Treatment Log

Field Definitions

fixApplied.type Values

Storage Location

{slide_tmpdir}/treatment-{index:03d}-attempt-{n}.json

Where:

• {slide_tmpdir} is the per-slide cache directory (e.g. {cache_dir}/slide-002/)
• {index:03d} is the 0-based slide index zero-padded to 3 digits
• {n} is the 1-based attempt number

Example: {cache_dir}/slide-002/treatment-002-attempt-1.json

Promotion Criteria

A fix is a promotion candidate (promotionCandidate.recommend = true) if:

1. The fix was required on 2 or more distinct decks, OR
2. The triage finding had confidence < 0.75 AND the model-resolved type differed from the classify_elements() default type.

A fix is not a promotion candidate if:

• It was an EDL geometry tweak unique to this deck's specific layout
• It addressed a typo or content error in the source HTML
• It was marked REVIEW (no fix possible — inherent format limitation)

Promotion Process

Promotion is a manual step — the model reads treatment log entries and updates known-patterns.md with a new entry.

To promote a candidate:

1. Read all treatment log entries for the candidate pattern (across decks).
2. Identify the common observable signature (CSS class, rect dimensions, relative position, element relationship).
3. Verify the fix is deterministic: the same geometry transform should apply to any future occurrence of the signature.
4. Write a new entry in known-patterns.md following the entry format.
5. Update the PATTERN-NNN ID (next sequential ID after existing entries).
6. Optionally update classify_elements() confidence scores if the pattern indicates the classifier is systematically wrong about element type.
7. Update the treatment log entry with promotionCandidate.promote = true (to distinguish "recommended" from "completed").

Example: Recording a Treatment Log Entry

After render-compare attempt 1 on slide 2:

import json
from datetime import datetime, timezone

entry = {
    "schemaVersion": 1,
    "deckSlug": "sales-kickoff",
    "slideIndex": 2,
    "sourceFile": "slide-03.html",
    "attempt": 1,
    "timestamp": datetime.now(timezone.utc).isoformat(),

    "triageSummary": {
        "flaggedCount": 2,
        "patterns": ["PATTERN-002"],
        "complexity": "low"
    },

    "prepSummary": {
        "autoResolved": 18,
        "modelResolved": 2,
        "modelReasonings": [
            "e-014: badge (not richtext) — small emoji circle",
            "e-021: skip — flex container"
        ]
    },

    "buildSummary": {
        "shapesBuilt": 12,
        "richtextBuilt": 8,
        "svgsBuilt": 1,
        "tablesBuilt": 0
    },

    "renderCompareVerdict": "FIX",
    "issueCategories": ["text_error"],
    "issueDescription": "Stat '42%' overlaps badge icon at x=40",

    "fixApplied": {
        "type": "edl",
        "description": "Moved stat text box 54px right to clear badge"
    },

    "nextAttemptVerdict": "PASS",

    "promotionCandidate": {
        "recommend": True,
        "patternName": "stat_badge_text_offset",
        "signature": "numeric richtext adjacent to small badge, same row",
        "suggestedFix": "shift x to badge.right + 8; reduce w by same amount"
    }
}

log_path = f"{slide_tmpdir}/treatment-002-attempt-1.json"
with open(log_path, 'w') as f:
    json.dump(entry, f, indent=2)

Reference Files