Assessment SOP — Pyze Execution Intelligence

Regulatory Context Assessment quality is the primary driver of signal detection and regulatory judgment. Subject to audit under EMA GVP Module VI.

Why this SOP is audit-grade Every step below is captured at the DOM element level (button labels, field IDs, section panels) — not as screenshot coordinates. Every metric is derived from 2,656 real events across 793 cases, with confidence and case evidence visible per step.

Step Journey

Click a task to see the application journey on the same case, then click an application to drill into the execution detail Pyze captured for it.

Level 2 · Application Journey

Open Adverse Event Review — 10 apps touched within ±15 min of this task on the same case

Case counts per app are not mutually exclusive — a case that touched both Veeva and Phobos in this window is counted in both rows. Overlap reflects real cross-app work, not double-counted cases.

Level 3 · Execution Detail

Veeva Safety

DOM-level capture

Pyze JS captures every element interaction — button labels, field IDs, section panels.

pyzeClick <span> "Adverse Events" 964×

pyzeClick <span> "Complete" 549×

pyzeClick <span> "Narrative" 400×

pyzeClick <a> "Show more" 361×

pyzeClick <span> "Products" 294×

pyzeClick <span> "Patient" 224×

pyzeClick <span> "Submissions & Distributions" 224×

pyzeClick <span> "Documents" 206×

Level 3 · Execution Detail

Phobos

DOM-level capture

Pyze JS captures every element interaction — button labels, field IDs, section panels.

pyzeClick <button> "Proceed" 152×

pyzeClick <select> "Select Attachment description · Legal · Legal Pre-Sca" 129×

pyzeClick <button> "Save Comments" 113×

pyzeClick <button> "Advance" 92×

inputld <select> "Select Attachment description · Legal · Legal Pre-Sca" 74×

pyzeClick <button> "Send to process" 70×

pyzeClick <select> "Select Compound · - sitagliptin phosphate Tablet (G" 62×

pyzeClick <div> "Edit info and/or add comments · × · Compound · Select C" 38×

Level 3 · Execution Detail

Microsoft Word

Document-level capture

Task Mining captures the specific file open during work — file names often encode case context.

file V-Narrative 552979 (English) (26689105_0_2_348832) 1076×

file V-Narrative 543823 (English) (26544362_0_2_343646) 895×

file V-Narrative 549707 (English) (26645831_0_1_347236) 556×

file V-Narrative 549550 (English) (26643764_0_1_347593) 526×

file V-Narrative 534252 (English) (26399499_0_2_343143) 301×

file V-Narrative 533936 (English) (26397592_0_1_338360) 264×

file V-Narrative 542512 (English) (26522671_0_2_344761) 255×

file V-Narrative 540358 (English) (26494608_0_2_342069) 228×

Level 3 · Execution Detail

Acrobat

App-level capture

Only application name and dwell time available. Direct instrumentation would unlock deeper detail.

4,151 granular events captured, but no DOM, document, or message-level metadata is available for this application. Instrumenting this app directly with Pyze would unlock element-level detail.

Level 3 · Execution Detail

Microsoft Teams

App-level capture

Only application name and dwell time available. Direct instrumentation would unlock deeper detail.

2,848 granular events captured, but no DOM, document, or message-level metadata is available for this application. Instrumenting this app directly with Pyze would unlock element-level detail.

