Build a reusable Beverage Storage Shelf-Life & Quality-Co…Food and Drink
How each model performed on this task. Every chart states its N, and the reward carries a 95% confidence interval once there is more than one run to measure.
Core metrics
Check Pass %
Mean share of the task's checks passing, averaged over runs. N = 1 run per model.
100%
Claude Opus 5.5
100%
GPT-6 sol
100%
GPT-5.6 sol
0%
Claude Opus 5
Extended metrics
Median Time
Wall-clock minutes, median across runs. N = 1 run per model.
84m
Claude Opus 5
16m
GPT-5.6 sol
Tool Calls
Tool invocations, median across runs — counted from the trajectories. N = 1 run per model.
118
Claude Opus 5
58
GPT-6 sol
39
Claude Opus 5.5
29
GPT-5.6 sol
Verifier × model pass rate
How often each verifier passed, per model. Green is passing, red is failing — a red band across a row is a verifier nothing clears.
| Verifier | Claude Opus 5.5n=1 | GPT-6 soln=1 | GPT-5.6 soln=1 | Claude Opus 5n=1 |
|---|---|---|---|---|
| checks (15) | ||||
| 100% | 100% | 100% | 0% | |
How to passEstablish exactly what the tracker evaluates: storage, monitoring, quality assessment, handling, retention, or discard decisions. Define the beverage categories and situations the framework can cover without assuming that every beverage behaves identically. Separate food-safety decisions, quality decisions, manufacturer instructions, regulatory requirements, and personal preferences. This distinction is essential because a beverage becoming less pleasant does not automatically mean it is unsafe, while the absence of obvious sensory changes does not automatically prove safety. Establish decision states such as 'retain as stored', 'recheck', 'follow manufacturer guidance', 'escalate for safety review', or 'discard' where supported by evidence. Define what information is mandatory before making a decision and when uncertainty should prevent a confident conclusion. The framework should remain useful months later when stored products are revisited or new beverage types are added. Avoid creating universal storage rules when authoritative sources specify product-specific conditions. The objective is to establish a disciplined boundary between what can be concluded from evidence and what requires additional information or professional food-safety guidance. | ||||
| 100% | 100% | 100% | 0% | |
How to passResearch and define a source hierarchy for beverage storage and quality decisions. Prioritize government food-safety agencies, public-health authorities, universities, recognized food-science organizations, official manufacturer documentation, standards bodies, and established technical publications. Use books, peer-reviewed research, beverage-industry publications, and professional magazines to explain deterioration mechanisms and storage science. Define which sources should establish safety principles, storage temperatures, labelling interpretation, shelf-life concepts, packaging risks, and microbiological considerations. Record source identity, publication or revision information, scope, accessibility, and claims supported. Do not treat general consumer advice as equivalent to product-specific manufacturer guidance. Likewise, do not assume manufacturer documentation resolves every scientific or safety question. Where credible sources differ, document the difference and investigate the reason rather than choosing whichever instruction appears easiest. The hierarchy should remain useful months later when a product's documentation changes or new scientific evidence becomes available. The objective is to ensure that every important tracker decision is traceable to an appropriate source and that uncertainty is preserved rather than replaced with invented certainty. | ||||
| 100% | 100% | 100% | 0% | |
How to passCreate a master register for each beverage being evaluated. Record product identity, beverage type, formulation or relevant category, packaging material, container size where relevant, sealed/opened status, label information, manufacturer guidance, storage instructions, and source references. Distinguish between similar products whose formulations or packaging may create different storage requirements. Record whether information is directly documented, independently verified, observed, inferred, or unknown. Capture relevant labelling terms such as storage instructions, opening instructions, date markings, or handling warnings without assuming that every date has the same meaning. Identify packaging conditions, including damaged seals, leaks, swelling, broken closures, or unusual container changes where relevant. Months later, the register should allow a reviewer to reconstruct exactly which product was evaluated rather than relying on a generic beverage category. The objective is to prevent a broad rule such as “all juice” or “all bottled drinks” from being applied to a product whose formulation, packaging, or labelling creates materially different requirements. | ||||
| 100% | 100% | 100% | 0% | |
