Active Frontier Source Registry

This registry tracks where to monitor the active autonomy research frontier and how to query each source. It is a manual-first operating page for perception, SLAM/mapping, world models, VLA/VLM, datasets/benchmarks, and validation.

It owns source discovery metadata only. Candidate status remains in the canonical owners: Perception Coverage Audit, SLAM Coverage Audit, Knowledge Gap Backlog, and Continuous Research Loop.

How To Use This Registry

1. Scan a source or saved query at its review cadence.
2. Verify the primary source before routing: paper, project page, dataset page, benchmark page, release note, standard, or official repository.
3. Dedupe against existing registry rows, audits, canonical pages, and gap backlogs.
4. Route manually:
   • route-to-backlog: durable gap belongs in Knowledge Gap Backlog.
   • route-to-audit: perception or SLAM evidence belongs in the relevant audit.
   • route-to-loop: process or cadence change belongs in Continuous Research Loop.
   • watch: relevant but not yet actionable; if it creates ongoing research work, place it in the canonical watchlist/P2 row.
   • ignore: out of scope, duplicate, superseded, or insufficiently primary.
5. Link the registry row to the canonical destination instead of restating the backlog item.

Scope

Included track	Watch for
perception	3D detection, BEV, occupancy, open-world/OOD, sensor fusion, radar, event, thermal, and robustness methods.
slam-mapping	LiDAR-inertial, visual-inertial, radar, Gaussian/neural SLAM, loop closure, pose graphs, lifelong mapping, and map-change methods.
world-models	Occupancy world models, scene generation, neural simulation, 4D forecasting, diffusion planning, and closed-loop simulation.
vla-vlm	Driving VLM/VLA reliability, spatial reasoning, grounded language, action heads, and closed-loop evaluation.
datasets-benchmarks	Driving, adverse-weather, corruption, FOD, V2X, 4D radar, occupancy, and SLAM datasets/benchmarks.
validation	Closed-loop evaluation, scenario testing, uncertainty calibration, runtime monitoring, safety evidence, and fault injection.

Out of scope: broad company tracking, market intelligence, general operations, and platform maintenance unless they directly affect one of the six active frontier tracks.

Registry Content Model

Each source row uses stable IDs and parseable fields so a future Node checker can validate the page without changing the document model.

Field	Purpose
Source ID	Stable machine-readable identifier.
Source	Human-readable source name.
Source Type	Controlled source class such as preprint, venue, publisher-index, dataset, benchmark, code, model, standard, regulator, industry-lab, or weak-signal.
Authority Tier	T1, T2, T3, or T4.
Source Role	Discovery role: primary evidence, aggregator, code artifact, benchmark, dataset, weak signal, or standards source.
Frontier Tracks	Comma-separated track IDs.
Evidence Objects	Expected primary evidence: paper, DOI, arXiv/OpenReview ID, repo, project page, dataset page, benchmark page, standard, release note.
Native Filters	Native categories, venue filters, search operators, alert options, or API parameters.
Manual Query Pattern	Human-readable query seed.
Watch Method	RSS, API, email alert, saved search, page watch, manual page review, or issue digest.
Cadence	Weekly, monthly, quarterly, event-driven, or conference-cycle.
Last Checked	ISO date.
Next Review	ISO date.
Source Health	Controlled health value.
Automation	high, medium, low, or none.
Verification Rule	Required primary-source check before routing.
Caveats	Short source-specific warnings.

Authority tiers:

• T1: peer-reviewed venue, official benchmark/dataset, official standard, or regulator source.
• T2: author/lab page, official repository/model release, industry technical report, or preprint with artifacts.
• T3: scholarly aggregator or metadata index.
• T4: social, newsletter, community, vendor, or other weak signal.

Automation values:

• high: stable API, RSS, export, or predictable identifiers.
• medium: stable pages, but scraping/page-diff/manual query shaping likely.
• low: useful manually, brittle for scripts.
• none: manual-only.

Source health values:

• active: source works and returns relevant frontier material.
• degraded: filters, feeds, or result quality have weakened but remain useful.
• paused: temporarily noisy, unavailable, or off-cycle.
• retired: no longer useful; replacement or rationale belongs in caveats.
• moved: source appears relocated; keep old URL until manually verified.
• stale: review date has expired; content is not automatically wrong.
• unreachable: last check failed; requires retry or manual review.
• withdrawn-or-retracted: only set after primary-source confirmation.
• license-or-access-limited: source exists but use or redistribution is constrained.
• unknown: automation could not determine status.