Level 2 · Application Journey

Categorize Case — 10 apps touched within ±15 min of this task on the same case

Level 3 · Execution Detail

Phobos

DOM-level capture

Pyze JS captures every element interaction — button labels, field IDs, section panels.

pyzeClick <button> "Proceed" 462×

pyzeClick <select> "Select Attachment description · Legal · Legal Pre-Sca" 451×

pyzeClick <button> "Save Comments" 328×

inputld <select> "Select Attachment description · Legal · Legal Pre-Sca" 310×

pyzeClick <button> "Advance" 216×

pyzeClick <button> "Send to process" 208×

pyzeClick <select> "Select Compound · - sitagliptin phosphate Tablet (G" 152×

pyzeClick <a> "Cases" 145×

Level 3 · Execution Detail

Veeva Safety

DOM-level capture

Pyze JS captures every element interaction — button labels, field IDs, section panels.

pyzeClick <span> "Complete" 539×

pyzeClick <a> "Show more" 180×

pyzeClick <span> "Adverse Events" 125×

pyzeClick <span> "Narrative" 116×

pyzeClick <li> "Narrative" 105×

pyzeClick <li> "Details" 90×

pyzeClick <span> "Approval Complete" 85×

pyzeClick <span> "Products" 82×

Level 3 · Execution Detail

Microsoft Word

Document-level capture

Task Mining captures the specific file open during work — file names often encode case context.

file V-Narrative 538518 (English) (26459147_0_1_344628) 198×

file V-Narrative 542172 (English) (26517690_0_2_345412) 181×

file V-Narrative 542162 (English) (26518018_0_1_342783) 170×

file V-Narrative 538589 (English) (26461195_0_1_346748) 160×

file V-Narrative 527377 (English) (26286859_0_1_339572) 113×

file V-Narrative 525025 (English) (26253159_0_3_342205) 109×

file V-Narrative 537919 (English) (26452534_0_2_341686) 99×

file V-Narrative 541519 (English) (26506626_0_2_342362) 75×

Level 3 · Execution Detail

Acrobat

App-level capture

Only application name and dwell time available. Direct instrumentation would unlock deeper detail.

1,357 granular events captured, but no DOM, document, or message-level metadata is available for this application. Instrumenting this app directly with Pyze would unlock element-level detail.

Level 3 · Execution Detail

Microsoft Teams

App-level capture

Only application name and dwell time available. Direct instrumentation would unlock deeper detail.

1,141 granular events captured, but no DOM, document, or message-level metadata is available for this application. Instrumenting this app directly with Pyze would unlock element-level detail.

Level 2 · Application Journey

Review Product Information — 10 apps touched within ±15 min of this task on the same case

Level 3 · Execution Detail

Veeva Safety

DOM-level capture

Pyze JS captures every element interaction — button labels, field IDs, section panels.

pyzeClick <span> "Products" 541×

pyzeClick <span> "Complete" 368×

pyzeClick <span> "Adverse Events" 348×

pyzeClick <span> "Narrative" 278×

pyzeClick <a> "Show more" 226×

pyzeClick <span> "Patient" 179×

pyzeClick <span> "Documents" 178×

pyzeClick <span> "Submissions & Distributions" 173×

Level 3 · Execution Detail

Phobos

DOM-level capture

Pyze JS captures every element interaction — button labels, field IDs, section panels.

pyzeClick <button> "Proceed" 108×

pyzeClick <button> "Save Comments" 80×

pyzeClick <select> "Select Attachment description · Legal · Legal Pre-Sca" 64×

pyzeClick <button> "Advance" 62×

pyzeClick <button> "Send to process" 61×

pyzeClick <select> "Select Compound · - sitagliptin phosphate Tablet (G" 40×

pyzeClick <a> "Cases" 38×

pyzeClick <select> "Select Report Type · Legal · Literature · Medical Recor" 34×

Level 3 · Execution Detail

Microsoft Word

Document-level capture

Task Mining captures the specific file open during work — file names often encode case context.

file V-Narrative 552979 (English) (26689105_0_2_348832) 932×

file V-Narrative 533164 (English) (26378749_0_4_342805) 791×

file V-Narrative 549707 (English) (26645831_0_1_347236) 499×

file V-Narrative 533533 (English) (26384397_0_2_337752) 380×

file Document1 303×

file V-Narrative 514031 (English) (26063743_0_1_327856) 264×

file Narrative 497939 (English) 232×

file V-Narrative 549428 (English) (26642399_0_2_349187) 213×

Level 3 · Execution Detail

Acrobat

App-level capture

Only application name and dwell time available. Direct instrumentation would unlock deeper detail.

