Search what System One
can take on.

capability · Agent flow · Decision

Ask or proceed

Identify missing information that blocks a concrete next step.

Research value not rated. About 133 input tokens in the sample. At 10,000 uses, about $0.06 at the Jev list rate.

capability · Agent flow · Decision

Choose the next tool

Choose among a small set of tools with explicit capabilities.

Research value medium. About 113 input tokens in the sample. At 10,000 uses, about $0.05 at the Jev list rate.

capability · Agent flow · Decision

Choose who should handle a request

Send a request to deterministic code, a specialist model, or a person.

Research value not rated. About 233 input tokens in the sample. At 10,000 uses, about $0.10 at the Jev list rate.

capability · Agent flow · Decision

Interrupt or stay quiet

Decide whether a background result deserves attention now.

Research value not rated. About 72 input tokens in the sample. At 10,000 uses, about $0.03 at the Jev list rate.

capability · Agent flow · Decision

Match a command to a request

Map an informal request to an available command ID.

Research value medium. About 92 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Agent flow · Decision

Recognize the current intent

Separate a question, a change request and a status request.

Research value not rated. About 94 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Agent flow · Decision

Review a proposed tool call

Classify a proposed tool call as clear or caution before anything runs.

Research value not rated. About 233 input tokens in the sample. At 10,000 uses, about $0.10 at the Jev list rate.

capability · Agent flow · Decision

Understand a tool result

Recognize success, failure or missing evidence without another long response.

Research value high. About 110 input tokens in the sample. At 10,000 uses, about $0.05 at the Jev list rate.

capability · Classification · Catalog selection

Navigate a decision tree

Use jev-tree to choose a known category with a bounded call budget.

Research value medium. About 83 input tokens in the sample. At 10,000 uses, about $0.03 at the Jev list rate.

capability · Classification · Decision

Route a support request

Choose a queue and flag an explicit time-sensitive problem.

Research value not rated. About 195 input tokens in the sample. At 10,000 uses, about $0.08 at the Jev list rate.

capability · Coding · Decision

Compare issue reports

Flag whether two reports describe the same observed failure.

Research value not rated. About 99 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Coding · Decision

Route a code change

Choose the right review path from a short change description.

Research value not rated. About 96 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Computer use · Decision

Check a browser outcome

Ask whether the final screen supports the result the user requested.

Research value high. About 98 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Computer use · Decision

Check a form against the task

Compare requested values with a visible form before the agent submits it.

Research value high. About 92 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Computer use · Decision

Choose the next browser operation

Choose a small next step from the current page state and task.

Research value high. About 170 input tokens in the sample. At 10,000 uses, about $0.07 at the Jev list rate.

capability · Computer use · Decision

Match a visible control

Choose a control ID from descriptions captured by your browser tool.

Research value high. About 102 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Computer use · Decision

Read what changed on screen

Classify observed progress after an action without inventing a result.

Research value high. About 143 input tokens in the sample. At 10,000 uses, about $0.06 at the Jev list rate.

capability · Computer use · Decision

Return a stuck browser task

Choose whether to wait, observe again or return control to the main agent.

Research value high. About 126 input tokens in the sample. At 10,000 uses, about $0.05 at the Jev list rate.

capability · Context · Decision

Find a useful preference

Distinguish a reusable explicit preference from incidental conversation.

Research value high. About 104 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Context · Decision

Flag how a passage relates to a task

Separate useful evidence, contradiction and embedded instructions.

Research value not rated. About 134 input tokens in the sample. At 10,000 uses, about $0.06 at the Jev list rate.

capability · Context · Decision

Keep useful context

Evaluate candidate context before adding it to a longer prompt.

Research value high. About 100 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Context · Decision

Select a useful source snippet

Select a source ID using the meaning of the request.

Research value high. About 101 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Context · Decision

Spot stale evidence

Identify whether a supplied fact needs a fresh lookup.

Research value not rated. About 65 input tokens in the sample. At 10,000 uses, about $0.03 at the Jev list rate.

capability · Corpus search · Decision

Choose the next corpus search step

Choose among retrieval expansion, keyword filtering, excerpt reading, reformulation or conclusion.

