Why math and code permit objective rewards, how outcome and format signals work, where reward hacking appears, and how RLVR differs from preference RL.
This chapter follows the series' four-layer pyramid: intuition first, then consequences, system design, and finally implementation-level checks. It is written to be useful both as a first explanation and as a review sheet before reading the primary papers.
Verifiable tasks replace an expensive learned judge with executable correctness checks, enabling scalable reasoning RL—but only for properties the verifier truly measures.
What you should be able to do after reading
- Explain the mechanism without relying on the feature name.
- Trace the relevant tensors, losses, or messages through one concrete example.
- Distinguish a paper claim from an inference, implementation choice, or marketing shorthand.
- Design a minimal experiment that could prove the idea wrong.
Where this chapter fits in the ten-phase map
Reasoning post-training starts from the V3-family base engineered in Phases 1–8. RL changes which trajectories the model emits; it does not repeal attention, routing, precision, or serving constraints. Longer reasoning also makes context management and test-time cost more important.
The dependency is useful when debugging. If the model-level equation is correct but the measured result is poor, walk backward through representation, numerical format, memory layout, routing or communication, and finally the evaluation harness. The first broken contract is usually more actionable than the final benchmark delta.
1. RLHF and RLVR answer different questions
Preference RL asks which response humans prefer and often trains a reward model. RLVR checks objective outcomes such as an exact math answer, unit tests, or a formally verified proof. It reduces judge ambiguity but narrows the task domain.
2. R1 uses rule-based signals
The paper describes accuracy rewards for verifiable problems and a format reward that enforces reasoning and answer structure. It avoids neural outcome or process reward models in the core R1-Zero setup, limiting reward-model exploitation.
3. Outcome rewards leave credit assignment hard
A binary final reward does not say which intermediate step helped or harmed. Policy optimization must infer credit across a long trajectory. Group-relative baselines reduce variance, while diverse sampling gives the optimizer contrasting outcomes.
4. Verifiers have attack surfaces
A code solution can hard-code examples, exploit test weaknesses, time out selectively, or emit an answer that passes parsing without solving the task. Math extraction can accept malformed equivalences. The verifier is part of the security boundary.
5. Correctness is not the whole product
A correct but unreadable, needlessly long, unsafe, or language-mixed response may earn full task reward. The full R1 pipeline adds data and broader RL objectives because verifiable correctness alone does not define assistant quality.
Engineering lens. For every concept above, identify the tensor, state, metric, or system boundary that makes it observable. Then ask which assumption would make the claim fail. This keeps the chapter testable instead of leaving it as architecture vocabulary.
Worked example
Design a code reward: sandbox execution, hidden tests, resource limits, deterministic parsing, partial credit, and exploit checks. Then list behaviors that could maximize reward without implementing the intended algorithm.
Do the arithmetic with small dimensions first. Small examples expose index shifts, hidden assumptions, and missing denominators that disappear inside a billion-parameter headline. Once the hand-worked result is correct, automate it and compare the program output against the same values.
Implementation and measurement plan
Version rewards like code. Unit-test parsers, isolate execution, hold out adversarial tests, and audit high-reward samples manually. Track pass rate alongside response length, diversity, exploit indicators, and out-of-domain regression.
- State the exact model, checkpoint, hardware, and date behind every numerical claim.
- Separate algorithmic complexity, theoretical FLOPs, measured latency, memory, and end-to-end cost.
- Build a small reference implementation before optimizing kernels or distributing it.
- Compare against an equal-compute or equal-parameter baseline and report the denominator.
- Record failure cases and scope limits beside the successful result.
From a paper claim to an engineering contract
The primary anchor for this chapter is DeepSeek-R1: Incentivizing Reasoning Capability via Reinforcement Learning from DeepSeek-R1. Reading a number from that source is only the first step. A reproducible contract has four layers:
| Layer | Question to write down | Evidence |
|---|---|---|
| Mechanism | What operation, loss, state, or routing decision changes? | Equation, pseudocode, tensor shapes |
| Implementation | How is it realized on the named hardware and software stack? | Kernel, precision, layout, process groups |
| Measurement | Which denominator and baseline make the comparison fair? | Raw metrics, config, repeated runs |
| Scope | Where should the claim stop being trusted? | Failure cases, ablations, dated limitations |
This separation prevents a frequent error in frontier-model writing: converting a theoretical reduction into a latency promise, or converting one internal benchmark into a universal quality ranking. The implementation can fail to realize the algorithm, and the workload can fail to expose the intended benefit.
