Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsSome links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
To stop an LLM from returning broken JSON, treat the response as an untrusted interface rather than a string to parse. Constrain generation with a supported schema, parse defensively, validate that schema, then run application-level checks for meaning and authorization. JSON mode can reduce syntax errors, but it does not guarantee required fields or correct types; stricter structured-output features improve that layer without eliminating refusals, truncation, or domain mistakes.
The reliable design is a pipeline: one canonical schema, constrained decoding, strict parsing, JSON Schema validation, semantic checks, bounded recovery, and observability. Each stage answers a different question, so passing one stage must never be mistaken for passing all of them.
As an Amazon Associate I earn from qualifying purchases.
What counts as “broken JSON” in an LLM response?
“Broken” includes more than text that a JSON parser rejects. There are three failure classes, and each needs a different control.
Syntactic failure: the text is not JSON
RFC 8259 defines JSON as a sequence of tokens. A response fails this layer when it has a missing comma or brace, an invalid escape, an unquoted key, a trailing explanation, Markdown fences, or truncation in the middle of a value. A stream that ends before the closing brackets is syntactically incomplete even if every token received so far looks plausible.
#1 Best Overall
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
Duplicate object names are another hazard. The RFC says object names should be unique, but implementations vary when they are not: some keep the first value, some keep the last, and others expose both. Reject duplicates or detect them during parsing instead of allowing a language-specific parser choice to decide which value reaches your application.
Contract failure: valid JSON with the wrong shape
A parser can accept {"total":"12.50"} even when your contract requires a number. It can also accept an empty object, an unknown enum value, a negative quantity, an unbounded array, or a string that does not satisfy a declared pattern. JSON Schema’s Core and Validation vocabularies express required properties, primitive types, enums, ranges, array limits, and string constraints.
Semantic failure: valid shape, unusable meaning
Schema validation still cannot answer every business question. A date may have the required format but fall outside the booking window. An identifier may match its pattern but refer to a customer the caller is not authorized to access. Cross-field relationships, referential integrity, permissions, and business invariants belong in ordinary application code after structural validation.
Why does JSON that parses still break my application?
Parsing proves only that a decoder can build a data structure. It does not prove that the structure contains the keys your code reads, that values have the expected types, or that the values are safe to use.
Suppose an order endpoint expects currency, items, and a non-negative integer quantity. The following is syntactically valid but contractually and semantically unsafe:
{"currency":"USD","items":[{"sku":"A-17","quantity":-1}],"note":null}A typed model or JSON Schema validator should reject the negative quantity before inventory code runs. A separate authorization check should confirm that the caller can order the SKU. Do not “fix” this by silently coercing "-1" to -1 or filling missing fields with defaults; record the original response and classify the failure first.
Should I use prompt-only JSON, JSON mode, Structured Outputs, or a validator?
These options solve different parts of the problem. The most dependable production design combines constrained generation with parsing and post-generation validation rather than choosing a single feature.
Rank #2
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
| Approach | Syntax guarantee | Schema guarantee | Semantic checks | Recovery burden |
|---|---|---|---|---|
| Prompt-only JSON request | Weak; the model may add prose or malformed syntax | None | Owned entirely by your application | High |
| JSON mode | Designed to produce valid JSON, with documented edge cases | None; it does not guarantee a particular schema | Owned by your application | Medium to high |
| Structured Outputs or a strict function schema | Valid JSON plus adherence to the supported schema subset | Strong within that subset | Still owned by your application | Lower, but refusals and truncation remain |
| Parse plus a JSON Schema validator | The parser rejects invalid syntax | Strong after generation, regardless of model behavior | Owned by your application | Medium; you still need a retry and repair policy |
What JSON mode does and does not promise
OpenAI’s documentation distinguishes JSON mode from schema-constrained generation: JSON mode is intended to produce valid JSON, but it “will not guarantee the output matches any specific schema.” You must still check required properties, types, enum values, limits, and domain rules. Configure the request so the model is explicitly instructed to emit JSON; otherwise provider-specific behavior can include an incomplete or unusable response.