Research value high. About 223 input tokens in the sample. At 10,000 uses, about $0.09 at the Jev list rate.

capability · Corpus search · Decision

Reformulate agent search query

Choose how to refine a search query when BM25 or grep yields zero or noisy matches.

Research value high. About 170 input tokens in the sample. At 10,000 uses, about $0.07 at the Jev list rate.

capability · Corpus search · Decision

Select top candidate document

Pick the most promising candidate document from BM25 retrieval to stage or inspect first.

Research value high. About 138 input tokens in the sample. At 10,000 uses, about $0.06 at the Jev list rate.

capability · Corpus search · Decision

Triage grep match lines

Separate high-signal matches containing relevant evidence from routine boilerplate.

Research value high. About 120 input tokens in the sample. At 10,000 uses, about $0.05 at the Jev list rate.

capability · Corpus search · Decision

Verify claim against candidate passage

Compare a factual claim with a concrete line range or passage retrieved from the corpus.

Research value high. About 95 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Design · Decision

Match a color palette

Choose an approved palette that fits a written brief.

Research value medium. About 112 input tokens in the sample. At 10,000 uses, about $0.05 at the Jev list rate.

capability · Experiences · Dialogue

Choose a natural moment

Let an optional adviser choose an eligible recorded reaction or silence.

Research value medium. About 65 input tokens in the sample. At 10,000 uses, about $0.03 at the Jev list rate.

capability · Operations · Decision

Choose a diagnostic test

Rank proposed checks using current failure evidence.

Research value high. About 115 input tokens in the sample. At 10,000 uses, about $0.05 at the Jev list rate.

capability · Operations · Log triage

Prioritize log inspection

Use jevlogs to separate routine events from records that deserve investigation.

Research value low. About 43 input tokens in the sample. At 10,000 uses, about $0.02 at the Jev list rate.

capability · Research · Decision

Screen a document against criteria

Check an abstract against explicit inclusion criteria.

Research value not rated. About 119 input tokens in the sample. At 10,000 uses, about $0.05 at the Jev list rate.

capability · Selection · Decision

Compare two item records

Flag whether two descriptions likely refer to the same item.

Research value medium. About 102 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Selection · Decision

Match an item to a description

Choose a supplied asset, template or catalog item by its description.

Research value high. About 111 input tokens in the sample. At 10,000 uses, about $0.05 at the Jev list rate.

capability · Selection · Decision

Select values for known fields

Map a request to allowed field values without generating arguments.

Research value not rated. About 108 input tokens in the sample. At 10,000 uses, about $0.05 at the Jev list rate.

capability · Verification · Decision

Check a citation in context

Compare a claim with the supplied source passage.

Research value high. About 116 input tokens in the sample. At 10,000 uses, about $0.05 at the Jev list rate.

capability · Verification · Decision

Check a claim

Compare a short claim with the evidence already collected.

Research value high. About 86 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Verification · Decision

Check a repair against its invariant

Distinguish temporary recovery from a repair that meets stated requirements.

Research value high. About 122 input tokens in the sample. At 10,000 uses, about $0.05 at the Jev list rate.

capability · Verification · Decision

Check launcher acceptance

Batch independent checks over one observed result.

Research value not rated. About 118 input tokens in the sample. At 10,000 uses, about $0.05 at the Jev list rate.

capability · Verification · Decision

Review a draft before it is sent

Check a draft reply against the evidence, then recommend send, hold, or review.

Research value not rated. About 248 input tokens in the sample. At 10,000 uses, about $0.10 at the Jev list rate.

capability · Verification · Decision

Review a proposed refund

Check whether a refund request names an amount and looks like a duplicate charge.

Research value not rated. About 254 input tokens in the sample. At 10,000 uses, about $0.11 at the Jev list rate.

capability · Verification · Decision

Score a small rubric

Evaluate a bounded quality scale with explicit anchors.

Research value not rated. About 101 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

capability · Writing · Decision

Check named writing rules

Flag vague wording and missing specifics before an agent edits a draft.

Research value medium. About 129 input tokens in the sample. At 10,000 uses, about $0.05 at the Jev list rate.

capability · Writing · Decision

Label a text block

Assign a known structural label to supplied text.

