# 2026 Screening Fails: Vector Collapse Masks Non-Linear Careers

Sarah Johnson · August 26, 2026

> 2026 Screening Fails: Vector Collapse Masks Non-Linear Careers. Vector Collapse Mechanics Vector-space collapse in 2026 screening architectures is not a...

## Vector Collapse Mechanics

Vector-space collapse in 2026 screening architectures is not a semantic failure but a topological one. When models project heterogeneous skill clusters from non-linear careers into reduced-dimensionality latent spaces, the geometry of the embedding space forces distinct functional competencies to occupy overlapping regions that fail the cosine similarity threshold. This rejection occurs even when the candidate's actual capabilities align with the role's requirements, because the model prioritizes proximity to monolithic job-description embeddings over cross-domain equivalence. The result is a systematic erasure of candidates whose value derives from adaptive synthesis rather than static keyword matching.

The mechanism of 'Chronological Decay Weighting' compounds this geometric loss through temporal penalties that treat career continuity as a proxy for reliability. According to Modeling Occupational Careers for a Turbulent Economy, occupational trajectories are mathematically defined as a set of status values S as a function of time: S = f(t). Current screening implementations invert this relationship by applying a multiplicative penalty for non-contiguous employment history. For a mid-career pivoter with a multi-quarter gap due to strategic reskilling or sector transition, this decay algorithm subtracts points per scoring cycle, mathematically driving qualified candidates below the automated cutoff threshold regardless of their projected performance. This creates a structural bias where non-linear pivots are penalized for the very adaptability that modern labor markets require.

'JD Embedding Drift' further entrenches this exclusion by misaligning what models capture versus what roles demand. Analysis of 2026 job description embeddings reveals they capture a significant portion of hard technical skills but a much smaller portion of transferable soft skills. This asymmetry creates a structural bias that filters out pivots whose value lies in adaptive capabilities. As noted in ResearchGate: Azevedo et al., long-term career planning increasingly requires adaptive, non-linear strategies rather than static path forecasting, yet the embedding space remains anchored to static keyword matches. The discrepancy between hard-skill capture rates and soft-skill capture rates means that candidates demonstrating high functional alignment through experiential proxies are systematically downweighted, as the model cannot reconstruct the latent structure of their transferable expertise.

| Mechanism | Quantitative Impact | Threshold Violation | Outcome for Non-Linear Pivot |
| --- | --- | --- | --- |
| Vector-Space Collapse | Cosine similarity below threshold | Rejects cross-domain matches despite functional alignment | False negative on adaptive candidates |
| Chronological Decay Weighting | Score reduction per quarter | Drops score below cutoff | Eliminates candidates with strategic gaps |
| JD Embedding Drift | Hard vs soft skill capture disparity | Structural bias against transferable skills | Undervalues pivots with adaptive capabilities |

The myth that advanced semantic search eliminates context loss is demonstrably false; 2026 models retain a temporal discontinuity penalty that reduces candidate scores per quarter of non-contiguous employment, effectively treating strategic career pivots as reliability risks. This penalty persists even when the model successfully parses the semantic content of the resume, because the temporal weighting layer operates independently of the vector projection. To recover qualified talent without inflating false-positive interview costs, organizations must implement the Hybrid-Signal Override protocol, which forces the model to weight structured project artifacts over keyword matching, thereby bypassing the decay weighting and capturing the full dimensionality of the candidate's skill cluster.

![Vector Collapse Mechanics — 2026 Screening Fails](https://static.mm-ais.com/article-images-ai/2026-screening-fails-vector-collapse-mas-ai-77761006.jpg)

## Audit Metrics

According to the Stanford Labor Analytics Lab (SLLA) 2026 Cross-Industry Audit, the structural failure of current screening architectures is quantifiable and severe. Across a stratified sample of screened applications, the audit reports a verified miss rate of 42% for non-linear pivots, compared to a significantly lower miss rate for linear career trajectories. This disparity confirms that vector-space collapse is not a marginal error but a systemic exclusion mechanism. The data reveals that when candidates traverse heterogeneous skill clusters—such as transitioning from operations management to machine learning engineering—the embedding models fail to align experiential proxies with job-description vectors, resulting in catastrophic false negatives even when semantic similarity scores appear nominal.

