bear-counter

bear-counter · Find opposing literature

One task: you hand over a viewpoint, and it brings back genuinely existing opposing literature. You rely on instinct to find support; the picky reviewers who read your work read the things you skipped. Run bear-support and bear-counter once each on the same passage, and you can see both sides at the same time.

First read references/sci-cli.md—CLI detection steps, usage, and ironclad rules. Then read references/output-system.md—the three-tier output system and HTML shell specification.

Steps (fixed order)

Step 0 — Confirm the CLI is available Run sci --version. If it is not installed, give the installation command and stop; if there is an authentication error, prompt sci init and stop.

Step 1 — Lock down the claim Compress the viewpoint into one precise, falsifiable sentence. Dig out its implicit scope (which population, under what conditions, how measured)—the places most vulnerable to attack are often hidden in unstated assumptions.

Number the claim as C1. If the user inputs multiple claims, first split them into C1、C2、C3, but by default prioritize attacking the most central one; handle the remaining claims only if the retrieval budget allows.

Step 2 — Build five categories of opposing queries, and inform the user Based on the real claim, generate five directions:

1. Direct contradiction (opposite results)
2. Boundary / complication (under which conditions it weakens or reverses)
3. Alternative explanation (same phenomenon, different mechanism)
4. Methodological critique (known flaws of the method or measurement tool)
5. Replication failure / null results

List these five directions in the output (with one line each explaining what query terms will be used to search), thenstart retrieval directly without waiting for user confirmation—unless the claim itself is ambiguous in meaning, in which case confirm first and then execute. Searching everything is far more efficient than having the user approve one by one; if the user wants to adjust the directions afterward, tell them they can specify new ones.

Step 3 — Retrieve for each direction, and show the query terms For broad scanning use ultra_low(free), and for directions with useful results rerun with low . Use a different --prefix。for each direction and embed in the output the query terms actually used for each direction—the user needs to see "what it searched" in order to judge the credibility of the results, and it also makes later adjustment easier.

Step 4 — Classify and assign threat levels For each genuinely retrieved piece of opposing literature:

• Which category of challenge it falls into (one of the five)
• One line on "how it specifically threatens this claim"
• Threat level: high / medium / low (depending on how central the rebutted point is, and the persuasiveness of the literature itself)
• One line on "what would need to be shown to respond to it"

Criteria for determining threat level:

1. High: directly challenges the core causal relationship, core conclusion, or core scope of applicability
2. Medium: challenges boundary conditions, alternative mechanisms, measurement validity, or the scope of extrapolation
3. Low: only constitutes a background limitation, or the evidence is relatively distant from the claim

Do not write "insufficient research" as "the view has been refuted". A null result can only indicate that this search found no usable counter-evidence.

Step 5 — If it is quiet, report that it is quiet If a direction comes back from searching with no genuine opposing literature, truthfully say "no counter-evidence was found for this direction", and state clearly which possibility is more likely: the view has not been challenged, or the question is under-researched. Do not fabricate an opposing side just to appear comprehensive.

Output format

After the list of search directions, put the most dangerous paper at the top, and group the rest by type:

[!] 最危险
Smith et al. 2023 — "论文标题"
威胁程度 ████████░░ 高 · 直接矛盾
→ 具体怎么威胁你的主张（一句话）
回应：应对方向（一句话）
查询词：actual query used

The rest grouped by type (direct contradiction / boundary conditions / alternative explanations / methodological criticism / replication failure), each paper in the same format as above. Threat level: high = ████████░░, medium = █████░░░░░, low = ███░░░░░░░

Types for which no counter-evidence was found:— [类型名] 此方向未检索到反对文献, do not disappear from the list.

Deliverables

After the search ends, create the {topic-slug}/ subfolder and output four things:

1. Terminal — directly print the TUI format
2. report.md — generate according to the structured Markdown specification of output-system.md : stable YAML front matter + at-a-glance conclusion + threat matrix + suggested research actions + evidence table + detailed evidence + search transparency + reusable JSON
3. report.html — self-contained HTML, generated according to the output-system.md specification
4. references.bib — merge the .bib files generated by all queries, deduplicate, and keep only the literature entries actually cited in the report

HTML signature experience:threat panel, accent color #dc2626(red). The literature section is divided into tabs by the five categories of challenge directions. For HTML component specifications, see the bear-counter section of output-system.md .

For simple queries, give only 1–2 lightweight next steps. When writing a discussion or replying to reviewers, give specific suggestions: which boundaries should be proactively acknowledged, how to respond, and which robustness checks should be added.

Honesty boundaries

Do not manufacture critics, do not build straw men. Only genuine opposition found through search counts. Finding a large blank itself is a discovery — state it clearly, and do not read it as "you are right". Organizing opposing opinions into a Discussion or point-by-point rebuttal is a product-level task; hand it to SciMaster.

Gotchas

• The five categories of challenge directions cannot be merged: even if a category of direction finds only 1 paper or 0 papers, it must still be listed separately and the possible meaning of the null result explained; do not compress the five categories into "some opposing literature".
• Do not inflate threat levels: marking every paper as "high" will make users lose their sense of judgment; medium/low threat papers have clear criteria for determination, follow them.
• "Insufficient research" and "the view has been refuted" are different: a direction having no counter-evidence may mean the question has not yet been sufficiently researched; it cannot be inferred as "the view holds".