02:01+00:00 | model: qwen2.5:14b

Model

qwen2.5:14b

Requests

Completed

Failed

Success Rate

100.0%

Avg Client

1m 37.1s

Avg Server

1m 29.4s

Avg Tokens/sec

580.1

Total Tokens

42,475

LLM Calls

Client-Side Latency

Generate click → Download ready

Min

1m 37.1s

p50

1m 37.1s

p75

1m 37.1s

p85

1m 37.1s

p90

1m 37.1s

p95

1m 37.1s

p99

1m 37.1s

Max

1m 37.1s

Mean: 1m 37.1s | Std Dev: 0.0s

Server-Side Duration

From API response

Min

1m 29.4s

p50

1m 29.4s

p75

1m 29.4s

p85

1m 29.4s

p90

1m 29.4s

p95

1m 29.4s

p99

1m 29.4s

Max

1m 29.4s

Mean: 1m 29.4s | Std Dev: 0.0s

Tokens/sec (higher = faster)

Min

580.1

p50

580.1

p75

580.1

p85

580.1

p90

580.1

p95

580.1

p99

580.1

Max

580.1

Per-Request Breakdown

#	Status	Client	Server	Tokens	LLM Calls	Tok/s	Job ID
1	completed	1m 37.1s	1m 29.4s	42,475	26	580.1	24612c9b-4cba-424d-b293-4d0136730e0e

AI Editor — Edit & Generate Text

What it does: edits protocol text from plain instructions (for example, “change the wording to past tense”) and can generate new content. What to look for: the typical time each edit and each generation took, and how many succeeded.

AI Editor mixed | 2 test(s) | http://100.107.115.34 | 2026-06-02T06:02:01+00:00 | model: phi3 (vision: qwen2.5vl:7b)

Model

phi3 (vision: qwen2.5vl:7b)

Tests

Edit Passed

2/2

Generate Passed

2/2

Avg Edit LLM

17.4s

Avg Generate LLM

18.1s

Avg Total Time

2m 41.9s

Avg Edit Tok/s

229.8

Avg Generate Tok/s

203.1

Total Tokens

1,879

Edit LLM Time

"change tense to past"

Min

16.9s

p50

16.9s

p75

17.8s

p85

17.8s

p90

17.8s

p95

17.8s

p99

17.8s

Max

17.8s

Mean: 17.4s | Std Dev: 0.5s

Generate LLM Time

"add table same as in image"

Min

18.1s

p50

18.1s

p75

18.1s

p85

18.1s

p90

18.1s

p95

18.1s

p99

18.1s

Max

18.1s

Mean: 18.1s | Std Dev: 0.0s

Edit Tokens/sec

higher = faster

Min

203.0

p50

203.0

p75

256.7

p85

256.7

p90

256.7

p95

256.7

p99

256.7

Max

256.7

Mean: 229.8 | Std Dev: 26.8

Generate Tokens/sec

higher = faster

Min

202.9

p50

202.9

p75

203.3

p85

203.3

p90

203.3

p95

203.3

p99

203.3

Max

203.3

Mean: 203.1 | Std Dev: 0.2

Per-Test Breakdown

#	Setup	Edit	Edit LLM	Edit Tokens	Edit Tok/s	Generate	Gen LLM	Gen Tokens	Gen Tok/s	Total	Error
1	2m 1.7s	passed	17.8s	493	203.0	passed	18.1s	498	202.9	2m 37.6s	-
2	2m 11.2s	passed	16.9s	390	256.7	passed	18.1s	498	203.3	2m 46.2s	-

Data Analysis — Understand Your Data

What it does: reads an uploaded data file and prepares it for analysis. What to look for: Avg Analysis and Avg Processing are the usual wait times; Success Rate is how often it finished cleanly.

Data Analysis — AI Mode

No Data Analysis results found.

AI Chunking — Reading Long Documents

What it does: splits a long document into smaller sections the AI can read and understand. What to look for: the typical time to process a document, and how often it completed.

AI Chunking mixed | 2 run(s) | http://100.107.115.34 | 2026-06-02T06:02:01+00:00 | model: phi3

Model

phi3

Runs

Completed

Failed

Success Rate

100.0%

Avg Chunking

10.3s

Avg Total Time

28.6s

Avg Tokens/sec

782.8

Total Doc Tokens

16,046

Chunking Time

Process Document → Done

Min

10.2s

p50

10.2s

p75

10.3s

p85

10.3s

p90

10.3s

p95

10.3s

p99

10.3s

Max

10.3s

Mean: 10.3s | Std Dev: 0.1s

Tokens/sec

doc tokens ÷ chunking time

Min

778.9

p50

778.9

p75

786.6

p85

786.6

p90

786.6

p95

786.6

p99

786.6

Max

786.6

Mean: 782.8 | Std Dev: 3.9

Per-Run Breakdown

#	Status	Setup	Chunking	Total	Doc Tokens	Tok/s	Error
1	completed	19.0s	10.3s	29.3s	8,023	778.9	-
2	completed	17.7s	10.2s	27.9s	8,023	786.6	-

How to read the numbers

Success rate — out of every attempt, how many finished correctly. Higher is better — 100% means nothing failed.

Typical wait time (Avg / Average) — the usual time a person waited for a result. Shown in seconds (s) or minutes (m). Lower is better.

p50, p85, p99 — a quick way to show the range of waits. p50 = half of people waited less than this. p85 = 85% waited less. p99 = almost everyone (99%) waited less. Lower is better.

Tokens / second — roughly how fast the AI writes its answer (a “token” is about ¾ of a word). Higher is better.

Setup — the time to open the page and get ready, before the AI work even begins.

Model — the specific AI engine used for that tool (for example, “qwen2.5:14b” or “phi3”). Different tools can use different engines.