How to passConvert manufacturer and authoritative guidance into specific storage variables that can be monitored. Record temperature, exposure duration, light conditions, humidity where relevant, container orientation where relevant, opening status, handling frequency, and other product-specific conditions. Distinguish between required conditions, recommended conditions, and conditions that merely improve quality. Define how each condition can be verified and what evidence should be recorded. Avoid relying on vague descriptions such as “kept somewhere cool” when a source provides a measurable requirement. Where no reliable numerical requirement exists, preserve the source's wording rather than inventing one. Months later, the same structure should allow the team to compare actual storage history with documented requirements. The objective is to transform storage advice into observable information that can support a decision. This also makes repeated failures easier to diagnose. A beverage that repeatedly experiences unsuitable conditions can be identified as a process problem rather than evaluated as an isolated incident. The tracker should distinguish between known storage conditions, estimated conditions, and completely unknown history. | ||||
| 100% | 100% | 100% | 0% | |
How to passDevelop a framework for interpreting available shelf-life evidence without inventing unsupported expiration periods. Research authoritative food-safety guidance, manufacturer information, food-science literature, and relevant labelling resources concerning best-before concepts, quality dates, safety-related dates, unopened versus opened products, and storage-dependent shelf life. Record the exact source, product state, storage assumptions, and limitations behind every shelf-life conclusion. Distinguish between a manufacturer-established shelf-life statement and a generalised recommendation from another source. Do not automatically treat a printed date as proof that a product is unsafe immediately afterward, nor assume that a product remains safe indefinitely because the date has not passed. Where opening changes the applicable guidance, record the change explicitly. Months later, the tracker should allow a reviewer to revisit the decision when a product has moved from sealed to opened status. The objective is to prevent date-only decision-making and instead evaluate date information together with product identity, storage conditions, packaging integrity, and authoritative guidance. When evidence is insufficient, the system should clearly mark the limitation rather than generate a speculative shelf-life estimate. | ||||
| 100% | 100% | 100% | 0% | |
How to passCreate a technical framework for evaluating how packaging and environmental exposure can affect beverage quality or safety. Research food-science and beverage-industry sources covering oxygen exposure, light sensitivity, carbonation retention, container-material interactions, temperature cycling, seal integrity, and other relevant deterioration mechanisms. Identify which mechanisms are applicable to each beverage category and which are unsupported assumptions. Record the exposure, expected effect, evidence source, confidence, and observable indicators. Distinguish between quality deterioration and evidence of a safety hazard. A beverage may lose flavour, colour, carbonation, or freshness without presenting the same type of risk as microbial contamination. Conversely, packaging damage may require a different response from ordinary quality decline. Months later, the tracker should help explain why similar beverages stored under different conditions experienced different outcomes. The objective is to move beyond generic advice such as “keep it cool” and understand which environmental factors actually matter for a specific beverage and package. The resulting framework should also identify which observations are useful evidence and which are too ambiguous to support a confident conclusion. | ||||
| 100% | 100% | 100% | 0% | |
How to passDevelop a framework for beverages after opening or after handling conditions change. Research authoritative food-safety guidance concerning post-opening storage, contamination pathways, handling practices, refrigeration where applicable, container contact, resealing, and repeated exposure. Record the opening event, subsequent storage conditions, handling history, and relevant source guidance. Distinguish between manufacturer-specific instructions and general food-safety principles. Do not assume that all opened beverages have the same risk profile. Identify when the available evidence is insufficient to determine a safe retention period. Record whether the product's condition can be assessed meaningfully through documented indicators or whether appearance and smell provide inadequate reassurance. Months later, the tracker should allow a reviewer to determine whether the original storage history remains credible or whether too much uncertainty has accumulated. The objective is to ensure that opening status becomes a formal change in the product record rather than a casual note. This makes the system more reliable for beverages whose storage requirements materially change after opening. | ||||
| 100% | 100% | 100% | 0% | |
