04. Model routing — match capability to task difficulty without hiding failure

~22 min read. A lead AI engineer does not optimize a single number; they make the tradeoff visible, measured, and reversible.

Builds on 00-eli5.md. The dispatch board chooses model routes; the fuel ledger proves whether lower spend preserved the outcome.

---

What previous chapters solved before this pressure appears

The first three chapters taught us to count workflow cost, locate latency rooms, and reuse stable prompt work. What still breaks is sending every request through the same expensive lane because the team has not separated easy, hard, risky, and high-value work. Routing turns optimization into capability allocation, but a bad route can fail quietly.

The accumulated lesson is already visible in the taxi fleet. meter ticks expose money, the ETA call exposes perceived wait, the dispatch board exposes route choice, and the fuel ledger keeps those choices honest. This file adds the next constraint without forgetting the earlier ones: every optimization relieves one pressure and creates another that some subsystem must absorb.

What this file solves

This file starts from 80% of traffic moves to a small model, spend falls, and support resolution drops three points. It shows the concrete design move a lead engineer makes, the artifact to inspect, and the signal that proves the optimization helped instead of merely moving pain elsewhere.

The opening failure shows up in a concrete artifact

The failure is not abstract: 80% of traffic moves to a small model, spend falls, and support resolution drops three points. Here is the early artifact a reviewer can inspect.

A routing review starts with a confusion matrix by request segment.

Router slice for week 31:

Segment	Routed small	Small success	Escalation	Product risk
FAQ answer	84%	96%	3%	low
refund dispute	61%	81%	16%	medium
policy exception	44%	68%	27%	high
billing action	30%	72%	22%	high

A smart team might try to fix the most visible line in that artifact. That is tempting, and it is incomplete. The root cause is unmeasured difficulty classification: the route changed before the team knew which requests the small model could safely handle. So how do we save money without routing hard work into silent failure? This is the root-cause pivot: not a local metric problem, a boundary-and-pressure problem.

A tiny version exposes the whole mechanism

A toy request makes the pressure visible: one request looks fine in isolation, but after route mix, retries, context, or output length changes, the user-visible workflow changes. The smallest example is enough to show why the lever must be measured at the workflow boundary.

Rule: Use the cheapest route that meets the task’s quality, safety, latency, and business-value budget with measured confidence.

Why this rule exists. Model routing is the dispatch board deciding which cab fits the ride: a short local trip does not need the airport van, but a hazardous route should not be assigned to a bicycle. The fuel ledger matters because it shows whether the new pressure landed in cost, latency, memory, quality, or operator attention.

---

1) Classify difficulty, risk, and value before choosing a model

Start with the workflow, not the vendor feature. In Maya's review, the team takes one request and follows it from API ingress to model call, tool call, runtime behavior, and resource consequence. That cross-layer trace is the shortest path from symptom to lever. If the symptom is cost, the trace follows meter ticks. If the symptom is silence, it follows the ETA call. If the symptom is serving pressure, it follows the carpool lane and boot space.

user request
    │
    ▼
API gateway ── route/version ──► model/runtime
    │                              │
    │                              ├─ tokens / queue / KV / output
    │                              ▼
    └──────── outcome ◄──────── fuel ledger row

The counterintuitive part is that the most obvious metric can improve while the product gets worse. A smaller bill can hide more failed outcomes. Higher tokens/sec can hide longer queueing. A shorter prompt can hide missing evidence. The mechanism in this chapter is useful only when the trace keeps the relieved pressure and the newly created pressure in the same picture.

2) The route table before the model call

Picture the chapter as a pressure transfer, not a free lunch.

Before optimization                    After optimization
┌──────────────────────┐              ┌──────────────────────┐
│ visible pain          │              │ relieved pain         │
│ cost / wait / memory  │──change──►   │ lower local metric    │
└──────────┬───────────┘              └──────────┬───────────┘
           │                                      │
           ▼                                      ▼
 hidden cause not named                 new pressure appears
 retries, route mix, context,           quality, queueing, cache,
 output, provider limits                memory, fallback, ops

The diagram is the reason this module keeps returning to the fuel ledger. The ledger is where the second box becomes visible instead of surfacing as an invoice surprise, p99 incident, or quality complaint weeks later.

3) Maya builds a support-router confusion matrix

Maya threads one workload through the design review: a production assistant with real traffic, route versions, prompt versions, and outcome labels.

Attempt A — optimize the visible line

The first attempt changes the local knob associated with the opening symptom. The local dashboard improves. The team celebrates too early because the request boundary is still broken: retries, quality loss, queueing, cache misses, or memory pressure move elsewhere.

Attempt B — optimize with the pressure chain

The second attempt keeps the artifact, the rule, and the guardrail together. Maya writes the expected improvement, the pressure that may worsen, the owner of that pressure, and the rollback trigger. The dispatch board may change a route, the memorized route may change a prompt prefix, or the carpool lane may change scheduling, but the same request ID proves whether the user outcome survived.

