8020 in Everything
8020 in Everything

Brief: excel

Research brief for excel · how it was made.

Research brief — excel

FieldValue
Slugexcel
Primary query / seo_intentWhere Spreadsheet Mistakes Actually Concentrate
Template (A–E)C
Evidence tier (1–3)1 (Panko / EuSpRIG error research) + 2 (workbook review protocol)
Audience (one line)Analysts, managers, and ops people who live in workbooks - blunt risk talk, not an Excel tutorial
Public byline8020.in Editorial
ReviewerPractitioner pass against Panko / EuSpRIG framing
Date2026-07-20

1. Concentration claim (one sentence)

In Excel work that drives decisions, expensive failure concentrates in a few bottlenecks - untested formulas, mixed inputs and logic, silent bad lookups, broken copy-fills, missing reconciliations, and version chaos - not in equal time polishing every chart or learning every new function.

2. Hard anchors (2–5)

  1. Panko spreadsheet development experiments / reviews — Lab studies of individuals building spreadsheets show cell error rates commonly in the ~1%–5% range; a multi-study roll-up averages about 3.9% CER across hundreds of participants. https://panko.com/ssr/DevelopmentExperiments.html · synthesis: https://arxiv.org/pdf/1602.02601
  2. Panko (2016) arXiv / EuSpRIG tradition — At those rates, large workbooks are very likely to contain errors; individual inspectors in one experiment caught about 63% of seeded errors (teams of three ~83%). https://arxiv.org/pdf/1602.02601
  3. EuSpRIG horror stories — Reinhart–Rogoff coding error — Documented case where an AVERAGE range omitted countries (L30:L44 vs L30:L49), altering a widely cited growth result. https://eusprig.org/research-info/horror-stories/
  4. Fifteen-minute workbook risk walkthrough → ranked top-10 — personal protocol, not a lab instrument.

2b. Field 80/20 examples

Approx % claimField / contextSourceWhere
~1%–5% cell error rates (lab); ~3.9% average across studiesSpreadsheet development researchPanko site / 2016 reviewFormula testing bottleneck
Individual inspection caught ~63% of errors in one experimentError detection limitsPanko 2016 citing 1999 inspection workWhy “I looked at it” is weak
Famous omitted-range average (Reinhart–Rogoff)Real-world formula omissionEuSpRIG horror storiesSilent range / formula risk
Illustrative top-10 workbook registerPersonal protocolIllustrativeOnly-on-8020

Never invent “80% of Excel mistakes are VLOOKUP.”

3. Original observation (only-on-8020 seed)

Workbook risk walkthrough → ranked top-10 bottleneck register (critical formulas, input cells, lookups, reconciliations, file versions) — weekly attention on the ranked list, not equal feature learning.

4. Ignored majority (named)

Learning obscure functions before testing the model; dashboard cosmetics; equal cell-by-cell staring; “it opened without #REF! so it’s fine”; emailing File_FINAL_v7.xlsx as the control system.

5. Composite policy

ScenarioKeep as Illustrative?
Top-10 workbook register (ops forecast style)Yes

6. Vital few (bottleneck sections)

  1. Untested formulas / no inspection time
  2. Inputs mixed with calculations (hardcoded constants)
  3. Lookups and joins that fail silently
  4. Copy-fill / absolute-reference breaks
  5. No reconciliation before a decision
  6. Version chaos (shorter)

7. Device budget

Warning sign / Action today each section; ≤1 8020 move; viz: none

7b. Viz & misreads

Viz: none Misreads: 3 (no error means correct; more functions = safer; staring at every cell is thorough)

  1. Panko DevelopmentExperiments
  2. arXiv 1602.02601
  3. EuSpRIG horror stories
  4. Internal: risk-management, decision-making, data-science, testing

9. Sign-off

  • ☑ Brief complete
  • ☑ Not an Excel how-to curriculum
  • ☑ Reinhart–Rogoff framed as documented formula omission among other critique issues (EuSpRIG notes selection/weighting too - mention formula piece clearly)