How to passCreate a structured method for recording observable changes in beverage quality. Include appearance, colour, odour, texture, carbonation, sediment, package condition, taste (where appropriate and safe), and other product-specific indicators. Research food-science literature to understand which changes can reasonably indicate deterioration and which are normal characteristics of particular beverage types. Do not turn sensory observations into universal safety tests. Clearly distinguish between normal product characteristics, likely quality deterioration, abnormal observations requiring further investigation, and observations that cannot establish safety. Record the observation, date, storage history, source guidance, and resulting decision. Months later, repeated observations may reveal patterns associated with temperature exposure, opening practices, or packaging failures. The objective is to create useful evidence without encouraging unsupported conclusions. A strong tracker should prevent a reviewer from saying “it smells fine, so it must be safe” when the available science does not support that inference. Likewise, it should prevent normal characteristics such as sediment or color variation from being incorrectly treated as evidence of spoilage. | ||||
| 100% | 100% | 100% | 0% | |
How to passMove beyond individual beverages and evaluate the storage process itself. Record recurring problems such as temperature excursions, inconsistent refrigeration, repeated exposure to light, damaged packaging, poor stock rotation, incomplete labeling, unknown opening dates, or inconsistent handling. Research food-safety management principles and quality-control practices to identify useful process controls. Separate isolated incidents from recurring systemic failures. If multiple products experience similar storage deviations, determine whether the process rather than the individual product is the primary problem. Months later, the tracker should reveal whether corrective actions actually reduced repeated failures. Create fields for incident, root cause where supported, corrective action, verification, and recurrence. The objective is to turn individual beverage observations into a practical quality-control system. A beverage can be handled correctly today while the overall storage process remains unreliable. Identifying that difference makes the system more useful for long-term improvement. The framework should also allow corrective actions to be tested rather than simply recorded as completed. | ||||
| 100% | 100% | 100% | 0% | |
How to passCreate the central decision framework for determining what action is justified. Evaluate product identity, storage conditions, opening status, packaging integrity, shelf-life evidence, relevant observations, source authority, uncertainty, and known deviations. Establish mandatory safety gates where available evidence indicates that a conservative action is required. Do not allow favorable sensory observations or an unexpired date to automatically override a serious documented handling concern. Likewise, do not automatically treat every minor quality change as proof of a safety problem. For each record, provide Current Status, Evidence Supporting It, Important Uncertainty, Required Action, and Re-Check Trigger. Avoid false precision when evidence does not support a numerical risk estimate. Months later, new information should be added to the existing decision history and trigger reassessment when materially relevant. The objective is to make decisions transparent and evidence-based while avoiding unsupported medical or food-safety claims. The system should make clear when it can support a quality decision and when a safety question exceeds the available evidence and should be escalated to an appropriate authoritative source. | ||||
| 100% | 100% | 100% | 0% | |
How to passCreate a long-horizon workflow for maintaining the beverage tracker. Define triggers such as opening a container, storage-condition deviation, damaged packaging, updated manufacturer guidance, new safety information, repeated quality problems, missing documentation, or uncertainty about handling history. Preserve previous observations and decisions rather than overwriting them. Record the event, evidence, action, responsible reviewer, and whether the underlying storage process needs correction. Include a review queue for products whose status cannot be confidently determined from the available evidence. Establish periodic checks for storage conditions and documentation without inventing unnecessary review frequencies. Months later, the system should show which products were retained, which were discarded, which experienced deviations, and whether corrective actions reduced recurrence. The objective is to make the tracker useful as an ongoing quality-management tool rather than a one-time spreadsheet. A future reviewer should be able to understand not only the current status but also how the product arrived there and what evidence would cause the decision to change. | ||||
| 100% | 100% | 100% | 0% | |