What Structured Outputs adds
OpenAI describes Structured Outputs as ensuring that responses adhere to a supplied JSON Schema. That guarantee applies to the provider’s supported subset and to successful, non-refused, non-truncated responses. It is a generation control, not a replacement for parsing, authorization, or business validation.
How do I define one canonical schema?
Start with the type that your application actually consumes, then derive the wire schema from it. Keeping a separately edited prompt schema, API schema, and language model invites drift: a field becomes optional in one place, an enum changes in another, and failures appear only at runtime.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Model the domain before writing the prompt
- Name fields for their meaning, not for how the model might phrase them.
- Mark every required property explicitly; do not rely on a prose instruction such as “always include these fields.”
- Use the narrowest primitive type: integer instead of a numeric string, boolean instead of
"yes", and an enum instead of free text when the set is closed. - Bound arrays, strings, and numbers where the domain has real limits. Bounds protect downstream systems from accidental or adversarially large output.
- Describe units, timezone expectations, nullability, and examples in field descriptions.
Generate and diff the schema
Native helpers reduce disagreement between code and the schema sent to the model. For a Pydantic model, model_json_schema() generates a JSON Schema (including Draft 2020-12 constructs where applicable). Store the generated artifact or its hash, and fail CI when a model change alters the schema without an intentional review. The same principle applies to TypeScript, Java, Kotlin, or any other typed stack: generate from the source of truth or compare the two representations automatically.
Keep provider limits in view
Schema-constrained APIs support a defined subset. OpenAI documents support for strings, numbers, booleans, integers, objects, arrays, enums, and selected combinations such as anyOf, along with root-level restrictions. A general JSON Schema can be valid according to the standard yet rejected by the provider. Check the provider’s current schema-support list, keep the model-facing schema simple, and perform richer checks in your validator and domain code.
How should I constrain generation without making the schema brittle?
Use the strictest native decoding available for the provider and model you selected, but design for the subset it can enforce. Put the contract in the API’s schema parameter or strict tool/function definition rather than repeating a long description in the prompt.
- Select a supported response format. Choose Structured Outputs or an equivalent strict schema option when available. Use JSON mode only when you can tolerate application-owned schema enforcement.
- Send the same schema your validator will use. Avoid maintaining a simplified prompt-only copy.
- Give fields clear descriptions. Explain units, allowed enum meanings, and whether an empty array is valid. Descriptions guide generation but are not validation.
- Constrain bounded values. Add minimum, maximum, length, and item limits that reflect real domain rules and provider support.
- Test refusal and truncation paths. A strict decoder cannot make an answer appear when the model refuses or reaches its output limit.
What is the safest parsing sequence?
Parse only after checking that a response exists and was completed. Preserve the raw, access-controlled response for diagnostics, then classify failures instead of attempting invisible repairs.
Rank #3
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
- Inspect transport and finish metadata. Distinguish a normal completion from a refusal, cancellation, timeout, or maximum-token finish reason.
- Reject empty content. An absent message is not an empty JSON object.
- Reject presentation wrappers unless explicitly supported. Markdown fences, prefatory prose, and trailing commentary should fail the parser rather than be stripped silently.
- Parse with duplicate-key detection. Configure the parser, or a pre-parser, to reject repeated object names.
- Handle streams as one document. Buffer and assemble chunks according to the provider protocol; do not parse each chunk as an independent JSON document.
- Record the failure class. Store a redacted excerpt, parser error location, model and version, schema hash, and finish reason.
Do not use a permissive “JSON repair” routine as the first line of defense. If you intentionally support a repair step for a non-critical workflow, run it in a separate, observable branch and validate the repaired value against the original contract before use.
How do I validate the parsed value with JSON Schema?
