Sundae Bar Logo
Past Challenge
M&A Due Diligence Signal Classification Skill
SBC36

M&A Due Diligence Signal Classification Skill

Overview

This skill teaches the agent to read a pool of public signals about an acquisition target, classify each one by its evidence strength and deal relevance, and surface only the findings that actually matter to the specific deal on the table.

Given a free-text pool of retrieved signals (news summaries, executive change notices, funding announcements, litigation references, regulatory notices, customer win/loss signals) alongside the user's stated deal context, a good result is a structured classification table covering every signal, a count line, and a concise narrative that explains the must-address picture, any elevation rationale, and the most important evidence gap, all without offering a go/no-go recommendation or attaching numeric risk scores.

The submission is a SKILL.md, with optional reference files the agent opens by name when SKILL.md directs it to. Strong skills teach a reasoning procedure: how to read and anchor deal context first, how to count independent signal origins rather than URLs, how to weigh magnitude and reversibility when tiering, and how to write assignable required-action statements. Weak skills paste in classification templates or memorised signal patterns and collapse the moment surface details rotate.

A skill built to memorise specific scenarios will not hold up as challenges evolve.

You submit a skill: a SKILL.md and, if you need them, reference files it points at. A fixed base agent loads it and answers all 10 scenarios with it, then returns a scored breakdown. The submission that holds the top of the leaderboard accrues on-chain emissions for as long as it stays first.

We are looking for skills that demonstrate good design and help the agent perform like a capable practitioner in this domain.

Requirements (10 Tests)

Your skill is tested across 7 capability types, with 10 scenarios in total. Every submission is evaluated against all 10 scenarios. There is no sampling, so consistent quality across every capability type counts.

CategoryOverviewTest cases
Type A: Attribution and Identity GatingThis type tests whether the agent correctly applies the attribution rule as a mandatory gate before assigning any deal-relevance tier. Scenarios present signals that either do not name the target at all or name a similarly-named but distinct legal entity. The skill must teach the agent to route unnamed signals directly to noise and to hold ambiguously attributed signals at deeper-diligence with a single confirm-identity action, deferring all further tiering until attribution is resolved.1
Type B: Attribution with Zero Must-Address DisciplineThis type tests whether the agent maintains noise and deeper-diligence discipline even when attribution problems appear alongside signals that might seem urgent on topic alone. Scenarios are constructed so that correct attribution gating results in a must-address count of zero. The skill must teach the agent to resist inflating tiers when no properly attributed, sufficiently evidenced finding actually meets the stated elevation bar.1
Type C: Compound Cluster and Price Condition ElevationThis type tests whether the agent identifies compound signal clusters, where multiple individually ambiguous signals combine to bear on a single price condition or thesis dependency, and whether it elevates correctly and only when all elevation conditions are simultaneously satisfied. The skill must teach the agent to split, link, and re-evaluate signals as a cluster rather than processing each in isolation, and to name the specific price condition the cluster touches.1
Type D: Evidence Sufficiency and CorroborationThis type tests whether the agent assigns the correct three-tier evidence-sufficiency label to each signal by counting independent origins rather than counting URLs or headlines. Scenarios include signals with multiple surface appearances that trace to a single origin, and signals where absence of evidence in a searched register must be distinguished from absence across all registers. The skill must teach the agent to write the label that matches the actual origin count performed, not the label that matches the topic's apparent importance.1
Type E: Walk-Away and Price Condition Dual Must-AddressThis type tests whether the agent correctly elevates absence-of-evidence rows to must-address when two or more stated conditions (a walk-away threshold and a price condition, for example) are simultaneously at risk from the same evidentiary gap. Scenarios present deals with multiple explicit user-stated thresholds and gaps that bear on more than one of them. The skill must teach the agent to evaluate each absence row against every stated condition and to produce separate, clearly justified must-address rows when the gap independently decides each condition.1
Type F: Walk-Away Threshold and Single-Source ElevationThis type tests whether the agent correctly distinguishes between single-source claims that are primary and binding (regulatory filings, docket entries, charging documents) and those that are not, and applies that distinction when a walk-away threshold is in play. Scenarios present high-stakes topics backed by a single non-primary source alongside a stated walk-away condition. The skill must teach the agent to hold such signals at deeper-diligence rather than elevating them to must-address solely because the topic is alarming.1
Type G: Tier Discipline and Cross-Signal IntegrationThis type tests whether the agent keeps tier elevation rare and explicitly justified across a full signal pool, and whether it performs a genuine cross-signal check that catches compound clusters and entity overlaps. Scenarios are constructed so that disciplined tiering results in a low or zero must-address count even when several signals address sensitive topics. The skill must teach the agent to weigh magnitude, probability, and reversibility together, to identify when one signal changes the reading of another, and to explain in the narrative why a seemingly dramatic finding was held at a lower tier.1

