Per-table card variant · Experimental
Labeled-faithfulness card
Full v0.6.1 card with every section explicitly tagged by provenance type (verbatim-cells, inherited, summarized, inferred).
In one paragraph
Same content as v0.6.1 but every section header carries an explicit `[VERBATIM-*]`, `[INHERITED]`, `[SUMMARIZED]`, or `[INFERRED]` tag. Tests whether explicit provenance labels improve a model's ability to refuse correctly on inferred-scope claims rather than accepting them as verbatim data.
How the inputs are generated
Generation · 01Generator script
evaluation_runs/generate_card_variants.py:render_labeledInput sources
- • pipeline-v0.6.1 cards
AI use
No — pure deterministic transformation
OCR / re-OCR
Inherits from the upstream pipeline variant
Approximate processing time
~2 seconds for all 407 cards.
Resource intensity
Low — CPU-only post-processing, runs in seconds
Determinism
Deterministic (same input → same output, byte-identical)
Output location
card_sets/pipeline-v0.7-labeled/Cards produced
407 cards
Introduced
v0.7 variant family, 2026-05-22.
Evaluation results
Diagnostic · 02Avg open-tier pass rate
49%
Typical card size
~2.4 KB per card
Evaluation cycle
Cycle 18
Relative to v0.6.1 baseline
+22 percentage points open-tier versus v0.6.1 baseline