3,201 granular events captured, but no DOM, document, or message-level metadata is available for this application. Instrumenting this app directly with Pyze would unlock element-level detail.

Level 3 · Execution Detail

Microsoft Teams

App-level capture

Only application name and dwell time available. Direct instrumentation would unlock deeper detail.

1,719 granular events captured, but no DOM, document, or message-level metadata is available for this application. Instrumenting this app directly with Pyze would unlock element-level detail.

Standard Flow

Representative application journey — the dominant app-to-app transitions observed in the pilot data. Tier badge shows how deeply Pyze captures each application.

Veeva Safety

DOM

12,762 events · 724 cases

Phobos

DOM

3,577 events · 332 cases

Veeva Safety

DOM

12,762 events · 724 cases

Microsoft Word

Document

618 events · 122 cases

Veeva Safety

DOM

12,762 events · 724 cases

Dominant transitions chained from real observations. Specific case journeys appear as variants below. Use the Step Journey above to drill into any task.

Step Variants

Distinct task-sequence patterns observed across cases in this step. Click a variant on the left to see its task sequence, business impact, and recommended remediations on the right.

Reference Assessment (step-level) Reference Path 222 cases · 12 analysts

Task Sequence

1. Open Adverse Event Review

Automated Explanation rule-based

This is the mined reference path — 222 instances across 12 analysts follow this exact sequence. Treat as the proposed Standard Operating Procedure for Assessment (step-level).

Genuine Complexity Complex-Case Execution Path 135 cases · 8 analysts

Task Sequence

1. Categorize Case

Automated Explanation rule-based

Genuine Complexity — this sequence differs from the reference path (1 sub-steps vs 1 in reference) but shows no strong pain-point signals: adherence score 0 vs reference 100.0 (delta +0.0). Likely a legitimate case-complexity path — not an automation candidate, but worth cataloging.

Time

1,849 hrs/yr

Cost

$138,675/yr

At 1,000

108,765 hrs/yr

Reviewer Sweep Reviewer Verification Pass 60 cases · 11 analysts

Task Sequence

1. Review Product Information

Automated Explanation rule-based

Reviewer Sweep — a viewing-heavy pattern with 3.2 avg sub-steps and zero edits captured. Consistent with a downstream reviewer re-opening the task to verify earlier work rather than progressing new data. 11 analysts across 60 cases.

Backfill Loop Mid-Flow Data Backfill 34 cases · 7 analysts · 3× same-task rework

Task Sequence

1. Open Adverse Event Review → 2. Categorize Case

Automated Explanation rule-based

Backfill Loop — instances revisit the same activity 3.0× on average, consistent with mid-flow detours to fetch or enter data that wasn't ready at the right step. Ordering issue, not an app issue — 7 analysts across 34 cases. Suggests upstream data readiness or UX sequencing is at fault.

Backfill Loop Mid-Flow Data Backfill 31 cases · 6 analysts · 3× same-task rework

Task Sequence

1. Open Adverse Event Review → 2. Review Product Information

Automated Explanation rule-based

Backfill Loop — instances revisit the same activity 3.0× on average, consistent with mid-flow detours to fetch or enter data that wasn't ready at the right step. Ordering issue, not an app issue — 6 analysts across 31 cases. Suggests upstream data readiness or UX sequencing is at fault.

Reviewer Sweep Reviewer Verification Pass 21 cases · 5 analysts

Task Sequence

1. Review Product Information → 2. Open Adverse Event Review

Automated Explanation rule-based

Reviewer Sweep — a viewing-heavy pattern with 8.2 avg sub-steps and zero edits captured. Consistent with a downstream reviewer re-opening the task to verify earlier work rather than progressing new data. 5 analysts across 21 cases.