4) Why measured cascades beat all-big or all-small routing

The plausible alternative is attractive because it is simpler to explain in a status update: change one knob, quote one percentage, and move on. That works for demos. It fails for lead-level ownership because it cannot answer which workload benefits and which workload pays.

Use this chapter's mechanism when the workload has the shape named in the opening artifact. Use the alternative when the product is small enough, stable enough, or low-risk enough that the extra machinery would cost more than it saves. The decision is not about elegance; it is about whether the signal-to-operator cost is worth it.

5) Repair rate decides whether weighted-average savings survive

Concrete numbers make the tradeoff review honest. The sample prices and memory figures below are illustrative; replace them with the provider, hardware, and workload numbers in your own stack.

Scenario	Fresh input	Cached input	Output	Extra condition	Lesson
All strong model	2000	0	400	0% repair	$0.0320
80% small / 20% strong	2000	0	400	0% repair	$0.0077
Small route, 10% repair	2000	0	400	10% repair	$0.0048 easy
Small route, 60% repair	2000	0	400	60% repair	$0.0208 easy
Risk override strong	2000	0	400	0% repair	$0.0320 but safer

The table teaches the design habit: every row says what improved and what might have worsened. If a row cannot name both, the proposal is not ready for production review.

6) The small model that was confidently wrong

Walk the failure from top to bottom. The user action enters the API. The application builds a prompt or route. The runtime spends tokens, queue time, cache memory, or output steps. The dashboard records a local improvement. Then the user-visible metric moves the wrong way.

That failure is not bad luck. It is what happens when the optimization changes one layer and the observation stops one layer too early. In a review, Maya asks for the missing link: where did the pressure go after the local metric improved? If nobody can answer, the change ships behind a small canary or does not ship.

7) Signals that reveal whether model routing is healthy

• Healthy behavior: cost per successful outcome improves while acceptance and safety stay stable.
• First degrading metric: small-route repair and escalation rate rises.
• Misleading beginner metric: percentage on cheap model, because it ignores outcome loss.
• Expert graph: route confusion matrix by segment with false-easy and false-hard counts.

Mini-FAQ. "Why not watch the simplest metric?" Because the simplest metric is often the one the optimization directly manipulates. You need a paired guardrail that shows whether the system merely moved pain into another layer.

8) Boundaries where the chapter's lever works and where it turns pathological

• Strong fit: high-volume traffic with separable easy and hard segments.
• Pathology: ambiguous tasks where confidence signals are weak and mistakes are costly.
• Scale or workload limit: when eval labels or outcome feedback are too sparse to tune the router.

This boundary is not a disclaimer. It is a routing rule for engineering attention. The best optimization in one endpoint can be the wrong default for another endpoint with different latency tolerance, risk, context length, or outcome value.

9) Wrong mental model to replace

The wrong model is “small model whenever possible.” The better model is capability matching: route by difficulty, risk, confidence, and business value, then let outcomes veto savings.

The replacement model should change how you speak in design review. Do not say, "this reduces cost" or "this improves latency" without naming the request slice, expected magnitude, guardrail, and rollback trigger. Say which meter ticks, ETA call, carpool lane, or boot space pressure changed.

10) Other failure shapes you will recognize

1. False-easy failure. false-easy routing of hard requests.
2. Weak failure. weak model approves its own answer.
3. Repair failure. repair loop erases savings.
4. Premium-tier failure. premium-tier policy confused with task difficulty.
5. Route failure. route drift after prompt or model update.
6. Safety-sensitive failure. safety-sensitive requests missing overrides.
7. Fallback failure. fallback latency not included in user budget.

11) Cross-topic reinforcement — the same pressure shape returns

• Cost anatomy turns routing into a weighted average with repair multipliers.
• Latency anatomy checks whether cascades add unacceptable delay.
• Streaming hides none of a bad route; it can reveal wrong content earlier.
• Dashboards later need route version and outcome joins.

12) Design-review questions that catch shallow plans

1. Do routes have labeled evals by task segment?
2. Can you explain every override rule?
3. Can you compute savings after repair and fallback?
4. Can product metrics veto a cheaper route?

Where this shows up in production

• Enterprise support bot — turns route, token, cache, retry, and outcome rows into cost per resolved ticket rather than model spend per message.
• Coding assistant — separates inline completions from agentic edits because typing flow, repo context, and repair loops have different budgets.
• Search answer product — pays for rewriting, retrieval, reranking, synthesis, citations, and judge calls as one user-visible answer.
• Voice assistant — treats dead air, cancellation, and local fallback as product features because users notice 100 ms gaps.
• Back-office summarizer — uses larger queues and batches because humans care about daily throughput more than first-token immediacy.
• Commerce assistant — protects purchase-changing actions with stronger routes while letting read-only advice run cheaper.
• Internal data copilot — attributes spend by tenant, dataset, prompt version, and tool path so one team cannot hide another team's budget.
• Education tutor — spends tokens on safety and pedagogy rules, then watches whether shorter answers still teach well.
• Legal review workflow — keeps evidence and citation context even when compression pressure is high because unsupported claims are worse than cost.
• Healthcare intake helper — uses conservative routing and buffered streaming because safety checks are part of the latency path.
• Marketing content tool — controls output length and variant count because creative generation can silently explode spend.
• Incident-response copilot — prefers predictable latency and logs over clever savings during high-severity operations.

