"Fixing the Most Dangerous Dam in the World" — Practical Engineering

Why this is in the vault

Canonical Grady Hillhouse exposition on Mosul Dam — a structure built on dissolving gypsum where the only "fix" is continuous never-ending grouting at half-a-billion-dollar scale. Filed for the load-bearing kernel: this is an exceptionally clean real-world instance of the chronic-maintenance-as-only-feasible-fix archetype, where the proper solution (cutoff wall, downstream Badouch dam) is permanently out of reach for capital/political reasons and the operating org must instead institutionalize a forever-grouting loop. Direct analog to the COO-harness pattern of continuous low-level remediation (skill drift, memory conflicts, channel-state hygiene) where the "proper fix" — a single-author rewrite of the entire harness — is not feasible, so the right move is to industrialize the maintenance loop.

Episode summary

Mosul Dam in northern Iraq was built in the 1980s on a foundation of gypsum — a sedimentary rock ~200x more soluble in water than limestone. Since the reservoir first filled, water has continuously dissolved the foundation, requiring 24/7 grouting (injecting sand/cement/bentonite slurry into bore holes) to maintain structural integrity. The US Army Corps of Engineers called it "the most dangerous dam in the world" in 2006. A 3-year, $500M+ rehabilitation project finished in 2019 (5,000+ bore holes, 250 mi of drilling, 50,000 cu yd of grout) brought the dam back from acute crisis to a sustainable maintenance posture — but the underlying problem (dissolving foundation) is unchanged, so the grouting must continue forever.

Key arguments / segments

• [00:00:00] Cold open: 800 L/sec seepage measured one year after reservoir filled — enough to fill an Olympic pool every hour. Once the dissolution feedback loop starts it's "pretty hard to stop."
• [00:01:01] Stakes framing: full breach would wipe towns off the Tigris and partially submerge Baghdad hundreds of miles downstream. US Army Corps called it "unequivocally the most dangerous dam in the world" in 2006.
• [00:02:00] Site overview: earthen embankment dam, 4-turbine hydro plant, service spillway + auxiliary spillway with fuse plugs. The dam itself is conventional; the foundation is the problem.
• [00:03:01] Foundation geology: gypsum dissolves in water (~200x more soluble than limestone). Consulting consensus: site selection was "fundamentally flawed."
• [00:04:00] Dissolution is non-linear/positive-feedback (more dissolution → more space → more flow → more dissolution). Internal-erosion-filter fixes don't work because dissolved gypsum can't be filtered out.
• [00:05:01] Original design considered bentonite blanket and cutoff wall but settled on a grout curtain (same approach as Teton Dam) — political pressure pushed completion before grouting was done, with the plan to keep grouting via the Gallery Tunnel forever.
• [00:06:02] Sink holes started appearing downstream and migrating toward the dam — surface evidence of subsurface caverns. The grouting program has been continuous since construction.
• [00:07:00] Iraqi backup plan: half-built Badouch Dam downstream to catch a Mosul breach. Stalled by geopolitics; never finished.
• [00:08:00] 2003 US-led invasion → US Army Corps assessment → "astronomical" risk classification.
• [00:09:00] First US intervention was a near-comedy of procurement failure: a contractor delivered concrete-mixing plants instead of grout-mixing plants and the review committee missed it; equipment didn't fit the gallery tunnels; staff received "a few weeks of training." Most of 21 contracts spent "millions of dollars for almost no benefit."
• [00:11:00] ISIS seized Mosul Dam for 8 days in August 2014, halting grouting and looting equipment. Even short-term neglect was acutely dangerous.
• [00:13:00] Iraqi Minister of Water Resources publicly downplayed risk as "one in a thousand" — multiplied by Corps fatality estimate that's 500-1500 expected annual fatalities, "nowhere in the world would anyone consider that acceptable."
• [00:14:01] Lt Gen Sean McFarland: "If this dam were in the United States, we would have drained the lake behind it." This is the load-bearing one-line framing for the whole video.
• [00:15:00] 2016-2019 Italian-led rehab project: helicopter-inserted engineers operating near front lines; permanent camp eventually established; modern grout infrastructure installed; sophisticated computer system tracking every bore hole's pressure/depth/mix/flow.
• [00:17:01] Outcome: 98% of post-grout permeability tests below 3 lujons. Project won 2022 Deep Foundations Institute award. Dam loses "most dangerous" title.
• [00:18:00] The kicker: rehab was a "half a billion dollar band-aid." Grouting has never been considered a permanent solution — gypsum is still soluble, reservoir is still pushing water through it. A permanent fix (cutoff wall) was estimated at $3-5B in 2018 and likely won't be funded.
• [00:20:01] Sponsor segment: Nebula / Real Engineering "The Anatomy of" CT-scanner series.

Notable claims

• [00:00:00] 800 L/sec foundation seepage one year after reservoir fill (verifiable, sourced via Army Corps of Engineers).
• [00:04:00] Gypsum is roughly 200x more soluble in water than limestone (verifiable, basic geochemistry).
• [00:09:00] Concrete-plant-vs-grout-plant procurement failure is a remarkable institutional-failure anecdote — "Millions of dollars had been spent on 21 contracts for almost no benefit to the dam."
• [00:15:00] Italian rehab: 5,000+ bore holes, 400 km / 250 mi of total drilling, 41,000 m³ / 50,000 yd³ of grout injected over 3 years for ~$500M.
• [00:18:00] Permanent fix (cutoff wall) estimated at $3-5B in 2018. Effectively unfundable; grouting forever is the operating model.

Guests

• Grady Hillhouse — civil engineer and host of Practical Engineering. Solo presenter; no guests this episode.

Mapping against Ray Data Co

Medium — chronic-maintenance-as-only-feasible-fix archetype. Direct analog to the harness-engineering posture where the proper solution (single-author rewrite of CLAUDE.md + memory files + skills) is not realistically fundable in attention-capital, so the right move is to industrialize the maintenance loop. Mosul Dam tells you the maintenance loop can in fact work — but it requires: (1) modern instrumentation tracking every "bore hole" (analog: every skill invocation, every memory conflict, every channel-state event), (2) explicit training-the-staff phase so the original engineers don't become the perpetual single point of failure, and (3) accepting that this is the operating model forever, not a transition state to a "real" fix. The harness-engineering Ch 2 precedence work ([[~/rdco-vault/06-reference/2026-05-18-agentway-harness-engineering-claude-code-design-guide.md]]) is structurally similar to the grout-program-instrumentation phase: not a rewrite, just better visibility into where the existing structure is leaking.

Secondary mapping: the Iraqi Minister's "one in a thousand" handwave maps to overconfidence-vs-evidence in [[feedback_calibrate_overconfidence]] — official statements that sound calibrated until you compute the actual implied expected loss.

Related

• [[feedback_calibrate_overconfidence]] — official-statement-vs-implied-loss anti-pattern
• [[~/rdco-vault/06-reference/2026-05-18-agentway-harness-engineering-claude-code-design-guide.md]] — maintenance-loop-as-operating-model
• Other Practical Engineering dam videos already in vault: [[2026-04-20-practical-engineering-sawing-a-dam-in-half]], [[2026-04-20-practical-engineering-taum-sauk-dam-failure]], [[2026-04-20-practical-engineering-why-no-short-arch-dams]], [[2026-04-21-practical-engineering-teton-dam-failure]] (the Teton Dam grout-curtain comparison referenced at 5:01)