Digital Justice, Interrupted

The standard explanations are inadequate training and insufficient funding. Both have been addressed repeatedly, and neither fixed anything. Our research question became: what if the technology is not failing the institution, but the institution is rejecting the technology?

We analysed the system using IAD and Resilience Engineering, working from policy documents, court data, and published studies. The answer was not a broken feature. It was a structural pattern that no amount of software could fix.

The formal system was designed for judges, lawyers, clerks, and litigants, but largely without them. Policymakers and vendors built for an abstract user who follows mandated procedure. The real arena holds eight actor groups, and none of them optimize for the same outcome: policymakers want compliance, judges want discretion, staff want manageable workloads, litigants want resolution, and lawyers, inconveniently, profit when the system stays complex.

We mapped the case filing journey lane by lane, separating what humans do from what the IT system does. Nearly every breakdown sits at a handoff between lanes: where a rigid validation rule meets a messy human document, where a failed notification meets a litigant who never knew to check.

These adaptations are intelligent. They are also the reason nothing gets fixed: because the workarounds keep cases moving, the system appears to work, so its design defects never surface in any official metric. The improvisation masks the brittleness. Remove the exceptional effort of clerks and judges, and the whole thing collapses, which is exactly what happened during COVID lockdowns.

Backlog creates time pressure. Published time-use estimates suggest hearings compress to five or ten minutes. Rushed hearings produce weak judgments, an estimated 15 to 25% of which come back as appeals and remands. Remanded cases rejoin the backlog, which grows faster than courts can dispose of it. That is the Backlog Amplification loop, and it explains the central paradox: every conventional fix, more judges, more computers, more funding, pushes against a loop that regenerates the problem.

A judge who spends half the day on clerical coordination is not slow. The judge is doing two jobs, and only one of them requires a judge. If procedural time could fall to 10 or 15%, roughly a third of every judge's day returns to actual adjudication, without hiring anyone. Time pressure drops, hearings lengthen, judgment quality rises, remands fall, and the dominant loop starts running backwards.

That conditional is doing honest work: this is the hypothesis the diagnosis points to, not a measured result.

Every pillar of PAWS is a legitimized version of something practitioners already do. The misscan trick becomes a transparent validation flow with a real correction window. The phone-call search becomes search by name with automatic status updates. The paper shadow disappears only when digital becomes faster than paper, never by mandate. And AI drafts judgment structure and finds precedents, while the judge makes every decision. The clerk, never the judge.

The proposal that matters most is not software. It is a rule change: measure judges jointly on disposal rate and remand rate, so speed and quality stop being opposed incentives. The current metric, disposal alone, is the mechanism that converts backlog into five-minute hearings. Change the payoff rule and the loop weakens even before any code ships.

What the project changed for me: I stopped seeing interfaces as the unit of design. The eCourts portal is not badly designed because of its screens. It is badly designed because of its incentives, and no redesign of the screens touches those. The workarounds taught me that users are not the problem to be trained away. They are the system's most accurate documentation of what the design got wrong.