Source Categories

Category	Example sources	Role
Preprints and metadata	arXiv, OpenReview, Semantic Scholar, OpenAlex, Crossref, DBLP	Early discovery and metadata cross-checking.
Core venues	CVPR, ICCV, ECCV, WACV, NeurIPS, ICML, ICLR, CoRL, RSS, ICRA, IROS, IEEE IV, ITSC	Accepted paper and workshop tracking.
Publisher indexes	IEEE Xplore, ACM Digital Library, SpringerLink, ScienceDirect, SAGE/IJRR, SAE	Final proceedings, journals, robotics/control/sensor papers.
Datasets and benchmarks	Waymo Open Dataset, nuScenes, Argoverse, KITTI/SemanticKITTI, nuPlan, CARLA Leaderboard, NAVSIM, Bench2Drive, OpenLane/OpenLane-V2, V2X, adverse-weather, corruption, and SLAM benchmarks	Dataset and evaluation freshness.
Code and model artifacts	GitHub, Hugging Face, official project pages, OpenMMLab, OpenPCDet, Autoware/ROS ecosystems	Reproducibility and release signals.
Simulation and validation infrastructure	CARLA, CommonRoad, Scenic, MetaDrive, Waymax, ASAM OpenSCENARIO, ASAM OpenODD, ASAM OpenDRIVE	Closed-loop and scenario-evidence monitoring.
Safety and regulatory frontier	ISO 21448/SOTIF, ISO 26262, UL 4600, SAE, UNECE, NHTSA ADS materials	Validation, assurance, and standards signals.
Industry research	Waymo, NVIDIA, Waabi, Toyota/Woven, Mobileye, Bosch, Aurora, Zoox, Tesla AI	Production-adjacent technical signals.
Sensor frontier	LiDAR, 4D radar, FMCW LiDAR, event-camera, and thermal vendors; SPIE/SAE/IEEE sensor papers; patents and application notes	Hardware-driven perception and SLAM frontier signals.
Weak signals	Lab pages, newsletters, challenge pages, curated GitHub lists, social/community posts	Discovery only; never sufficient for promotion.