Time

9 hrs/yr

Cost

$675/yr

At 1,000

529 hrs/yr

Reviewer Sweep Reviewer Verification Pass 15 cases · 4 analysts · 2× same-task rework

Task Sequence

1. Open Adverse Event Review → 2. Review Product Information → 3. Open Adverse Event Review

Automated Explanation rule-based

Reviewer Sweep — a viewing-heavy pattern with 7.9 avg sub-steps and zero edits captured. Consistent with a downstream reviewer re-opening the task to verify earlier work rather than progressing new data. 4 analysts across 15 cases.

Time

1,919 hrs/yr

Cost

$143,925/yr

At 1,000

112,882 hrs/yr

Reviewer Sweep Reviewer Verification Pass 14 cases · 7 analysts · 2× same-task rework

Task Sequence

1. Review Product Information → 2. Categorize Case

Automated Explanation rule-based

Reviewer Sweep — a viewing-heavy pattern with 7.1 avg sub-steps and zero edits captured. Consistent with a downstream reviewer re-opening the task to verify earlier work rather than progressing new data. 7 analysts across 14 cases.

Reviewer Sweep Reviewer Verification Pass 9 cases · 5 analysts · 2× same-task rework

Task Sequence

1. Categorize Case → 2. Open Adverse Event Review

Automated Explanation rule-based

Reviewer Sweep — a viewing-heavy pattern with 9.1 avg sub-steps and zero edits captured. Consistent with a downstream reviewer re-opening the task to verify earlier work rather than progressing new data. 5 analysts across 9 cases.

📄

Detailed Step Reference

Full DOM element detail + case evidence for every step (3 total) — mostly for BA and RPA-developer use

Open Adverse Event Review

Open Adverse Events Section None

856 observations 459 cases 12 analysts Median 33.0s 100% high confidence

Categorize Case

Enter Case Type None

702 observations 297 cases 8 analysts Median 3.0s 100% high confidence

Review Product Information

Open Products Section None

479 observations 261 cases 14 analysts Median 55.0s 100% high confidence

Top Application Transitions

Top app-to-app hops observed during this stage. Each pair is a single swivel-chair move, ranked by frequency.

Rank	From App		To App	Cases	Frequency	Share
1	Veeva Safety DOM	→	Phobos DOM	240	531	26.1%
2	Phobos DOM	→	Veeva Safety DOM	206	433	21.3%
3	Microsoft Word Document	→	Veeva Safety DOM	113	327	16.1%
4	Veeva Safety DOM	→	Microsoft Word Document	105	284	14.0%
5	Acrobat App	→	Veeva Safety DOM	84	236	11.6%
6	Veeva Safety DOM	→	Acrobat App	84	220	10.8%

Path Variants

Top 10 full application journeys observed through this stage (consecutive duplicates collapsed). Shows the actual end-to-end paths cases take — not just single transitions.

Rank	Hops	Application sequence	Cases	Share of cases
1	1	Veeva Safety DOM	402	50.7%
2	1	Phobos DOM	65	8.2%
3	2	Veeva Safety DOM → Phobos DOM	28	3.5%
4	3	Veeva Safety DOM → Phobos DOM → Veeva Safety DOM	13	1.6%
5	3	Phobos DOM → Veeva Safety DOM → Phobos DOM	9	1.1%
6	4	Veeva Safety DOM → Phobos DOM → Veeva Safety DOM → Phobos DOM	8	1.0%
7	8	Veeva Safety DOM → Phobos DOM → Veeva Safety DOM → Phobos DOM → Veeva Safety DOM → Phobos DOM → Veeva Safety DOM → Phobos DOM	7	0.9%
8	7	Veeva Safety DOM → Phobos DOM → Veeva Safety DOM → Phobos DOM → Veeva Safety DOM → Phobos DOM → Veeva Safety DOM	6	0.8%
9	7	Phobos DOM → Veeva Safety DOM → Phobos DOM → Veeva Safety DOM → Phobos DOM → Veeva Safety DOM → Phobos DOM	6	0.8%
10	6	Veeva Safety DOM → Phobos DOM → Veeva Safety DOM → Phobos DOM → Veeva Safety DOM → Phobos DOM	5	0.6%

A "1-hop" path means the case stayed in a single application for this stage. Longer paths indicate cross-app work — those are the ones to investigate for swivel-chair reduction opportunities.