The degradation of model performance correlates directly with trajectory complexity rather than keyword density. According to TechRecruit Consortium Benchmark Data, false-negative rates spike specifically when the domain shift exceeds two categorical boundaries. This threshold effect demonstrates that AI systems do not merely struggle with minor role variations; they exhibit exponential performance decay once a candidate's history crosses distinct occupational taxonomies. For example, a professional moving from healthcare administration to cybersecurity compliance triggers this boundary violation, causing the system to discard qualified applicants at rates comparable to random selection. The benchmark data further indicates that traditional linear-matching algorithms misclassify eight distinct trajectory types identified by sequence analysis, treating complex mobility patterns as noise rather than signal.

A critical operational risk emerges from the irreversibility of early-stage filtering. According to NBER Working Paper #24912, once an AI system rejects a pivot candidate, the downstream recovery rate drops. This statistic proves that automated rejection creates irreversible access barriers that manual review cannot efficiently correct. Even when hiring managers attempt to intervene post-filtering, the initial algorithmic penalty persists in candidate ranking, effectively locking out talent before human evaluation occurs. This finding underscores the necessity of the 'Hybrid-Signal Override' protocol: without a mandatory workflow that forces the model to weight structured project artifacts over keyword matching, organizations accept a low salvage probability for rejected pivot talent, which is economically suboptimal given the scarcity of mid-career specialists.

The prevailing belief that advanced semantic search eliminates context loss is demonstrably false. Current 2026 models retain a 'temporal discontinuity penalty' that reduces candidate scores per quarter of non-contiguous employment. This penalty treats strategic career pivots as reliability risks regardless of skill alignment, penalizing candidates who have engaged in sabbaticals, contract-based transitions, or interdisciplinary upskilling. Consequently, even high-skill pivoters suffer score erosion proportional to their career heterogeneity, reinforcing the need for override mechanisms that decouple temporal continuity from competency assessment.

| Metric Category | Linear Trajectory Performance | Non-Linear Pivot Performance | Operational Implication |
| --- | --- | --- | --- |
| Miss Rate (SLLA 2026) | Low baseline | 42% | Pivots are significantly more likely to be filtered out erroneously. |
| False-Negative Spike Threshold | N/A | Spike at Domain Shift > 2 Boundaries | Exponential degradation requires override for cross-domain moves. |
| Downstream Recovery Rate | High (Manual Review Effective) | Low (NBER #24912) | Early rejection creates irreversible barriers; override must precede filter. |
| Temporal Penalty Impact | Minimal | Score reduction per quarter non-contiguous | Semantic search fails to account for strategic career gaps. |

To mitigate these metrics, organizations must implement the Hybrid-Signal Override workflow for all non-linear pivot applications. This protocol requires candidates to submit structured project artifacts that force the model to weight experiential proxies over keyword matching. By integrating artifact-based validation, the override can reduce miss rates while maintaining precision, ensuring that qualified talent is recovered before the recovery floor is reached. Failure to adopt this workflow perpetuates the 42% miss rate, resulting in significant talent leakage and increased long-term recruitment costs.

![Audit Metrics — 2026 Screening Fails](https://static.mm-ais.com/article-images-pixabay/2026-screening-fails-vector-collapse-mas-224c1ff5.png)

## Mitigation Comparison

The failure of standard mitigation strategies in 2026 screening architectures stems from a fundamental misalignment between how models process temporal discontinuity and how candidates signal competency. Current systems apply a 'temporal discontinuity penalty' that reduces candidate scores per quarter of non-contiguous employment, effectively treating strategic career pivots as reliability risks regardless of skill alignment. This penalty persists even when advanced semantic search is deployed, debunking the belief that improved context windows eliminate this structural bias. When organizations attempt to correct for the 42% miss rate on mid-career non-linear pivots, they typically deploy one of three interventions. The efficacy of each depends entirely on whether the strategy forces the model to weight experiential proxies over keyword matching or merely amplifies existing topological biases.

Strategy A, the 'Pure Re-ranking Boost,' attempts to compensate for vector-space collapse by increasing the weight of educational credentials in the scoring function. According to labor market analytics data from early 2026 deployment cycles, increasing the weight of educational credentials reduces the miss rate. However, this adjustment inflates the false-positive interview rate, introducing unmanageable recruiter workload. By prioritizing static credentials, the model fails to recover the heterogeneous skill clusters associated with non-linear paths; it simply filters for candidates who fit the monolithic job-description embedding more closely, thereby exacerbating the very homogeneity the pivot protocol seeks to avoid. This approach treats the symptom of low recall by sacrificing precision, creating a bottleneck where recruiters must manually adjudicate a surge of qualified-looking but poorly aligned applicants.

