Evaluating OCR-to-Markdown Systems Is Fundamentally Broken (and Why That’s Hard to Fix)

Evaluating OCR methods that convert PDFs or doc photographs into Markdown is way extra complicated than it seems. Not like plain textual content OCR, OCR-to-Markdown requires fashions to get better content material, format, studying order, and illustration decisions concurrently. As we speak’s benchmarks try to attain this with a mixture of string matching, heuristic alignment, and format-specific guidelines—however in apply, these approaches routinely misclassify appropriate outputs as failures.

This put up outlines why OCR-to-Markdown analysis is inherently underspecified, examines widespread analysis methods and their failure modes, highlights concrete points noticed in two extensively used benchmarks, and explains why LLM-as-judge is presently probably the most sensible technique to consider these methods—regardless of its imperfections .

Why OCR-to-Markdown Is Laborious to Consider

At its core, OCR-to-Markdown doesn’t have a single appropriate output.

A number of outputs will be equally legitimate:

• Multi-column layouts will be linearized in numerous studying orders.
• Equations will be represented utilizing LaTeX, Unicode, HTML, or hybrids.
• Headers, footers, watermarks, and marginal textual content could or will not be thought-about “content material” relying on job intent.
• Spacing, punctuation, and Unicode normalization usually differ with out affecting which means.

From a human or downstream-system perspective, these outputs are equal. From a benchmark’s perspective, they usually should not.

Frequent Analysis Methods and Their Limitations

1. String-Primarily based Metrics (Edit Distance, Actual Match)

Most OCR-to-Markdown benchmarks depend on normalized string comparability or edit distance.

Limitations

• Markdown is handled as a flat character sequence, ignoring construction.
• Minor formatting variations produce massive penalties.
• Structurally incorrect outputs can rating properly if textual content overlaps.
• Scores correlate poorly with human judgment.

These metrics reward formatting compliance reasonably than correctness.

2. Order-Delicate Block Matching

Some benchmarks section paperwork into blocks and rating ordering and proximity.

Limitations

• Legitimate various studying orders (e.g., multi-column paperwork) are penalized.
• Small footer or marginal textual content can break strict ordering constraints.
• Matching heuristics degrade quickly as format complexity will increase.

Right content material is usually marked mistaken attributable to ordering assumptions.

3. Equation Matching through LaTeX Normalization

Math-heavy benchmarks sometimes anticipate equations to be rendered as full LaTeX.

Limitations

• Unicode or partially rendered equations are penalized.
• Equal LaTeX expressions utilizing completely different macros fail to match.
• Combined LaTeX/Markdown/HTML representations should not dealt with.
• Rendering-correct equations nonetheless fail string-level checks.

This conflates illustration alternative with mathematical correctness.

4. Format-Particular Assumptions

Benchmarks implicitly encode a most popular output type.

Limitations

• HTML tags (e.g., ) trigger matching failures.
• Unicode symbols (e.g., km²) are penalized in opposition to LaTeX equivalents.
• Spacing and punctuation inconsistencies in floor fact amplify errors.

Fashions aligned to benchmark formatting outperform extra common OCR methods.

Points Noticed in Present Benchmarks

Benchmark A: olmOCRBench

Guide inspection reveals that a number of subsets embed implicit content material omission guidelines:

• Headers, footers, and watermarks which are visibly current in paperwork are explicitly marked as absent in floor fact.
• Fashions skilled to extract all seen textual content are penalized for being appropriate.
• These subsets successfully consider selective suppression, not OCR high quality.

Moreover:

• Math-heavy subsets fail when equations should not totally normalized LaTeX.
• Right predictions are penalized attributable to illustration variations.

Consequently, scores strongly depend upon whether or not a mannequin’s output philosophy matches the benchmark’s hidden assumptions.

Instance 1

For the above picture, Nanonets-OCR2 appropriately predicts the watermark to the suitable facet of the picture, however within the floor fact annotation penalizes the mannequin for predicting it appropriately.

{
"pdf": "headers_footers/ef5e1f5960b9f865c8257f9ce4ff152a13a2559c_page_26.pdf", 
"web page": 1, 
"id": "ef5e1f5960b9f865c8257f9ce4ff152a13a2559c_page_26.pdf_manual_01", 
"kind": "absent", 
"textual content": "Doc tu00e9lu00e9chargu00e9 depuis www.cairn.data - Universitu00e9 de Marne-la-Vallu00e9e - - 193.50.159.70 - 20/03/2014 09h07. u00a9 S.A.C.", "case_sensitive": false, "max_diffs": 3, "checked": "verified", "first_n": null, "last_n": null, "url": ""}

Sort absent signifies that within the prediction information, that textual content shouldn’t be current.

Instance 2

The benchmark additionally doesn’t take into account texts which are current within the doc footer.

Instance on this doc, the Alcoholics Namelessu00ae and www.aa.org shouldn’t be current within the doc in accordance with the ground-truth, which is wrong

{
	"pdf": "headers_footers/3754542bf828b42b268defe21db8526945928834_page_4.pdf", 
	"web page": 1, 
	"id": "3754542bf828b42b268defe21db8526945928834_page_4_header_00", 
	"kind": "absent", 
	"max_diffs": 0, 
	"checked": "verified", 
	"url": "", 
	"textual content": "Alcoholics Namelessu00ae", 
	"case_sensitive": false, "first_n": null, "last_n": null
	}
{
	"pdf": "headers_footers/3754542bf828b42b268defe21db8526945928834_page_4.pdf", 
	"web page": 1, 
	"id": "3754542bf828b42b268defe21db8526945928834_page_4_header_01", 
	"kind": "absent", 
	"max_diffs": 0, 
	"checked": "verified", 
	"url": "", 
	"textual content": "www.aa.org", 
	"case_sensitive": false, "first_n": null, "last_n": null}

Benchmark B: OmniDocBench

OmniDocBench reveals related points, however extra broadly:

• Equation analysis depends on strict LaTeX string equivalence.
• Semantically an identical equations fail attributable to macro, spacing, or image variations.
• Quite a few ground-truth annotation errors have been noticed (lacking tokens, malformed math, incorrect spacing).
• Unicode normalization and spacing variations systematically scale back scores.
• Prediction choice heuristics can fail even when the right reply is totally current.

