How Hallmark Performs a Pre‑Emit Self‑Critique: The Six‑Axis Quality Gate

Hallmark evaluates every generated artifact against six independent quality dimensions before emitting it to the user, automatically revising any content that scores below threshold and stamping the final scores into a persistent comment header.

The Nutlope/hallmark repository implements a rigorous pre-emit self-critique mechanism that intercepts low-quality output before it reaches the user. This quality gate scores artifacts on six axes and enforces minimum standards through automated revision loops, ensuring only refined content passes to the final output stage.

The Six-Axis Scoring Framework

As defined in skills/hallmark/SKILL.md, Hallmark evaluates every artifact on a 1‑to‑5 scale across six independent dimensions:

• Philosophy (P) – Alignment with the brief’s conceptual intent and strategic vision.
• Hierarchy (H) – Logical structuring and intuitive ordering of content sections.
• Execution (E) – Technical correctness, validity, and completeness of implementation details.
• Specificity (S) – Level of concrete detail and actionable guidance provided.
• Restraint (R) – Conciseness and disciplined avoidance of unnecessary fluff or verbosity.
• Variety (V) – Diversity of ideas, formats, or visual treatments employed.

Each axis is scored independently, allowing precision in identifying weak areas without diluting strengths in other categories.

The Revision Trigger and Quality Threshold

The pre‑emit self‑critique enforces a strict minimum quality bar. If any axis receives a score lower than 3, Hallmark automatically enters a revision pass. The system recomposes the artifact iteratively until all six axes meet the minimum threshold of 3. This prevents substandard output from ever reaching the user interface.

Stamping the Critique for Future Discovery

After achieving passing scores, Hallmark records the six values in a single‑line comment at the very top of the generated file. The stamp follows the exact format specified in skills/hallmark/references/slop-test.md:

/* Hallmark · pre-emit critique: P5 H4 E5 S4 R5 V5 */
body {
  margin: 0;
  font-family: system-ui, sans-serif;
}

The comment format is rigid: Hallmark · pre-emit critique: followed by the six axis abbreviations and their integer scores. This persistent header serves as an immutable quality record attached to the artifact itself.

Discovery on Subsequent Runs

When Hallmark processes the same brief again, it scans for existing stamp comments to avoid repeating previous weaknesses. The system uses this discovered critique as a sanity check before generation, effectively remembering past performance and adjusting its approach. This creates a feedback loop where quality history informs future outputs.

You can see this pattern in practice in site/examples/wayfare/style.css, where generated CSS files carry the critique header, and in site/_tests/13-alma/index.html, which demonstrates the stamp in a test HTML environment.

Programmatic Validation

The stamp format enables straightforward programmatic validation. A consumer can verify whether content meets the emission criteria using a regular expression match:

function shouldEmit(content) {
  const stamp = content.match(/\/\*\s*Hallmark\s·\spre-emit\scritique:\sP(\d)\sH(\d)\sE(\d)\sS(\d)\sR(\d)\sV(\d)\s*\*\//);
  if (!stamp) return false;               // no critique → reject
  const scores = stamp.slice(1).map(Number);
  return scores.every(s => s >= 3);        // all axes meet minimum
}

This function returns false if the stamp is missing or if any score falls below the threshold of 3, ensuring that only fully vetted content proceeds to the user.

Summary

• Hallmark performs a pre‑emit self‑critique on every generated artifact before user delivery.
• Content is scored on six axes: Philosophy, Hierarchy, Execution, Specificity, Restraint, and Variety.
• Any score below 3 triggers an automatic revision loop until all axes meet the minimum threshold.
• Final scores are stamped into a standardized comment header following the format defined in skills/hallmark/references/slop-test.md.
• Future runs discover existing stamps to avoid repeating identified weaknesses and maintain quality consistency.

Frequently Asked Questions

What are the six axes of the Hallmark pre‑emit self‑critique?

The six axes are Philosophy (conceptual alignment), Hierarchy (logical structure), Execution (technical correctness), Specificity (detail level), Restraint (conciseness), and Variety (diversity of ideas). Each is scored from 1 to 5, with 3 serving as the minimum acceptable threshold for emission.

What happens if an artifact scores below 3 on any axis?

If any axis receives a score below 3, Hallmark automatically enters a revision pass. The system recomposes the artifact and re-evaluates it against the six axes, repeating this process iteratively until all scores meet or exceed the minimum threshold of 3.

How does Hallmark use the stamp comment on future runs?

When processing a brief that has been generated before, Hallmark scans for existing stamp comments in the artifact. It reads the previous scores to identify historical weaknesses and avoids repeating the same quality deficits, using the stamp as a quick sanity check before any new generation begins.

Where is the pre‑emit self‑critique process documented in the source code?

The process is defined in skills/hallmark/SKILL.md, which establishes the six-axis framework and revision logic. The exact stamp comment format is specified in skills/hallmark/references/slop-test.md. Real-world examples appear in site/examples/wayfare/style.css and site/_tests/13-alma/index.html.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}