openengineer/examples/survey-retaining-wall-failure.md

# Example: Retaining Wall Failure Survey

**Status:** Draft
**Phase:** The Bedrock Phase

## What This Example Demonstrates

The Survey practice (OE-0005): the structured assessment that serves as organized input to engineering understanding, distinct from understanding itself.

## The Observation

Between 2005 and 2015, a review of 47 retaining wall failures in transportation infrastructure was conducted. Each failure was documented with: the wall type (gravity, cantilever, anchored, mechanically stabilized earth), the failure mode (sliding, overturning, bearing capacity, internal failure, drainage-related), the soil conditions at the time of failure, the design assumptions, and whether the failure was related to a condition that was known at design time or one that developed during service. The raw data from these 47 cases, taken individually, tells a practitioner almost nothing useful. A single failure is an anecdote. Forty-seven failures, organized into a survey, reveal patterns that no single case could reveal.

## The Survey Structure

Mapping to OE-0005's five required sections, the retaining wall failure survey was structured as follows:

1. **Scope:** Retaining wall failures in transportation infrastructure, 2005-2015, 47 cases across four wall types (gravity, cantilever, anchored, mechanically stabilized earth). The scope was defined before data collection began, preventing the survey from expanding uncontrollably or shifting focus mid-study.

2. **Observations Collected:** Failure mode classification for all 47 cases. Soil condition data from post-failure investigations. Design documents and as-built records. Maintenance history. Rainfall and groundwater records for the 12 months preceding each failure. Each observation was collected using a consistent protocol, ensuring that the data from case 1 was comparable to the data from case 47.

3. **Patterns Identified:** (a) Drainage-related failures accounted for 62% of all cases across all wall types — this was the dominant pattern. (b) Mechanically stabilized earth (MSE) walls had the lowest failure rate per kilometer of installation, but their failures tended to be sudden and catastrophic (reinforcement strip rupture) rather than gradual. (c) 71% of failures involved a condition that existed at design time but was not accounted for (typically, groundwater level higher than assumed in design). These patterns emerged only after the observations were organized; they were not visible in any individual case file.

4. **Gaps Flagged:** Insufficient data on long-term drainage material degradation rates. No standardized post-failure investigation protocol — data quality varied significantly between cases. Limited data on walls that did NOT fail (survivorship bias in the dataset). These gaps are as important as the patterns, because they tell the practitioner where the survey's conclusions are weak and where future observation is needed.

5. **Relationship to Prior Surveys:** Extended a 1998 review by Dunnicliff (23 cases, 1980-1997). The 62% drainage-related failure rate is consistent with Dunnicliff's finding of 58%, reinforcing the pattern across a longer time period and larger dataset. This connection to prior work transforms the survey from an isolated study into a link in a chain of accumulating knowledge.

## Survey Is Not Understanding

The survey identifies patterns but does not explain them. Understanding — why drainage failures are so prevalent, why design-phase groundwater assumptions are so often wrong, what the mechanism is — requires the engineer to engage with the survey's findings and construct a contextualized model. The survey is the organized input; understanding is what the practitioner builds from it. This distinction matters because a survey without a practitioner's engagement is inert data. It becomes engineering knowledge only when a practitioner uses it to form understanding, verify that understanding against reality, and preserve the resulting context.

A survey that sits in a filing cabinet or a database is not engineering knowledge. It is potential knowledge, waiting for a practitioner to engage with it. The practitioner who reads the 62% drainage failure rate and asks "why?" has begun the transition from survey to understanding. The practitioner who then investigates the specific drainage mechanisms, tests the hypothesis against new observations, and records the reasoning has completed the transition. The survey enabled this process but did not replace it.

## Connection to the Process Chain

This survey connects to the foundational process (OE-0001): the observations were collected (step 1), patterns were recognized (step 2), and the findings can be translated into engineering language (step 3). The next step — verification against reality (step 4) — would require implementing the survey's findings in new designs and measuring whether drainage-related failure rates decrease. This is how a survey feeds the engineering process: it provides organized observations and identified patterns that a practitioner can use as input to understanding, which then informs decisions, which are then verified against new observations.

The process is cyclic, not linear. The verification step generates new observations that may feed back into a future survey, refining or challenging the original patterns. The 62% drainage failure rate found in this survey may change as new data is collected — and that is the point. The survey is a snapshot of understanding at a point in time, structured so that future practitioners can build on it, refine it, or replace it with better data. It is part of the engineering fabric, not a final answer.

## Self-Fading Assessment

This example builds a bridge from the abstract definition of a survey (OE-0005) to a concrete case where structured data collection revealed patterns invisible in individual cases. The reader has crossed this bridge when they can distinguish between a survey (organized observations and identified patterns) and understanding (the practitioner's contextualized model of why those patterns exist), and when they instinctively structure their own multi-case assessments using the five required sections. When that distinction and that structure are automatic, this example has faded, and the practice remains.