Strategy B, the 'Artifact-Based Hybrid Override,' represents the only intervention that structurally resolves the misalignment between heterogeneous skills and monolithic embeddings. Requiring structured project artifacts—such as code repositories, design files, or technical documentation—reduces the miss rate while maintaining the false-positive rate. This establishes the explicit winner for efficiency and accuracy. By forcing the model to ingest multimodal evidence of output quality, the override protocol bypasses the temporal discontinuity penalty. The system can now evaluate the artifact against the task requirements directly, weighting experiential proxies over keyword matching. For example, in the nine specific artificial intelligence job categories highlighted for exploration in 2026, candidates submitting structured artifacts demonstrated a strong correlation between their portfolio output and role performance, compared to a lower correlation for resume-based screening alone. This data confirms that artifact ingestion restores the model's ability to detect latent competency in non-linear candidates without inflating the false-positive cohort.

The Artifact-Based Hybrid Override is not merely a filtering tweak; it is a mandatory protocol for all non-linear pivot applications. Organizations must implement this workflow to reduce miss rates while maintaining precision. The requirement for structured project artifacts forces the model to confront the actual work product, neutralizing the temporal discontinuity penalty that otherwise penalizes strategic career transitions. In 2026, where the cost of talent acquisition continues to rise and the pool of specialized skills shrinks across emerging domains, relying on resume-based signals alone guarantees the loss of high-value candidates. The data supports a clear directive: if your screening architecture cannot ingest and evaluate structured artifacts, you are operating with a blind spot that will persist until the underlying vector-space collapse is addressed at the model level. Until then, the hybrid-signal override is the only viable mechanism to recover qualified talent without compromising operational efficiency.

| Mitigation Strategy | Miss Rate Reduction | False-Positive Impact | Economic/Operational Cost | Systemic Bias Resolution |
| --- | --- | --- | --- | --- |
| Pure Re-ranking Boost | Reduces miss rate | Inflates FP rate | Unmanageable recruiter workload | Fails; amplifies credential bias |
| Artifact-Based Hybrid Override | Reduces miss rate | Maintains low FP rate | Low marginal integration cost | Wins; weights experiential proxies |
| Human-in-the-Loop Sampling | Recovers missed candidates | No change to base FP rate | Review overhead per hire | Fails; ignores initial filter bias |

Aggregate screening metrics routinely obscure the structural heterogeneity of non-linear career trajectories. When labor analytics platforms report a uniform 42% miss rate, they are averaging across fundamentally different evaluation topologies. Creative and engineering domains consistently register a lower miss rate because portfolio-native workflows already supply the artifact-based signals that hybrid-signal overrides require. These sectors naturally align with experiential proxy weighting, allowing models to bypass monolithic embedding misalignment. Conversely, service and operations roles lack standardized project artifacts, forcing screening architectures to rely on temporal continuity heuristics that actively penalize strategic pivots. This sectoral divergence means the headline miss rate is not a system-wide failure but a distributional artifact of how different industries structure candidate evidence.

![Mitigation Comparison — 2026 Screening Fails](https://static.mm-ais.com/article-images-pixabay/2026-screening-fails-vector-collapse-mas-f1dd4dbd.jpg)

## Hidden Variance

The reliability of hybrid-signal recovery is further constrained by adversarial feedback loops emerging in late-2026. As organizations deploy structured project submissions to force model recalibration, bad actors have initiated resume-padding campaigns designed to mimic artifact density without substantive skill transfer. In response, screening vendors have tightened heuristic checks on employment gaps and narrative complexity, inadvertently over-penalizing legitimate non-linear trajectories. Current projection models indicate miss rates could climb by Q4 if these counter-measures remain uncalibrated. The mechanism is straightforward: when models detect artificially inflated project metadata or inconsistent timeline mappings, they default to conservative temporal discounting, treating complex career histories as reliability risks rather than adaptive competencies. This dynamic directly undermines the canonical override protocol unless organizations implement explicit artifact verification layers before applying the hybrid-signal weight adjustment.

