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

Evaluating OCR programs that convert PDFs or doc photos into Markdown is much 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 selections concurrently. At this time’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 broadly used benchmarks, and explains why LLM-as-judge is at the moment essentially the most sensible technique to consider these programs—regardless of its imperfections .

Why OCR-to-Markdown Is Exhausting to Consider

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

A number of outputs might be equally legitimate:

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

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

Widespread Analysis Methods and Their Limitations

1. String-Based mostly 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 giant penalties.
• Structurally incorrect outputs can rating effectively if textual content overlaps.
• Scores correlate poorly with human judgment.

These metrics reward formatting compliance fairly 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 commonly marked unsuitable as a consequence of ordering assumptions.

3. Equation Matching through LaTeX Normalization

Math-heavy benchmarks usually count on 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.
• Blended LaTeX/Markdown/HTML representations are usually not dealt with.
• Rendering-correct equations nonetheless fail string-level checks.

This conflates illustration selection with mathematical correctness.

4. Format-Particular Assumptions

Benchmarks implicitly encode a most popular output model.

Limitations

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

Fashions aligned to benchmark formatting outperform extra common OCR programs.

Points Noticed in Current Benchmarks

Benchmark A: olmOCRBench

Handbook 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 are usually not totally normalized LaTeX.
• Right predictions are penalized as a consequence of illustration variations.

In consequence, scores strongly rely 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 appropriate aspect 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", 
"sort": "absent", 
"textual content": "Doc tu00e9lu00e9chargu00e9 depuis www.cairn.information - 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 implies that within the prediction information, that textual content shouldn’t be current.

Instance 2

The benchmark additionally doesn’t contemplate 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 based on the ground-truth, which is wrong

{
	"pdf": "headers_footers/3754542bf828b42b268defe21db8526945928834_page_4.pdf", 
	"web page": 1, 
	"id": "3754542bf828b42b268defe21db8526945928834_page_4_header_00", 
	"sort": "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", 
	"sort": "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 comparable points, however extra broadly:

• Equation analysis depends on strict LaTeX string equivalence.
• Semantically an identical equations fail as a consequence of 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 cut back scores.
• Prediction choice heuristics can fail even when the proper 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 Fact and Prediction, and the check case shared beneath:

'pred': 'The collected eluant was concentrated by rotary evaporator to 1 ml. The extracts have been lastly handed by means of a last column crammed with 5 g silica + 3 g Al$_2$O$_3$ to take away any co-extractive compounds that 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 incorporates the analytes of curiosity. The extract was exchanged into n-hexane, concentrated to 1 ml to which 1 μg/ml of inside commonplace was added.'
'gt': 'The collected eluant was concentrated by rotary evaporator to 1 ml .The extracts have been lastly handed by means of a last 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 that will trigger instrumental
interferences through 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 incorporates the analytes of curiosity. The extract was exchanged into n - hexane, concentrated to 1 ml to which $ mumathrm{g / ml} $ of inside commonplace 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 means of a last 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 that will trigger instrumental interferences through 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 incorporates the analytes of curiosity. The extract was exchanged into n - hexane, concentrated to 1 ml to which $ mumathrm{g / ml} $ of inside commonplace was added.'

Why LLM-as-Decide Is the Least-Unhealthy Possibility At this time

Given these limitations, LLM-as-judge is at the moment essentially the most sensible technique to consider OCR-to-Markdown programs.

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

What LLM-as-Decide Handles Nicely

1. Semantic Equivalence Throughout Representations
   LLMs can acknowledge that:
   • LaTeX, Unicode, and HTML equations might be equal
   • Macro-level variations (A^T vs mathbf{A}^T) don’t change that 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 a different way
3. Context-Conscious Content material Inclusion
   LLMs can motive about whether or not:
   • Footers, headers, or watermarks ought to fairly 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 decide correctness relative to the doc, fairly 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 sort—scientific papers, scanned books, multilingual PDFs, kinds—introduces new edge circumstances. LLMs generalize throughout these circumstances with out task-specific rule engineering.

Acknowledged Limitations of LLM-as-Decide

LLM-based analysis has actual drawbacks:

• Non-determinism
• Sensitivity to immediate design
• Greater price and latency
• Diminished reproducibility in comparison with static scripts

Nevertheless, 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. Current benchmarks conflate formatting, illustration selections, and semantic correctness—typically 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 essentially the most dependable analysis sign obtainable immediately. Benchmark scores ought to due to this fact be handled as partial indicators, not definitive measures of OCR high quality.