assay

Scores Discovery Harness backlog candidates for viability before Spike or APM dispatch.

1|Updated Dec 13, 2025
One-click install
npx skills add https://github.com/ZK-Theory/TDL --skill assay-zk-theory
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: assay
Source: https://github.com/ZK-Theory/TDL/tree/main/.agents/skills/assay
Command: npx skills add https://github.com/ZK-Theory/TDL --skill assay-zk-theory

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Research backlog candidates can consume expensive compute on Spikes or APM worktrees before anyone checks whether the idea is viable. This Skill provides a cheap front-door viability gate that scores triaged Discovery Harness candidates on topology justification, data feasibility, and novelty, producing a PROMOTE, PARK, or KILL decision before any compute is spent. ## Core Features & Use Cases - Adversarial Topology Gate: Axis 1 forces the scorer to argue the null hypothesis, requiring a genuine metric space, a falsifiable feature-to-claim mapping, and a named non-TDA baseline before PROMOTE is possible. - Three-Axis Scorecard: Produces a machine-readable assay_scorecard YAML block with Axis 1 gate results plus 0-3 integer scores for data feasibility and novelty/publishability. - Contract Validation: Validates scorecard blocks with extract_scorecard_block and validate_assay_scorecard, including explicit iteration for multi-application notes. - Use Case: After /scout-review marks a candidate state: triaged in vault/00-Meta/Discovery/_backlog.md, run the assay to decide whether it earns a Spike, gets parked for more evidence, or is killed at zero compute cost. ## Quick Start Assay the triaged candidate in the Discovery backlog and write its scorecard note with a PROMOTE, PARK, or KILL decision.

Frequently Asked Questions about assay

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I score a Discovery Harness backlog candidate before running a Spike?

Run the assay on any candidate marked `state: triaged` in `vault/00-Meta/Discovery/_backlog.md`. It applies the three-axis rubric, writes a candidate note with a machine-readable scorecard block, and updates the backlog with the decision and next action.

What criteria decide PROMOTE, PARK, or KILL in an assay scorecard?

PROMOTE requires the Axis 1 topology gate to pass and Axis 2 plus Axis 3 scores to total at least 4 with neither at 0. KILL applies when the gate fails or a decisive red flag exists; PARK applies when evidence is insufficient.

What is the Axis 1 topology-earns-its-keep gate?

Axis 1 is an adversarial gate requiring a genuine metric space, a topological feature mapped to a falsifiable claim, a named non-TDA baseline, and a claim not reducible to clustering, PCA, GMM, or ordinary regression. If it fails, the candidate is killed at zero compute cost.

How do I validate an assay scorecard block?

Use `extract_scorecard_block` and `validate_assay_scorecard` from `trajectory_tda.discovery.assay_scorecard` against the candidate note. For multi-application notes, iterate and validate every fenced block explicitly since the extractor returns only the first.

When should a candidate be PARKed instead of KILLed?

PARK when evidence is insufficient, such as an abstract too thin to evaluate Axis 1, naming the missing evidence rather than guessing. KILL is reserved for gate failures, inaccessible data, or decisive red flags backed by directly verified citations.