Source Registry

Source ID	Source	Source Type	Authority Tier	Source Role	Frontier Tracks	Evidence Objects	Native Filters	Manual Query Pattern	Watch Method	Cadence	Last Checked	Next Review	Source Health	Automation	Verification Rule	Caveats
arxiv-cs-cv	arXiv cs.CV	preprint	T2	preprint discovery	perception, world-models, vla-vlm, datasets-benchmarks	paper, arXiv ID, project page, repo	cat:cs.CV, recent submissions	autonomous driving plus occupancy, BEV, 3D detection, OOD, VLM, world model	RSS/API	weekly	2026-05-12	2026-05-19	active	high	Verify paper page and artifact links before routing.	Preprint quality varies; dedupe against conference versions.
arxiv-cs-ro	arXiv cs.RO	preprint	T2	preprint discovery	slam-mapping, validation, datasets-benchmarks	paper, arXiv ID, repo, dataset page	cat:cs.RO, recent submissions	SLAM, odometry, mapping, pose graph, closed-loop evaluation, benchmark	RSS/API	weekly	2026-05-12	2026-05-19	active	high	Verify robotics relevance and primary artifacts before routing.	Many generic robotics papers need AV relevance filters.
arxiv-eess-iv	arXiv eess.IV	preprint	T2	preprint discovery	perception, datasets-benchmarks	paper, arXiv ID, repo	cat:eess.IV, recent submissions	image/video perception, event camera, thermal, sensor fusion, driving scenes	RSS/API	weekly	2026-05-12	2026-05-19	active	high	Verify driving or robotic perception fit before routing.	Medical and generic image papers require negative filtering.
arxiv-eess-sy	arXiv eess.SY	preprint	T2	preprint discovery	validation, slam-mapping	paper, arXiv ID, repo	cat:eess.SY, recent submissions	control, safety, runtime verification, uncertainty, fault injection, localization	RSS/API	weekly	2026-05-12	2026-05-19	active	high	Verify autonomy or safety-case relevance before routing.	Broad systems/control category can be noisy.
openreview-venues	OpenReview venues	venue	T1	open-review venue discovery	world-models, vla-vlm, perception, validation	paper, OpenReview ID, decision, reviews	venue invitation, title, abstract, decision	ICLR, ICML, NeurIPS, CoRL with driving, embodied, world model, VLA, robustness	API/manual	weekly	2026-05-12	2026-05-19	active	medium	Verify venue decision and official paper page before routing.	Venue fields differ and API versions vary.
cvf-openaccess	CVF Open Access	venue	T1	proceedings discovery	perception, world-models, vla-vlm, datasets-benchmarks	paper, proceedings page, project page, repo	conference year, title search	CVPR, ICCV, ECCV, WACV with occupancy, BEV, 3D, driving, VLM, dataset	manual/page watch	conference-cycle	2026-05-12	2026-08-12	active	medium	Verify official CVF page and artifact links before routing.	Static pages are stable but not a clean API.
pmlr-proceedings	PMLR Proceedings	venue	T1	proceedings discovery	world-models, vla-vlm, validation	paper, proceedings page, PDF	proceedings volume, title search	CoRL, ICML, AISTATS with robotics, driving, world model, policy, evaluation	manual/page watch	monthly	2026-05-12	2026-06-12	active	medium	Verify proceedings record and project artifacts before routing.	Some relevant venues use OpenReview instead.
rss-proceedings	Robotics: Science and Systems	venue	T1	proceedings discovery	slam-mapping, validation, datasets-benchmarks	paper, proceedings page, project page, repo	conference year, title search	SLAM, mapping, localization, multi-robot, benchmark, field robotics	manual/page watch	conference-cycle	2026-05-12	2026-08-12	active	medium	Verify official RSS paper page before routing.	Proceedings page structure changes by year.
icra	IEEE RAS ICRA	venue	T1	proceedings discovery	slam-mapping, perception, validation, datasets-benchmarks	paper, DOI, IEEE record, project page	program, title, keywords, IEEE Xplore	SLAM, odometry, mapping, field robotics, planning, evaluation, datasets	manual/page watch	conference-cycle	2026-05-12	2026-08-12	active	low	Verify official program or IEEE record before routing.	Final metadata often lands in IEEE Xplore.
iros	IEEE RAS IROS	venue	T1	proceedings discovery	slam-mapping, perception, validation	paper, DOI, IEEE record, project page	program, title, keywords, IEEE Xplore	robust perception, SLAM, localization, mapping, multi-robot, safety	manual/page watch	conference-cycle	2026-05-12	2026-08-12	active	low	Verify official program or IEEE record before routing.	Program pages and accepted-paper lists vary by year.
ieee-xplore	IEEE Xplore	publisher-index	T1	publisher record	perception, slam-mapping, validation, datasets-benchmarks	DOI, publisher page, abstract, venue	publication title, year, author, IEEE terms	IV, ITSC, ICRA, IROS, T-RO, RA-L with driving, SLAM, sensor fusion, safety	saved search/API	monthly	2026-05-12	2026-06-12	active	medium	Verify publisher record and avoid relying on abstract-only claims.	API access requires credentials; many records are paywalled.
acm-dl	ACM Digital Library	publisher-index	T1	publisher record	vla-vlm, world-models, validation	DOI, publisher page, conference page	title, abstract, venue, date	embodied AI, HRI, simulation, evaluation, foundation models, safety	saved search/manual	monthly	2026-05-12	2026-06-12	active	low	Verify DOI and official paper page before routing.	Less central for AV but useful for HRI and embodied-AI overlap.
semantic-scholar	Semantic Scholar	publisher-index	T3	metadata aggregator	perception, slam-mapping, world-models, vla-vlm, validation	paper metadata, DOI, arXiv ID, citations	query, field of study, date	exact method names, citation trails, related work, code links	API/manual	weekly	2026-05-12	2026-05-19	active	medium	Use only to discover or cross-check primary sources.	Aggregator metadata can lag or merge records incorrectly.