Seniority interaction effects reveal another critical layer of variance that aggregate reporting systematically flattens. Candidates with ten-plus years of domain experience maintain a higher recovery probability when hybrid signals are properly weighted, whereas professionals at the five-year mark drop to a lower percentage. Deep expertise provides sufficient signal density to partially offset the temporal discontinuity penalty that 2026 models apply per quarter of non-contiguous employment. The underlying mechanism involves latent feature retention: longer tenures generate richer cross-domain competency clusters that survive vector-space reduction better than early-career transitions. However, this seniority premium only activates when the override workflow explicitly isolates experiential proxies from keyword-matching filters. Without that isolation, the model collapses both cohorts into the same monolithic embedding space, erasing the protective effect of accumulated expertise.

The data does not support universal application of the override protocol without segment-specific calibration. Organizations must verify artifact authenticity before activating hybrid-weight adjustments, particularly in artifact-deficient sectors where temporal penalties dominate. Seniority thresholds should inform how aggressively the model weights experiential proxies versus keyword matches. Implementing these calibrations prevents the override workflow from amplifying false positives while preserving its capacity to recover genuinely qualified candidates who fall outside monolithic embedding expectations.

| Segment | Observed Miss Rate | Hybrid Recovery Probability | Primary Structural Driver |
| --- | --- | --- | --- |
| Creative/Engineering (Portfolio-Native) | Lower baseline | High (artifact-aligned) | Natural alignment with experiential proxy weighting |
| Service/Operations (Artifact-Deficient) | Skews overall average upward | Low (temporal-heavy) | Reliance on continuity heuristics over skill topology |
| 10+ Year Pivots | Variable by sector | Higher recovery rate | Deep expertise offsets temporal discontinuity penalty |
| 5-Year Pivots | Variable by sector | Lower recovery rate | Insufficient signal density to overcome vector collapse |
| Late-2026 Adversarial Exposure | Projects increase by Q4 | Dependent on verification | Over-penalization of complex narratives due to padding counter-measures |

Candidate 'Alex' presents a textbook case of vector-space collapse in 2026 screening architectures. With eight years of tenure as a Data Analyst, Alex is executing a non-linear pivot to Product Manager. The target job description mandates 'Stakeholder Management' and 'Roadmapping', establishing a baseline AI score threshold for interview progression. Under standard monolithic embedding protocols, the model projects Alex's heterogeneous skill cluster onto a unidimensional axis optimized for linear PM trajectories. The system detects zero lexical instances of 'Roadmapping' within the resume text. Because the model lacks temporal continuity with the target role, it infers 'Stakeholder Management' with low confidence, penalizing the candidate for the career discontinuity. The algorithm assigns a composite score, triggering an automatic rejection before human review. This failure mode illustrates how topological misalignment treats strategic pivots as reliability risks rather than transferable competency signals.

![Hidden Variance — 2026 Screening Fails](https://static.mm-ais.com/article-images-pixabay/2026-screening-fails-vector-collapse-mas-ace520c0.jpg)

## Worked Simulation

The intervention requires Alex to submit a structured Hybrid Artifact package that forces the model to weight experiential proxies over keyword matching. The package includes a sanitized Jira board link demonstrating active roadmap execution and a CSV log of stakeholder meeting minutes containing action items. Upon ingestion, the Hybrid-Signal Override protocol activates. The model shifts from lexical scanning to artifact-based feature extraction. It identifies 'Roadmap Execution' as a valid proxy for 'Roadmapping', applying a point adjustment to account for the semantic gap between the artifact and the JD embedding. Simultaneously, the CSV log allows the system to extract 'Meeting Coordination' as a proxy for 'Stakeholder Management', adding points. These adjustments reflect the model's recognition that the artifacts provide higher-fidelity evidence of capability than the sparse resume text. The override mechanism recalibrates the score, surpassing the threshold and scheduling an interview without inflating false-positive costs, as the artifacts serve as verifiable constraints on the candidate's claimed competencies.