Surface details such as target names, deal stages, threshold figures, signal sources, and date ranges rotate across variants while the underlying classification capabilities remain constant.

Datasets

Available Datasets

When you submit, we run your skill through a standardised base agent provided by sundae_bar. The base agent is the same for every submission and is not supplied by you. It receives the challenge's test scenarios one at a time, calls the load_skill tool to read your SKILL.md and the list of your reference files, opens a reference with read_reference when your SKILL.md tells it to, and follows the skill to answer each scenario.

Same base agent, same tests, only the submitted skill changes. Scores therefore reflect the quality of your skill, not agent configuration: the agent's answers are only as good as the procedure your skill teaches.

Graders then score each answer for output quality, correct skill use, and gaming behavior, and the per-scenario scores are averaged into your challenge score.

What is public at open: the scenario inputs, the rubric, and the sha256 hash of the sealed reference set (the same scenarios with their reference outputs and grading notes). Validators grade against the sealed set; the hash on the challenge page is your proof that it cannot change while the challenge runs.

What you see while it runs: your total, a score for every scenario and every grader, and the written notes of the gate checks (dataset derivation, skill alignment, redundancy, novelty). The quality judges' written explanations are held back, because they describe the reference answer.

What is published when the challenge closes: the sealed reference set, every judge explanation, and the full validator outputs for every submission. Check the file against the published hash and re-run any evaluation yourself.

What you submit: a folder or a .zip with SKILL.md at the top level and an optional references/ folder. A single SKILL.md is also accepted. All files are the submission.

Limits: SKILL.md is at most 10,000 bytes and is always loaded. references/ holds up to 30 files, each at most 32,000 characters, 128,000 characters in total. Text formats only: .md, .txt, .json, .yaml, .yml and .csv. Lower-case file names using letters, digits, dots, dashes and underscores. The current values are published by the platform API at GET /api/v2/submissions/limits; the submission form reads them from there.

How references are read: the agent always sees SKILL.md plus a list of your reference file names. It opens a reference only when SKILL.md tells it to, by name. Each read is one harness step within a cap of 10 steps per scenario, and the file's text counts against the scenario's token budget of 30,000 tokens. A reference the agent never opens does nothing for the score.

Rejected at upload: HTML comments, images, scripts, hidden or invisible characters, nested archives, disallowed file types, and anything over a limit. Text addressed to the evaluators, or that tries to override instructions, is accepted but places the entry under review and scores zero.

Important: the base-agent harness is fixed at evaluation time and its available tools are load_skill, read_reference. You submit only skill files; you cannot supply your own agent, add tools, or alter the base agent's configuration.

Tip: write your skill as though you are onboarding a capable new hire who needs a procedure, not scripting a chatbot.

Evaluation Criteria

Each scenario is graded by an LLM judge. Final scores combine four graders and a set of deterministic penalties.

CriteriaWeight

Response quality

An LLM judge scores each response against a task-specific quality bar for this challenge: how correct, complete, and usable the output is on its own terms, independent of which procedure produced it.

30 %

Skill use

An LLM judge evaluates whether the agent's output actually followed the procedures your skill teaches, rather than freelancing or pretending to use the skill.

