Munjal Auto, turning an undocumented site into a compliance publishing operation

WitsCode treated this as an operations problem rather than a queue of tickets, on three principles.

Verify, never assume. Every claim that something is live is backed by a real check. Documents are opened and read before filing rather than trusted by filename, which is how one arriving as a board-meeting outcome turned out to be the quarterly results. Every referenced file path is validated against disk before release, and publishing is confirmed by fetching the live page from an independent network rather than our own.

Write down what we learn. Institutional memory belongs in the repository, not in one person's head. A running task log records every change with its commit and release, a developer runbook captures conventions, deploy mechanics, and every known trap, and a full handover document means the project can change hands without losing context.

Raise what does not add up. We publish exactly what is asked for and flag discrepancies rather than guessing. A regulatory pack contained the same circular twice under different names, confirmed by checksum. A tax pack shipped Form 15H twice with no Form 15G, which would have left shareholders under 60 with no form to download. Both were raised before publication rather than after.

The engineering work followed the same logic. Three client-reported defects all traced to one root cause, a download component that fails silently when handed the wrong props, so we fixed the instances and then documented the contract so it stops recurring. The orphaned image pipeline was resolved by removing the risk rather than the dependency: 161 of the 166 assets were recovered and backed up locally with a SHA-256 per file and two documented recovery routes, while the CDN itself was kept for its edge caching and modern image formats. That judgment paid off when a later request to change customer logos needed originals we could no longer download. The release pipeline was rebuilt from a per-file process loop into a single batched operation, taking an end-to-end release from roughly 90 seconds to 23, with genuine failure detection so a broken push now fails loudly.