Run a standards-compliant validator immediately after parsing and before business logic. Pass the exact schema version and collect the failing path, keyword, and human-readable message so a retry can be targeted.
raw = receive_response()
meta = inspect_finish_and_refusal(raw)
if meta.refused or meta.incomplete:
return handle_non_content(meta)
value = parse_json(raw.content, reject_duplicate_keys=True)
errors = validator.iter_errors(value, schema=CANONICAL_SCHEMA)
if errors:
return retry_or_escalate(classify_schema_errors(errors), raw)
return validate_domain_rules(value)Remember that format is often annotation-only. A validator may accept an invalid email, date, or URI unless assertion behavior is enabled or you add an explicit check. Decide which formats are enforceable in your chosen library, configure assertion mode where appropriate, and test it rather than assuming the keyword performs validation.
Useful structural checks
- Required properties are present and no unexpected properties are accepted unless your contract permits them.
- Primitive types are exact, with deliberate handling of nullability.
- Enums contain only values your application recognizes.
- Numbers and string lengths are within operational limits.
- Arrays have bounded size and each item validates against its item schema.
- Nested alternatives (
anyOfor equivalent) resolve to an unambiguous branch.
What domain checks must happen after schema validation?
Domain validation is ordinary application code and should be explicit. Keep it separate from the model-facing schema so a change in business policy does not require pretending the model can enforce authorization.
Cross-field and temporal rules
Check relationships such as start < end, a total equaling the sum of line items, or a percentage allocation adding to 100. Parse dates into a timezone-aware representation and enforce the business window, not merely an ISO-like shape.
Authorization and referential integrity
Look up identifiers in your own database and verify that the current principal can access each referenced record. A well-formed account ID is not proof that the model was allowed to return it. Reject stale, unknown, or unauthorized references without sending sensitive lookup details back to the model.
Safety and operational limits
Apply limits on execution time, fan-out, monetary value, and resource consumption. Treat model-supplied URLs, file paths, regular expressions, and code as data that requires separate allowlists or sandboxing; schema validity does not make them safe to execute.
Rank #4
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
What should I do when the model returns truncated JSON or refuses?
Branch on the provider metadata instead of labeling every non-object response “malformed JSON.” Refusals and incomplete max-token responses are expected edge cases for structured generation.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteRefusal
If the API exposes refusal metadata, route it to a policy-aware response. Do not retry unchanged in a tight loop, and do not attempt to parse refusal text as the requested object. Depending on the workflow, ask the user for a safer reformulation, return a typed refusal result, or escalate to a human path.
Maximum-token truncation
A max-token finish reason means the document may be cut off before its closing token. Record the event and retry with a smaller input, a more compact schema, a lower-detail instruction, or a larger output allowance if the service permits it. Never append a brace and assume the object is complete; even a syntactically repaired value can omit required fields or end inside a string.
Transient transport errors
Use bounded retries with exponential backoff for timeouts and service errors, subject to your idempotency policy. Keep the same schema hash and request correlation ID, and avoid charging the downstream system twice for a response that may have completed remotely.
How should retries and repair be designed?
Retry only when the failure is plausibly recoverable, and cap the budget. A useful retry message contains the compact validation error, the expected schema version, and the original task—not an ever-growing transcript of previous failures.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →- Classify. Separate syntax, contract, semantic, refusal, truncation, and transport failures.
- Decide ownership. Fix deterministic application bugs locally; ask the model only to correct model-produced content.
- Retry once or a small, explicit number of times. Set a time and cost ceiling per request.
- Revalidate from scratch. Parse and run the full schema and domain pipeline on every attempt.
- Escalate deterministically. Return a typed error, queue human review, or use a safe fallback rather than passing an unverified value.
Automatic field-by-field patching is risky when a value has dependencies. If you do use a repair model, provide only the necessary error paths and schema, preserve the original response, and mark the result as repaired in telemetry.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What should I log to make JSON failures debuggable?
Observability turns an intermittent “JSON error” into a reproducible category. For each attempt, record:
Best Value
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
- model name and version, request and correlation identifiers, and the canonical schema hash;
- finish reason, refusal state, timeout or transport status, and token limits;
- parse success or parser error location, including duplicate-key detection;
- JSON Schema error keyword and instance path;
- domain-rule failure code, retry count, latency, and cost where available;
- a sanitized, access-controlled excerpt or digest of the response.
