Researcher · 2026 · client work

Nothing is invented.

A redesign is a migration with opinions, and its failure mode is not an ugly page — it is a fee that is now wrong, a module that quietly vanished, an address that moved a digit. Nobody notices for months, and by then the old site is gone.

Content archaeology448 pagesTwo languagesNot deployed
Overview02

The first deliverable was an archive.

My role
Executed — the whole rebuild
Dates, from version control
4 – 23 September 2026
Scope
47 templates rendering 448 static pages
Tests
209 — 94 in the site package, 115 in config
State
Complete and building. Not deployed
What the archive recorded03

The site, transcribed.

39

Qualifications

41

Add-on electives

91

Job listings

Recovered by crawl; ids 1133–1223, no gaps

13

Centres, with postal addresses

22

Accreditations

28

Gallery photographs

The rules04

Four rules, each of which cost something.

The rule
  • All copy is newly written
  • Every number, name, address, module and date survives exactly
  • Nothing is invented — a missing value renders a marked placeholder
  • No fabricated social proof, no dark patterns
What it cost
  • Every page had to be argued from scratch
  • No rounding, no tidying, no “that looks like a typo”
  • Visible holes in a site being shown to its owners
  • The most persuasive blocks on a sales page shipped mostly empty

They held from the first commit to the last. The cost is the only evidence that they were real.

Social proof05

Twelve testimonials recorded. Four published.

Before
Twelve testimonials were captured from the old site. For eight of them the archive holds a name, a role and the theme of what the student said — not the words.
What forced it
Writing a quote from a theme is the most ordinary thing in this industry, and it is putting sentences in a real person's mouth.
The decision
Only the four whose words the archive actually holds are published. The other eight are seeded unpublished, and the seeder refuses to publish them.
What it changed
The decision survives the next person who runs the seeder in a hurry. The same rule is why no aggregate rating ships, and why the placement pages carry no outcome figures — a placement count, a median salary and a time-to-placement were all asked for and none was available.
When the only honest answer is an inference06

An ambiguity you cannot leave blank.

Two job listings published a salary ceiling with no period attached, while every sibling listing on the same board said “per month” explicitly.

The problem

Rendering it raw is not neutral

A reader supplies the missing word, and half of them supply the wrong one.

Dropping the field hides a published fact

Both defaults are a decision; neither is abstention.

The argument

The annual reading is arithmetically impossible

It works out to a monthly wage below every entry-level role on the same board — for department-manager and store-manager posts.

How it ships

Read as monthly, like its siblings

The inferred value is what the page shows.

The raw string is kept beside it

Stored in sourceSalary, so the inference stays auditable and anyone who disagrees can overturn it.

An inference is allowed when leaving the field empty would itself mislead. It must be labelled, carry its reasoning, and keep the raw source next to it — otherwise it is a guess with better manners.

What the research produced07

Findings, and the register of what is still missing.

Where it stands
  • Fourteen published self-contradictions resolved, each with a stated default and a reason
  • Two corrections written back into the archive itself — 91 job URLs that did resolve, and two application cohorts rather than one
  • A bank account number and IFSC published in plain text on the old payment page, removed rather than migrated
  • An app-store link whose publisher could not be verified, deliberately left out of the footer
  • 1,664 of 1,664 page strings translated, none orphaned, using the English sentence as the lookup key
Still open
  • Thirty numbered gaps were still open at handover
  • Twenty-two accreditation registration numbers: none is published anywhere
  • Written consent from seven named students before any of them appears on the rebuilt site
  • Verified placement outcomes: count, median salary, time-to-placement
  • Employer names for all 91 listings, every one published under a generic label

The whole rebuild — content archaeology, information architecture, the design system, the bilingual architecture and the engineering. Executed. Counts are quoted from the repository's handover document, each of which names the command that regenerates it. No outcome is claimed, because the site has not been deployed.

All case studiesAcademy of Retail ExcellenceThe illustrated documentWork with me

B-SIDEOff the record
The archive before the redesign — case study — Ashish Narayan