Research value medium. About 95 input tokens in the sample. At 10,000 uses, about $0.04 at the Jev list rate.

guide · Coding and chat agents

Check a claim against tool output

Ask whether the evidence supports a statement before an agent repeats it as fact.

guide · Coding and research agents

Shortlist context for an agent

Evaluate a small set of retrieved passages without losing the source material.

guide · Agents with several connected tools

Choose the next available tool

Give an agent a tool recommendation from a list it can actually use.

guide · Anyone creating a System One pattern

Write a question Jev can evaluate

Define the evidence, possible answers and review policy before adjusting a threshold.

guide · Individual coding and chat agent users

Measure whether offloading helps

Compare completed tasks with and without System One, including review and fallback costs.

guide · Individual coding-agent users with a local stdio MCP client

Give an agent a short browser subtask

Use current control IDs, caller-supplied text and final-screen review with the System One browser companion.

guide · Agent and computer-use adapter developers

Give Jev useful accessibility evidence

Use labels, group context and exact field facts to turn a screen into a small decision.

finding · reported · independent experiment

Model routing: test whether Jev adds value

No-Jev ablation matched the hybrid result.

finding · reported · independent experiment

Choosing tools from real MCP inventories

Better tool prediction can still take longer.

finding · reported · synthetic test

Separate questions for separate hazards

Explicit checks helped; calibration and latency limits remain.

finding · first-party · public-dataset study

Log triage: filtering is not automatically saving

Conservative triage can add cost; reuse may help more.

finding · first-party · synthetic test

Navigating beyond a flat choice limit

Trees avoid truncation; structured data may need no model.

finding · reported · vendor benchmark

Typed evaluation versus chat-model wrappers

A reason to test typed decisions, not a savings guarantee.

finding · reference · official guidance

Define ambiguity before evaluating a workflow

Specify what uncertain and mixed cases should do.

finding · first-party · synthetic test

Batching two evidence checks

Fewer calls and tokens, with one extra error in this small test.

finding · reported · public-dataset study

Rank retrieved passages before reading them

Ranking gains depend on the dataset and how scores are averaged.

finding · reported · public-dataset study

Find the likely failure in an agent trace

Trace classification is promising, but most full diagnoses were wrong.

finding · reported · synthetic-data study

Separate signals from a final verdict

Question design helped, but a simple rule was a strong baseline.

finding · reported · public-dataset and synthetic study

Compare decision types before choosing a model

A broad comparison supports task-specific testing, not one universal winner.

finding · anecdotal · firsthand Reddit report

A personal app routes two recipe requests

Descriptions can distinguish two plausible routes; reliability remains untested.

finding · anecdotal · firsthand Reddit report

A user tries Jev to reduce a long agent history

A user found history selection useful; quality and cache costs need a test.

finding · anecdotal · X anecdotes and criticism

X posts disagree about compaction by filtering

Conflicting firsthand views identify a question to test, not a settled result.

finding · anecdotal · builder report

Choose a card design from a page description

A concrete design-selection integration exists; preference quality is unmeasured.

finding · reported · builder experiment

Select an editor command from an informal request

An authored demo maps descriptions to commands better than its name matcher.

finding · reported · application experiment

Choose diagnostic tests and review repair evidence

More attempts passed in one small study; some incidents regressed.

finding · reference · official examples

TypeSafe examples for decisions over supplied text

Official examples show how to frame the question; each adaptation needs testing.

finding · first-party · synthetic test

Parable dialogue advice chose silence in every first-run case

Conservative silence is not evidence of better dialogue.

finding · reported · independent experiment

Browser actions from observed control IDs

Less browser overhead helped a small matched comparison.

finding · reported · independent experiment

Mac actions from OCR and accessibility

Structured screen extraction is part of the workload.

finding · first-party · synthetic test

Shared page state improved our first browser diagnostic

Four of six saved states after a shared-input fix.

finding · reference · implementation

What Jev needs from a computer-use adapter

Supply explicit accessible controls and shared state.

finding · first-party · synthetic test

Six accessibility-informed next-action choices

Six correct choices; complete-task benefit still untested.