30 %

Scenario quality

An LLM judge evaluates how well each output meets the per-type quality bar, independent of whether your skill was followed. A reference good output anchors the judge's sense of the quality bar.

30 %

Output originality

A check that your skill drives genuine per-scenario reasoning rather than emitting near-identical, templated responses. Skills that reuse one canned output across scenarios lose credit here.

10 %

Scenarios are scored on the quality of the agent's classification work, and deterministic penalties apply to outputs that damage quality, for example by attributing a signal to the wrong entity, assigning an evidence-sufficiency label that contradicts the origin count, elevating a tier without meeting the stated conditions, or omitting signals from the table. Scenario scores roll up to per-capability-type averages, which combine into the final challenge score shown on the leaderboard.

Submission

How to submit

1
Login or create a sundaebar.ai account

Create an account and add your Bittensor wallet ID to your profile to earn alpha rewards in the challenge.

2
Review the requirements

Review the reference skills (good, mediocre, gaming) to understand the quality bar and calibration points.

3
Build your skill

Draft your SKILL.md (at most 10,000 bytes; it is always loaded) and, if the method needs them, put tables, templates and worked examples in reference files under references/. Tell the agent in SKILL.md which reference to open and when, by file name. Focus on teaching procedure rather than providing templates.

4
Test locally

Run your skill against the published scenario inputs and rubric to sanity-check output quality. The reference outputs are sealed until the challenge closes.

5
Submit via the platform

Submit through the sundae_bar challenge portal by uploading your skill folder, a .zip of it, or a single SKILL.md. The folder has SKILL.md at the top level and an optional references/ folder beside it.

6
Iterate

Review your scored breakdown (a score for every scenario and every grader, plus the gate checks' notes) and resubmit. The judges' written explanations are published when the challenge closes.

Constraints

Submission format: a folder or .zip with SKILL.md (YAML frontmatter with name and description, then a Markdown body) at the top level and an optional references/ folder. A single SKILL.md is also accepted.

Limits: SKILL.md at most 10,000 bytes. Up to 30 reference files, each at most 32,000 characters and 128,000 characters in total; text formats only (.md, .txt, .json, .yaml, .yml and .csv); lower-case file names.

Reference reads: the agent opens a reference only when SKILL.md names it. Each read is one of 10 steps per scenario and its text counts against the scenario's 30,000-token budget.

Base agent: fixed. Provided by sundae_bar.

Runtime: a fixed base-agent harness loads your skill and answers all 10 scenarios per submission with it.

Tests per submission: 10 (the full scenario set, no sampling).

Submission limits: 7 submissions per day. A rolling challenge limit also applies.

Rewards

Emissions accrue on-chain to whichever submission currently holds the top of the leaderboard. There is no fixed per-submission payout: the current leader earns for as long as it stays first.

Time at the top takes all. A strong skill submitted early starts accruing immediately and keeps earning until another submission displaces it. Once you are displaced, accrual stops until you reclaim the top.

Rankings shift as others iterate. Get to the top early, then keep refining your skill to defend the position.

Every file in your submission is scanned at submission time by a skill integrity gate: your SKILL.md and every reference file, whether or not your skill points at it. The gate checks that your skill teaches procedure rather than encoding scenario-specific content from our evaluation pool. Content placed in a reference file is checked exactly as content in SKILL.md.

A skill that fails the integrity gate scores 0 for the submission. The gate runs deterministically and reports its finding as part of your results. This is how the evaluation framework stays credible: a high score should reflect a skill that genuinely teaches procedure, not one that memorised our test set.

Text addressed to the evaluators or judges, or text that tries to override instructions, anywhere in the bundle, is accepted at upload but places the entry under review and scores zero. Files that are not plain text (HTML comments, images, scripts, hidden or invisible characters, nested archives, disallowed file types) or that exceed a limit are rejected at upload. The copy and similarity checks run over the whole bundle, chunk by chunk, so padding a copied skill with extra reference files does not pass them.