Redact secrets and personal data before logs leave the trust boundary. Keep enough context to replay the validator against a stored fixture, and add regression tests for every incident class.
How do I test an LLM JSON contract before production?
Test the pipeline, not only happy-path examples. Keep fixtures for valid responses, missing keys, wrong primitive types, unknown enums, boundary numbers, oversized arrays, duplicate names, Markdown wrappers, trailing prose, invalid escapes, and truncated streams.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Contract and drift tests
- Generate the schema from your typed model and compare its hash with the checked-in model-facing schema.
- Run provider compatibility tests whenever the model, API version, or schema changes.
- Verify that the validator rejects every intentionally invalid fixture and accepts representative valid values.
- Exercise refusal and max-token branches with mocked metadata; do not wait for a live incident to discover they are unhandled.
Property and boundary tests
Generate values at minimum, maximum, empty, null, and just-outside-boundary conditions. Test cross-field combinations that are individually valid but jointly impossible. Load-test array and string limits to ensure validation itself cannot become a denial-of-service vector.
Which architecture should I choose?
Choose based on the consequence of a wrong value, not on how clean the sample output looks.
| Situation | Recommended baseline | Why |
|---|---|---|
| Prototype or low-impact extraction | JSON mode, defensive parser, JSON Schema validation | Fast to integrate, but application code owns contract enforcement and recovery. |
| Writes to a database or external API | Structured Outputs or strict function schema plus independent validation | Generation is constrained, while authorization and invariants remain under your control. |
| Payments, permissions, or irreversible actions | Strict schema, domain checks, idempotency, human or deterministic approval gate | No model response should directly authorize an irreversible side effect. |
| Long streamed documents | Protocol-aware assembly, completion checks, bounded output, and resumable retry | Streaming increases the chance of incomplete content and complicates parsing. |
What is the practical implementation checklist?
- Define one typed domain model and generate the canonical schema.
- Hash and review schema changes in CI.
- Use provider-native strict decoding when its supported subset fits your contract.
- Keep the model-facing schema bounded and unambiguous.
- Inspect refusal, finish, and transport metadata before parsing.
- Reject empty content, wrappers, trailing prose, duplicate keys, and incomplete streams.
- Run JSON Schema validation with deliberate
formatassertion behavior. - Run authorization, cross-field, temporal, and referential checks in application code.
- Retry only recoverable failures with a small, observable budget.
- Log schema hashes, failure paths, model versions, and sanitized evidence.
- Use fixtures and boundary tests to prevent regressions.
The central design decision is to stop treating “valid JSON” as the finish line. It is the first gate in a chain that lets your application distinguish readable text from a value it can safely trust.
Frequently Asked Questions
Can I remove Markdown fences before parsing an LLM response?
You can support that as an explicit compatibility mode, but stripping fences and prose hides a contract violation. Prefer a strict path that rejects wrappers, records the raw response, and only enables a narrowly tested normalization branch for low-risk workflows.
Free tools Windows power users keep installed
One-click scans. No signup required.
Is a JSON Schema validator enough to secure an LLM-generated command?
No. A validator checks shape and declared constraints, not whether the caller is authorized or whether executing the command is safe. Use allowlists, permission checks, rate limits, and a confirmation or approval gate for consequential actions.
How can I detect duplicate keys reliably?
Many standard parsers silently keep one duplicate, so configure an object-pairs hook or equivalent parser option that raises on repetition. Add duplicate-key fixtures to your contract tests and reject the document before schema validation.
Should I stream structured JSON to users?
Stream only when your client can assemble provider chunks and wait for a completed finish signal before treating the value as usable. For side effects, buffer and validate the complete document first; partial objects should remain display-only and non-actionable.
What happens if a provider supports only part of my JSON Schema?
Keep the provider-facing schema within its documented subset and move unsupported assertions into your post-parse validator. Version both layers, test their equivalence, and treat a provider schema rejection as a deployment error rather than weakening the contract silently.
Recommended Free Tools
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