In lots of circumstances, low scores replicate benchmark artifacts, not mannequin errors.

Instance 1

Within the instance above, the Nanonets-OCR2-3B predicts 5 g silica + 3 g Al$_2$O$_3$ however the floor fact expects as $ 5g mathrm{ s i l i c a}+3g mathrm{ A l}*{2} mathrm{O*{3}} $ . This flags the mannequin prediction as incorrect, even when each are appropriate.

Full Floor Reality and Prediction, and the take a look at case shared beneath:

'pred': 'The collected eluant was concentrated by rotary evaporator to 1 ml. The extracts have been lastly handed by way of a remaining column crammed with 5 g silica + 3 g Al$_2$O$_3$ to take away any co-extractive compounds which will trigger instrumental interferences durin the evaluation. The extract was eluted with 120 ml of DCM:n-hexane (1:1), the primary 18 ml of eluent was discarded and the remaining have been collected, which comprises the analytes of curiosity. The extract was exchanged into n-hexane, concentrated to 1 ml to which 1 μg/ml of inside customary was added.'
'gt': 'The collected eluant was concentrated by rotary evaporator to 1 ml .The extracts have been lastly handed by way of a remaining column crammed with $ 5g mathrm{ s i l i c a}+3g mathrm{ A l}*{2} mathrm{O*{3}} $ to take away any co-extractive compounds which will trigger instrumental
interferences in the course of the evaluation. The extract was eluted with 120 ml of DCM:n-hexane (1:1), the primary 18 ml of eluent was discarded and the remaining have been collected, which comprises the analytes of curiosity. The extract was exchanged into n - hexane, concentrated to 1 ml to which $ mumathrm{g / ml} $ of inside customary was added.'

Instance 2

We discovered considerably extra incorrect annotations with OmniDocBench

Within the ground-truth annotation 1 is lacking in 1 ml .

'textual content': 'The collected eluant was concentrated by rotary evaporator to 1 ml .The extracts have been lastly handed by way of a remaining column crammed with $ 5g mathrm{ s i l i c a}+3g mathrm{ A l}*{2} mathrm{O*{3}} $ to take away any co-extractive compounds which will trigger instrumental interferences in the course of the evaluation. The extract was eluted with 120 ml of DCM:n-hexane (1:1), the primary 18 ml of eluent was discarded and the remaining have been collected, which comprises the analytes of curiosity. The extract was exchanged into n - hexane, concentrated to 1 ml to which $ mumathrm{g / ml} $ of inside customary was added.'

Why LLM-as-Choose Is the Least-Dangerous Choice As we speak

Given these limitations, LLM-as-judge is presently probably the most sensible technique to consider OCR-to-Markdown methods.

This isn’t as a result of LLM judges are good—however as a result of the issue is basically semantic.

What LLM-as-Choose Handles Effectively

1. Semantic Equivalence Throughout Representations
   LLMs can acknowledge that:
   • LaTeX, Unicode, and HTML equations will be equal
   • Macro-level variations (A^T vs mathbf{A}^T) don’t change which means
   • Spacing and normalization variations are irrelevant
2. Versatile Studying Order Reasoning
   LLMs can assess whether or not content material is full even when:
   • Sections are reordered
   • Multi-column layouts are linearized in another way
3. Context-Conscious Content material Inclusion
   LLMs can cause about whether or not:
   • Footers, headers, or watermarks ought to moderately be included
   • Textual content inside logos or figures counts as content material
4. Tolerance to Annotation Noise
   When floor fact is incomplete or incorrect, LLMs can nonetheless choose correctness relative to the doc, reasonably than blindly imposing flawed annotations.

Why Metric Engineering Doesn’t Scale

Many benchmark failures are addressed by:

• Including normalization guidelines
• Increasing equivalence lessons
• Introducing heuristic margins

These fixes don’t generalize. Each new doc kind—scientific papers, scanned books, multilingual PDFs, types—introduces new edge circumstances. LLMs generalize throughout these circumstances with out task-specific rule engineering.

Acknowledged Limitations of LLM-as-Choose

LLM-based analysis has actual drawbacks:

• Non-determinism
• Sensitivity to immediate design
• Increased value and latency
• Decreased reproducibility in comparison with static scripts

Nonetheless, these are operational limitations, not conceptual ones. In distinction, string- and rule-based metrics are conceptually misaligned with the duty itself.

Remaining Takeaway

OCR-to-Markdown analysis is underspecified by nature. Present benchmarks conflate formatting, illustration decisions, and semantic correctness—usually penalizing fashions for being appropriate in methods the benchmark didn’t anticipate.

Till benchmarks explicitly embrace semantic equivalence, LLM-as-judge stays the closest approximation to human judgment and probably the most dependable analysis sign accessible right now. Benchmark scores ought to subsequently be handled as partial indicators, not definitive measures of OCR high quality.