Recall — rebuild model routing from memory

1. What concrete failure opened this chapter, and which artifact made it inspectable?
2. What root cause made the naive fix insufficient?
3. State the rule in one sentence without using vendor language.
4. Which pressure does the mechanism relieve, and which new pressure can it create?
5. Which operational signal degrades first when the mechanism is misapplied?
6. Where is the boundary where this lever becomes pathological?
7. How does this chapter reuse the fuel ledger or dispatch board from earlier chapters?
8. What would you put in the rollback trigger for this optimization?

Interview Q&A

Q: What problem does model routing actually solve?

A: It relieves the specific pressure from the opening failure: 80% of traffic moves to a small model, spend falls, and support resolution drops three points. The important answer names the workflow metric, the moved pressure, and the guardrail that keeps the optimization honest.

Common wrong answer to avoid: It is just a generic way to make LLMs cheaper or faster.

Q: Why is the naive fix dangerous?

A: Because the naive fix attacks the visible symptom while the root cause is the root cause is unmeasured difficulty classification. It can improve a local metric and damage outcome quality, tail latency, memory, or operational clarity.

Common wrong answer to avoid: If the local metric improves, the product improved.

Q: What artifact would you inspect first?

A: The first artifact is the joined request trace or table for this chapter: route/version, token buckets, latency stages, outcome, and the topic-specific signal. Without it, the team debates guesses.

Common wrong answer to avoid: I would start by changing model parameters.

Q: How do you know the optimization worked?

A: The relieved pressure improves while the guardrail remains stable: cost per successful outcome improves while acceptance and safety stay stable. You also check the first degrading metric: small-route repair and escalation rate rises.

Common wrong answer to avoid: The dashboard line for cost or latency went down.

Q: When should you avoid this lever?

A: Avoid it in the pathological case: ambiguous tasks where confidence signals are weak and mistakes are costly. The workload shape must match the mechanism, or the optimization moves cost into quality, memory, or user wait.

Common wrong answer to avoid: Use it everywhere because it is a best practice.

Q: What is the common wrong mental model?

A: The wrong model is “small model whenever possible.” The better model is capability matching: route by difficulty, risk, confidence, and business value, then let outcomes veto savings.

Common wrong answer to avoid: The obvious intuition is good enough.

Q: How does this connect to earlier chapters?

A: It reuses the workflow boundary from cost anatomy, the stage trace from latency anatomy, and the fuel ledger discipline. The lever should be explained as pressure relief plus a new pressure, not as an isolated trick.

Common wrong answer to avoid: Earlier chapters are background only; this mechanism stands alone.

Apply now (10 min)

Step 1 — model the exercise. Use the modeled table in this chapter, then pick one feature you own or know. Fill in the same rows with rough numbers, change one assumption, and reproduce from memory which metric should improve, which guardrail could fail, and what rollback trigger you would set.

Step 2 — your turn. Pick a real LLM feature and write the same artifact with your own rough numbers. Name the pressure relieved, the pressure created, the owner, and the metric that would prove the change unsafe.

Step 3 — reproduce from memory. Close the file and redraw the two diagrams: request trace and pressure transfer. Then restate the rule and the first degrading metric without looking.

What you should remember

This chapter explained why the opening failure is not solved by changing one local knob. The useful move is to make the request boundary inspectable, apply the topic rule, and watch the paired guardrail so the optimization cannot hide its cost in another subsystem.

You learned to describe the lever as pressure movement: what it relieves, what it creates, and which team or resource absorbs the new cost. That is the difference between a trick and an operating practice.

Carry the diagnostic forward: if the dashboard cannot show the artifact, the route or version, the user outcome, and the first degrading signal in one place, the optimization is not yet reviewable.

Remember:

• Use the cheapest route that meets the task’s quality, safety, latency, and business-value budget with measured confidence.
• The first artifact to inspect is the trace/table that exposes 80% of traffic moves to a small model, spend falls, and support resolution drops three points.
• The first degrading signal is: small-route repair and escalation rate rises.
• The misleading beginner signal is: percentage on cheap model, because it ignores outcome loss.
• Every optimization must name what pressure it relieves, what pressure it creates, and who owns the new cost.

---

Bridge. Routing chooses the lane, but the user still waits while the chosen model works. The next chapter uses the ETA call to turn silence into useful progress and to stop meter ticks when the user cancels.

→ ./05-streaming-first-token.md