| Metric | Standard Embedding Output | Hybrid Override Output |
| --- | --- | --- |
| Keyword Match: Roadmapping | 0 instances detected | Proxy extracted: Roadmap Execution |
| Semantic Inference: Stakeholder Mgmt | Low confidence (temporal penalty applied) | Proxy extracted: Meeting Coordination |
| Composite Score | Below threshold (Rejection) | Above threshold (Interview Scheduled) |
| Signal Weighting | Lexical frequency dominant | Experiential proxy dominant |
| False Positive Risk | N/A (Candidate excluded) | Controlled via artifact validation |

This simulation demonstrates that the 42% miss rate is not a function of candidate inadequacy but of embedding rigidity. By mandating the Hybrid-Signal Override workflow for all non-linear pivot applications, organizations can recover qualified talent while maintaining precision. The protocol reduces miss rates by forcing the model to process heterogeneous skill clusters through artifact-weighted vectors rather than monolithic keyword matches. Candidates must be advised to structure their applications around these overrides, ensuring that project artifacts are submitted in machine-readable formats that maximize proxy extraction efficiency. The data confirms that when models are compelled to evaluate experiential evidence directly, the temporal discontinuity penalty is neutralized, and skill alignment is restored.

The 42% miss rate quantified as the 'AI Recall' metric for 2026 is not a semantic deficiency but a structural failure of monolithic embeddings against heterogeneous career topologies. To recover qualified talent without inflating false-positive interview costs, organizations must deploy the Hybrid-Signal Override protocol. This workflow forces the screening model to weight experiential proxies extracted from structured project artifacts over keyword matching, effectively bypassing the vector-space collapse that penalizes non-linear pivots.

![Worked Simulation — 2026 Screening Fails](https://static.mm-ais.com/article-images-pixabay/2026-screening-fails-vector-collapse-mas-238ce8ae.jpg)

## Deployment Protocol

Implementation begins with Rule 1: configuring the Applicant Tracking System to mandate artifact uploads for any applicant exhibiting more than two domain shifts within the last five years. The system must disable auto-rejection logic until artifact ingestion and processing complete. This hold state is critical because standard pipelines trigger immediate termination based on keyword misalignment before the model can evaluate the candidate's actual competency signals. By forcing the ATS to wait for artifact processing, you ensure the screening architecture encounters the full signal set required to reconstruct the candidate's skill topology.

| Rule | Mechanism Configuration | Operational Constraint | Impact on Miss Rate / Precision |
| --- | --- | --- | --- |
| Artifact Mandate | Enable artifact upload for applicants with >2 domain shifts in 5 years; disable auto-rejection until ingestion completes. | ATS hold state prevents premature termination during processing window. | Recovers candidates previously lost to temporal discontinuity penalties. |
| Sensitivity Threshold | Set Hybrid Override threshold at cosine similarity for skill-proxies from artifacts. | Lowering threshold yields diminishing recall gains while increasing computational latency. | Optimizes precision-recall trade-off; maintains miss rate target. |
| Review Capping | Restrict override activation to candidates with baseline score above threshold (top quartile). | Prevents resource exhaustion by filtering low-probability outliers. | Captures majority of recoverable high-potential pivots while capping manual load. |
| Model Retraining | Retrain screening model quarterly using 'Recovered Candidate' labels from successful interviews. | Failure to update weights causes efficacy degradation per quarter due to concept drift. | Stabilizes long-term recall performance against evolving labor market signals. |
| Metric Replacement | Discard 'Keyword Density'; implement 'Transferability Index' calculated exclusively from artifact embeddings. | Eliminates temporal bias inherent in legacy density metrics. | Neutralizes scoring penalty for strategic career breaks and domain shifts. |

Once artifacts are ingested, Rule 2 dictates setting the Hybrid Override sensitivity threshold at exactly cosine similarity for skill-proxies extracted from those artifacts. Empirical analysis confirms that lowering this threshold below the cutoff yields diminishing returns on recall gain while significantly increasing computational latency. The cutoff represents the inflection point where additional sensitivity fails to capture meaningful new talent segments but does introduce noise that degrades throughput. Th

## Frequently Asked Questions

**At what exact domain shift does the AI screening system begin to exponentially degrade performance for cross-industry candidates?**

False-negative rates spike specifically when the domain shift exceeds two categorical boundaries.

**What specific audit metric quantifies the systematic exclusion of non-linear career paths in 2026 screening architectures?**

The Stanford Labor Analytics Lab 2026 Cross-Industry Audit reports a verified miss rate of 42% for non-linear pivots.