Failure modes and misleading shortcuts
- A rule-based reward is not automatically unhackable.
- Exact match can reject equivalent math answers.
- Public tests invite memorization.
- Outcome rewards do not validate reasoning steps.
- Optimizing only verifiable domains can narrow behavior.
These are not footnotes. Frontier-model engineering is dominated by boundary conditions: a method can be mathematically correct and still lose to memory traffic, data skew, numerical drift, evaluation leakage, or a poorly stated comparison. A credible result makes those boundaries visible.
How to audit claims about this topic
Rewrite each claim with its missing boundary: name the exact mechanism, identify the tensor or resource it changes, and attach the workload and measurement. Then construct a counterexample at the edge of the claim. If a sentence cannot survive that rewrite, treat it as orientation—not evidence.
Next, trace provenance. Prefer the primary report for configuration and results, the released code for implementation behavior, and your own profiler for product performance. Secondary explainers are valuable for intuition but should not silently become the source of a numerical claim.
Decision guide: when should you use this idea?
Use it when the bottleneck named in the thesis appears in profiler traces or controlled quality experiments, the necessary kernels and runtime support exist, and the added system complexity can be observed in production. Start with the smallest configuration that exposes the bottleneck.
Delay it when a dense or higher-precision baseline does not yet converge, the evaluation harness is unstable, or the claimed resource is not limiting the workload. Sophisticated architecture cannot compensate for an invalid baseline.
Reject it when its benefit exists only under a denominator irrelevant to the product—for example, theoretical FLOPs while user latency worsens—or when numerical, safety, or operational regressions exceed the measured gain.
Hands-on study lab
- 1. Threat-model a verifier.
- 2. Compare outcome and process rewards.
- 3. Create equivalent-answer tests.
- 4. Design an out-of-domain regression suite.
For each exercise, save the configuration, a tiny deterministic fixture, the raw measurements, and one failed case. The goal is not merely to make the code run; it is to make the conclusion independently checkable.
Quick self-check
What is the central idea?
Verifiable tasks replace an expensive learned judge with executable correctness checks, enabling scalable reasoning RL—but only for properties the verifier truly measures.
What is the most common reading mistake?
A rule-based reward is not automatically unhackable.
What evidence should I demand?
An exact configuration, a fair baseline, primary-source support, end-to-end measurements, and failure cases at the limits of the claim.
How do I explain it to a new engineer?
Begin with the bottleneck, show one tiny worked example, trace the changed state, and only then introduce the official name. Finish by naming one situation where the method will not help.
How do I review an implementation?
Check indexing and masks, parameter sharing, dtype transitions, layouts, process-group scope, raw metric denominators, and behavior under an adversarial or worst-case fixture. A passing happy-path shape test is not enough.
Teach-back synthesis
Close the page and reconstruct the argument in five sentences: the bottleneck; the mechanism; the state or tensor that changes; the fair measurement; and the main failure mode. Then reopen the page and compare. If you can repeat the feature names but cannot state those five sentences, revisit the worked example.
Finally, connect the idea to two neighboring phases. DeepSeek's advantage is not one isolated invention: compressed attention changes the cache, sparse experts change active compute, FP8 changes arithmetic and bandwidth, distributed schedules hide communication, and reasoning training spends the resulting capacity differently. The series becomes useful when those dependencies form one mental model.
Key takeaways
- Why math and code permit objective rewards, how outcome and format signals work, where reward hacking appears, and how RLVR differs from preference RL.
- The mechanism, training recipe, runtime implementation, and measured product behavior are separate layers of evidence.
- Numbers remain meaningful only with their workload, precision, hardware, context length, and date attached.
- A small reproducible test is more valuable than a large uncheckable diagram.
Primary sources and further reading
- DeepSeek-R1. DeepSeek-R1: Incentivizing Reasoning Capability via Reinforcement Learning.
- DeepSeekMath. Pushing the Limits of Mathematical Reasoning in Open Language Models.
- Schulman et al.. Proximal Policy Optimization Algorithms.