Applications Footprint

Where work actually happens during this stage. Cross-app time flags potential integration or UX opportunities.

Application	Capture tier	Events	Cases touched	Avg dwell per event	Total minutes in stage
Veeva Safety	DOM	1,914	713	0.26s	8.2 min
Phobos	DOM	742	297	2.74s	33.9 min

Drift Monitoring

Compares observed behavior against the published SOP. Drift signals feed compliance events and trigger automatic SOP refresh.

Drift level

High

Top path concentration

13.6%

Distinct transitions

Emerging steps

Unstable steps

Rare apps

Uneven adoption apps

Fragmentation signal This stage shows 46 distinct transitions with the top path at only 13.6% — a classic fragmentation signal. Monitor for continued drift.

Emerging activities (not in standard flow)

Activity	Frequency	Cases	Action
Select Adverse Events	399	265	Candidate for SOP inclusion · Promote to step →
Select Report Type	190	143	Candidate for SOP inclusion · Promote to step →
Enter MARRS Case ID	150	114	Candidate for SOP inclusion · Promote to step →
View Adverse Events	101	90	Candidate for SOP inclusion · Promote to step →

Rare apps (used on <15% of cases)

Application	Capture tier	Cases using	Events	% of cases	Action
Acrobat	App	95	460	12.0%	Shadow tool signal · Flag for investigation →
Microsoft Excel	Document	77	204	9.7%	Shadow tool signal · Flag for investigation →
Microsoft Outlook	Message	74	146	9.3%	Shadow tool signal · Flag for investigation →
collaboration.merck.com	App	56	121	7.1%	Shadow tool signal · Flag for investigation →
ONENOTE	Document	27	52	3.4%	Shadow tool signal · Flag for investigation →
gpteal.merck.com	DOM	23	50	2.9%	Shadow tool signal · Flag for investigation →
Quality Vault	DOM	18	28	2.3%	Shadow tool signal · Flag for investigation →
YoudaoDict	App	17	30	2.1%	Shadow tool signal · Flag for investigation →

Uneven adoption apps (used by minority of analysts)

Application	Analysts using	Total events	Adoption	Action
Microsoft Outlook	6	146	35.0%	Training or knowledge-transfer opportunity · Review users →
RIM Vault	3	59	18.0%	Training or knowledge-transfer opportunity · Review users →
ONENOTE	4	52	24.0%	Training or knowledge-transfer opportunity · Review users →
gpteal.merck.com	6	50	35.0%	Training or knowledge-transfer opportunity · Review users →
apps.powerapps.com	5	46	29.0%	Training or knowledge-transfer opportunity · Review users →
evocpr01.apps.americas.cf.merck.com	2	36	12.0%	Training or knowledge-transfer opportunity · Review users →

Continuous monitoring active This SOP reflects observed behavior through the 19-day pilot. In production, drift signals are checked hourly and published SOPs auto-refresh when consensus shifts. Drift events appear in the admin audit log with a diff against the previous version.

Assessment

Tasks in this Step

Publish to QualityDocs

Step Journey

Standard Flow

Step Variants

Open Adverse Event Review

Categorize Case

Review Product Information

Top Application Transitions

Path Variants

Applications Footprint

Drift Monitoring

Emerging activities (not in standard flow)

Rare apps (used on <15% of cases)

Uneven adoption apps (used by minority of analysts)