| name | venue-targeting |
| description | Pick a target publication venue, define the venue-specific publication bar, and route loop-backs after proof, experiment, or verifier passes. |
Skill — Venue Targeting
Use this skill whenever the workflow needs to:
- choose a target venue for a candidate idea
- decide whether current results are publishable
- determine whether to loop back to Literature, Modeling, Proof, or
Experiment
- retarget to a different venue family without losing the core idea
The key rule: never judge "publishable enough" in the abstract.
Always judge against a named target venue or venue family.
Required project state
When a candidate is still in ideation, store venue choice in
notes/gap_briefs/<slug>.md.
When a candidate is promoted, copy the active venue state into
project.yaml:
target_venue: ICLR
backup_venues:
- NeurIPS
- AAAI
venue_family: ml-top
publication_bar:
headline_result: "clear ML-relevant conceptual or algorithmic result"
novelty: high
theory_depth: medium
empirical_depth: high
writing_style: "ML-first framing with strong related-work positioning"
loop_history: []
These fields are optional until a candidate is promoted, but once the
project is in mode: project, they should be treated as part of the
canonical state.
What to write in each gap brief
Every surviving gap brief should include:
Target venue
Backup venues
Venue-fit rationale
Minimum publishable bar
Loop triggers
Do not leave these vague. Name the missing ingredient that would cause a
loop.
Venue families
Use the most specific named venue that fits, but reason through the
family-level bar.
ml-top
Named venues:
- ICLR
- NeurIPS
- ICML
- AAAI
- COLM
Typical bar:
- Clear ML problem relevance.
- Strong novelty relative to recent ML literature.
- Either strong empirical evidence, or a concept/theorem with obvious ML
significance.
- Framing must explain why the result matters for learning, inference,
optimization, generalization, agents, or deployment.
Fatal weaknesses:
- Nice theorem with weak ML relevance.
- Pure toy model with no credible learning implication.
- Empirical section too thin for claims made.
algorithms-theory
Named venues:
Typical bar:
- Theorem-first contribution.
- Real novelty in technique, model, or bound.
- Tight proofs and clean statements.
- Simulations are optional and usually secondary.
Fatal weaknesses:
- Main theorem is incremental.
- Proof is mostly adaptation without conceptual payoff.
- Framing relies on experiments instead of theorem depth.
econ-theory
Named venues:
- Econometrica
- QJE
- AER
- Theoretical Economics
Typical bar:
- Economic significance and model insight must both be strong.
- Comparative statics, equilibrium characterization, mechanism insight,
or welfare implications should matter economically.
- For top general-interest venues like AER/QJE, pure theory usually needs
unusually broad importance.
Fatal weaknesses:
- Elegant math with weak economic interpretation.
- No clear welfare / equilibrium / identification / policy stake.
- Contribution fits specialist theory outlet better than general-interest
economics.
market-design-game-theory
Named venues:
- ACM EC
- WINE
- GEB
- SAGT
- Games and Economic Behavior
Typical bar:
- Incentives, equilibria, welfare, strategic interaction, or algorithmic
market design must be central.
- Theory can dominate, but mechanism intuition must be crisp.
- Empirical support can help, but does not replace theorem value.
Fatal weaknesses:
- Nice combinatorics with no strategic content.
- Mechanism claims without equilibrium clarity.
- Results better framed as pure CS theory or pure economics.
networks-social-computation
Named venues:
- WWW
- KDD
- AAMAS
- Network Science
- Management Science
Typical bar:
- Clear network or multi-agent problem.
- Good fit between theory, data, and mechanism/behavioral story.
- Strong baselines or realistic structural assumptions often matter.
Fatal weaknesses:
- Network framing is superficial.
- Literature positioning ignores adjacent social/network science work.
- No credible evaluation plan where venue expects one.
ai-agents-multiagent
Named venues:
Typical bar:
- Multi-agent reasoning, planning, uncertainty, or coordination should be
central.
- Theoretical contributions should connect to agent behavior or decision
making, not float as detached math.
Fatal weaknesses:
- Results read like generic optimization with agents renamed in.
- No connection to agentic benchmarks, planning settings, or strategic
environments where relevant.
Venue selection procedure
When ranking a candidate, score:
problem_fit — does the venue care about this question?
result_fit — theorem / mechanism / empirical / lower-bound shape.
novelty_fit — how high the incremental-novelty tolerance is.
evidence_fit — how much proof, experiment, or literature coverage is
required.
writing_fit — whether the story can be told in that venue's language.
Choose:
- one
Target venue
- one or two
Backup venues
Do not choose a prestige target that the contribution shape obviously does
not support.
Publication-bar outcomes
Every proof pass, experiment pass, or verifier pass should end with one
of these judgments:
meets bar
close-needs-proof
close-needs-experiment
close-needs-model
close-needs-literature
strong-but-wrong-venue
not-publishable-yet
These are routing judgments, not just prose labels.
Routing rubric
Route to proof-pass again when
- the model is good
- literature framing is stable
- main gap is proof completeness
- the venue bar is blocked mainly by missing lemmas, gaps, or tightness
Route to experiment-pass again when
- the venue expects stronger empirical validation
- theorem side is adequate but evidence breadth is weak
- baselines, ablations, stress tests, or figures are missing
Route to refine-model when
- proof attempts expose the wrong primitive, assumption, payoff, or
statement
- current theorem set lacks a headline result for the target venue
- the venue demands a cleaner or more general formulation than current
model supports
Route to literature-review when
- novelty risk rises
- a nearby paper weakens the gap
- framing needs a different subcommunity or result lineage
- stronger baselines, comparisons, or borrowable techniques are needed
Route to retarget venue when
- the results are real and coherent
- the target venue bar is too high or mismatched
- another named venue fits the exact contribution better
Retargeting is not failure. It is often correct research strategy.
Loop triggers by failure type
Map failure type to route:
| Failure type | Route |
|---|
| Theorem technically incomplete but on-track | proof-pass |
| Theorem true only under changed assumptions | refine-model |
| Main statement not novel enough | literature-review |
| Results solid but too narrow for target venue | retarget venue |
| ML venue but no convincing ML relevance | literature-review or retarget venue |
| Theory venue but contribution depends on weak experiments | refine-model |
| Econ venue but no economic stakes | refine-model or retarget venue |
| Market-design venue but weak incentive content | refine-model |
Output expectations for agents
When this skill is used, the run summary should make four things explicit:
- active target venue
- publication-bar judgment
- main reason for the judgment
- recommended next route
Example:
### Next actions
- Target venue: `ACM EC` — publication-bar judgment: `close-needs-model`.
- Reason: proofs suggest the current result is technically sound but not
yet mechanism-first enough for EC.
- Recommended route: `refine-model` to strengthen incentive / welfare
content before another proof pass.