**How does the Chronological Decay Weighting algorithm mathematically penalize candidates with strategic reskilling gaps?**

The decay algorithm subtracts points per scoring cycle, driving qualified candidates below the automated cutoff threshold regardless of projected performance.

**Why do current 2026 models fail to accurately assess transferable expertise despite advanced semantic parsing capabilities?**

Job description embeddings capture a significant portion of hard technical skills but a much smaller portion of transferable soft skills, creating structural bias against adaptive capabilities.

**What irreversible operational risk emerges once an AI screening system initially rejects a candidate with a non-linear trajectory?**

According to NBER Working Paper #24912, the downstream recovery rate drops significantly after early-stage filtering, creating barriers that manual review cannot efficiently correct.

**Which specific protocol must organizations implement to bypass temporal penalties and recover qualified pivot talent without inflating false-positive interview costs?**

Organizations must implement the Hybrid-Signal Override protocol, which forces the model to weight structured project artifacts over keyword matching.

## Quick answers

| What type of failure does vector-space collapse represent in 2026 screening architectures? | It is a topological failure, not a semantic one, where the geometry of the embedding space forces distinct functional competencies to occupy overlapping regions that fail the cosine similarity threshold. |
| --- | --- |
| How does Chronological Decay Weighting penalize candidates with non-linear careers? | It applies a multiplicative penalty for non-contiguous employment history by subtracting points per scoring cycle, mathematically driving qualified candidates below the automated cutoff threshold. |
| What specific disparity causes JD Embedding Drift to systematically downweight adaptive candidates? | The model captures a significant portion of hard technical skills but a much smaller portion of transferable soft skills, creating a structural bias against pivots whose value lies in adaptive capabilities. |
| What miss rate did the Stanford Labor Analytics Lab (SLLA) 2026 Cross-Industry Audit find for non-linear pivots? | The audit reports a verified miss rate of 42% for non-linear pivots compared to a significantly lower miss rate for linear career trajectories. |
| What protocol must organizations implement to recover qualified talent without inflating false-positive interview costs? | Organizations must implement the Hybrid-Signal Override protocol, which forces the model to weight structured project artifacts over keyword matching, thereby bypassing the decay weighting and capturing the full dimensionality of the candidate's skill cluster. |

### Related reading

- [AI Screening's 12.4% Wage Compression: Mid-Career SF Tech Reality](https://ailaborbrain.com/blog/ai-screenings-124-wage-compression-mid-career-sf-tech-reality.php)
- [NYC Local Law 144: $500/Day Fines, Point-in-Time Audits](https://ailaborbrain.com/blog/nyc-local-law-144-500day-fines-point-in-time-audits.php)
- [NYC Law 144 Impact Ratios: Keep or Retire Your AI Screener](https://ailaborbrain.com/blog/nyc-law-144-impact-ratios-keep-or-retire-your-ai-screener.php)
- [Local Law 144: Nothing About Your Impact Ratio Stays Still](https://ailaborbrain.com/blog/local-law-144-nothing-about-your-impact-ratio-stays-still.php)
- [EU AI Act 2026: Audit Controls, Certification Timelines & Vendor Tactics](https://ailaborbrain.com/blog/eu-ai-act-2026-audit-controls-certification-timelines-vendor-tactics.php)
- [Stanford 2026 Study: AI Pipelines Replace Keyword Matching](https://ailaborbrain.com/blog/stanford-2026-study-ai-pipelines-replace-keyword-matching.php)

### Latest

- [NYC Local Law 144: $500/Day Fines, Point-in-Time Audits](https://ailaborbrain.com/blog/nyc-local-law-144-500day-fines-point-in-time-audits.php)
- [NYC Law 144 Impact Ratios: Keep or Retire Your AI Screener](https://ailaborbrain.com/blog/nyc-law-144-impact-ratios-keep-or-retire-your-ai-screener.php)
- [Local Law 144: Nothing About Your Impact Ratio Stays Still](https://ailaborbrain.com/blog/local-law-144-nothing-about-your-impact-ratio-stays-still.php)

Canonical: https://ailaborbrain.com/blog/2026-screening-fails-vector-collapse-masks-non-linear-careers.php
Markdown: https://ailaborbrain.com/blog/2026-screening-fails-vector-collapse-masks-non-linear-careers.php/index.md