How to passCreate exactly 8 fictional stress-test scenarios and apply the complete tracker to each. Clearly label every scenario as fictional. Include: (1) a beverage exposed to an unsuitable temperature for an uncertain duration; (2) a container with damaged packaging but no obvious visible leakage; (3) an opened beverage whose storage history and opening date are unknown; (4) manufacturer guidance that appears inconsistent with general food-safety guidance and requires source comparison; (5) a beverage showing quality changes without sufficient evidence to conclude a safety problem; (6) missing or unclear labeling information for an otherwise identifiable product; (7) repeated storage-condition failures across several beverages revealing a process-level problem; and (8) a beverage that appears normal but has a high-risk, poorly documented handling history. For every fictional scenario, demonstrate identity verification, source evaluation, storage-condition analysis, packaging assessment, shelf-life interpretation, observation review, uncertainty handling, action decision, and re-verification requirements. These are fictional methodology tests only. | ||||
| 100% | 100% | 100% | 0% | |
How to passAttempt to break the completed system using common food-storage assumptions. Present a beverage with an unexpired date but a serious documented storage deviation and test whether the date receives excessive weight. Introduce a product with an unusual appearance that is actually a normal characteristic and determine whether the tracker incorrectly treats it as spoilage. Present an apparently normal beverage with unknown handling history and test whether sensory observations are treated as proof of safety. Add a generic online shelf-life recommendation that conflicts with product-specific manufacturer documentation. Introduce repeated temperature excursions and determine whether the tracker recognises a process failure rather than treating each event independently. Present a damaged package without obvious leakage and test whether packaging integrity is properly separated from sensory assessment. Audit the source hierarchy, product identity, storage requirements, shelf-life interpretation, packaging analysis, opening workflow, observation framework, decision model, incident history, and re-verification process. Add safeguards against date-only thinking, smell-test overconfidence, generic-rule bias, manufacturer-documentation overreach, and isolated-incident thinking. | ||||
| 100% | 100% | 100% | 0% | |
How to passTest the completed tracker as though it were handed to an unfamiliar food service or quality reviewer. The stranger should be able to register a beverage, identify authoritative guidance, record packaging and storage conditions, interpret date information, document opening status, assess relevant observations, identify uncertainty, record incidents, determine an evidence-supported action, and understand when re-verification is required. Add a concise How to Use This Tracker Workflow: Covering Product → Source → Storage → Packaging → Date → Opening → Observation → Risk/Uncertainty → Action → Incident → Re-Check. Remove unexplained terminology and hidden assumptions. Confirm that a new beverage can be added without redesigning the tracker. The Stranger Test is satisfied only when an unfamiliar reviewer can complete the process without asking follow-up questions. Verify that the tracker remains useful weeks or months later when products are revisited, storage processes change, guidance is updated, or recurring failures emerge. Make clear that the tracker must not invent safety conclusions where authoritative evidence is insufficient. The system should direct unresolved safety questions toward appropriate authoritative food-safety guidance rather than encouraging unsupported judgement. | ||||
| 100% | 100% | 100% | 0% | |
How to passAssemble the final Beverage Storage, Shelf-Life & Quality-Control Tracker with clearly separated sections for Instructions, Beverage Identity, Source Register, Storage Requirements, Actual Storage Conditions, Packaging & Opening Status, Shelf-Life Evidence, Quality Observations, Incidents, Uncertainty, Action Decision, Corrective Actions, Re-Verification Queue, and Decision History. Ensure product-specific guidance is distinguished from general food-science principles and that unsupported shelf-life estimates are never invented. Verify that quality deterioration is not automatically equated with safety risk and that normal sensory characteristics are not incorrectly treated as evidence of spoilage. Confirm that date markings are interpreted in context with product identity and storage history. Ensure the eight scenarios remain clearly fictional and that the adversarial audit produces concrete safeguards. Review terminology, evidence states, action categories, storage-condition fields, incident logic, and re-verification triggers for contradictions or ambiguity. Final Quality Check: a stranger must be able to use the tracker to evaluate an unfamiliar beverage record, identify the relevant evidence, document storage and handling history, distinguish quality concerns from unresolved safety questions, make an evidence-supported action decision, and know when further authoritative review is required—without asking follow-up questions. | ||||
Failure modes on this task
15 tagged verifier failures, by the stage they were assigned to. Higher is worse.
| Failure stage | Claude Opus 5.5 | GPT-6 sol | GPT-5.6 sol | Claude Opus 5 |
|---|---|---|---|---|
| Delivering the result1 mode | — | — | — | 9 |
| Reaching the page1 mode | — | — | — | 1 |
| No signal in the judge’s words1 mode · the grader | — | — | — | 5 |