openalex	OpenAlex	publisher-index	T3	metadata aggregator	perception, slam-mapping, world-models, validation	DOI, venue, authors, concepts	works search, concept, date	topic discovery, publication date checks, DOI lookup	API/manual	weekly	2026-05-12	2026-05-19	active	high	Use only to discover or cross-check primary sources.	Concept tags are broad and require manual validation.
dblp	DBLP	publisher-index	T3	bibliographic index	perception, slam-mapping, world-models, vla-vlm	venue record, author record, DOI link	author, title, venue, year	venue completeness, author/lab follow-up, accepted paper cross-check	manual/API	monthly	2026-05-12	2026-06-12	active	medium	Use for bibliographic cross-checking, not promotion alone.	Does not provide technical evidence.
github-search	GitHub Search	code	T2	code artifact discovery	perception, slam-mapping, world-models, vla-vlm, datasets-benchmarks	repo, release, license, README, issues	topic, language, stars, pushed date	autonomous-driving occupancy SLAM VLM dataset benchmark stars pushed	API/manual	weekly	2026-05-12	2026-05-19	active	high	Verify official repo link from paper/project page before routing.	Stars are weak evidence; do not clone arbitrary repos in CI.
huggingface-papers	Hugging Face Papers	model	T3	model and paper discovery	world-models, vla-vlm, perception, datasets-benchmarks	paper, model card, dataset card, license	papers, models, datasets, date	driving VLM, robot foundation model, world model, occupancy, dataset	manual/API	weekly	2026-05-12	2026-05-19	active	medium	Verify canonical paper and license before routing.	Trending pages are discovery signals only.
waymo-open	Waymo Open Dataset	dataset	T1	dataset and challenge source	datasets-benchmarks, perception, world-models, validation	dataset page, challenge page, paper, license	challenge, task, release note	3D detection, motion, occupancy, E2E, simulation, challenge updates	page watch/manual	monthly	2026-05-12	2026-06-12	active	medium	Verify official release notes and task definitions before routing.	Terms and benchmark splits matter for claims.
nuscenes	nuScenes	dataset	T1	dataset and benchmark source	perception, datasets-benchmarks, validation	dataset page, benchmark, devkit, license	task, benchmark, release	detection, tracking, prediction, occupancy, map, sensor fusion	page watch/manual	monthly	2026-05-12	2026-06-12	active	medium	Verify benchmark task and devkit version before routing.	Older but still central; watch for derivative benchmarks.
argoverse	Argoverse	dataset	T1	dataset and benchmark source	datasets-benchmarks, perception, world-models, validation	dataset page, benchmark, paper, license	task, benchmark, release	motion forecasting, sensor dataset, maps, E2E, scenario mining	page watch/manual	monthly	2026-05-12	2026-06-12	active	medium	Verify official task page before routing.	Derivative papers may use outdated versions.
kitti	KITTI	dataset	T1	legacy benchmark source	perception, slam-mapping, datasets-benchmarks	dataset page, benchmark page, paper	task, benchmark, raw data	odometry, object detection, tracking, semantic scene completion	manual	quarterly	2026-05-12	2026-08-12	active	low	Verify task relevance and avoid over-weighting legacy leaderboard claims.	Useful baseline but not representative of modern AV stacks alone.
carla-leaderboard	CARLA Leaderboard	benchmark	T1	closed-loop benchmark source	validation, world-models, vla-vlm	benchmark page, paper, code, scenario definitions	leaderboard, challenge, route, scenario	closed-loop driving, leaderboard, agent robustness, scenario evaluation	page watch/manual	monthly	2026-05-12	2026-06-12	active	medium	Verify scenario version and leaderboard rules before routing.	Synthetic benchmark claims need real-world caveats.
navsim	NAVSIM	benchmark	T1	closed-loop evaluation source	validation, world-models, vla-vlm	benchmark page, paper, code	task, metric, release	open-loop and closed-loop planning, E2E evaluation, driving score	page watch/manual	monthly	2026-05-12	2026-06-12	active	medium	Verify official task and metric definitions before routing.	Benchmark versions can change quickly.
bench2drive	Bench2Drive	benchmark	T1	closed-loop benchmark source	validation, vla-vlm, world-models	benchmark page, paper, code	task, metric, scenario	closed-loop driving, VLM/VLA evaluation, long-tail scenarios	page watch/manual	monthly	2026-05-12	2026-06-12	active	medium	Verify official benchmark page and code before routing.	CARLA-based results need sim-to-real caveats.
asam-openx	ASAM Standards	standard	T1	scenario and ODD standard source	validation, datasets-benchmarks	standard page, release notes, schema	OpenSCENARIO, OpenODD, OpenDRIVE	scenario testing, ODD, road network, simulation evidence	page watch/manual	quarterly	2026-05-12	2026-08-12	active	low	Verify official standard version before routing.	Access and versioning differ across standards.
nhtsa-ads	NHTSA Automated Vehicles Safety	regulator	T1	regulator source	validation	official guidance, rulemaking, reporting page	ADS, crash reporting, guidance, exemptions	ADS safety, reporting, post-market monitoring, incident evidence	page watch/manual	monthly	2026-05-12	2026-06-12	active	low	Verify official docket, rule, or guidance page before routing.	Date-sensitive; do not infer final rules from proposals.
nvidia-research	NVIDIA Research	industry-lab	T2	industry research source	perception, world-models, vla-vlm, validation	paper, project page, code, dataset, model	publication page, project page, date	Cosmos, simulation, 3D perception, world models, autonomous driving	page watch/manual	monthly	2026-05-12	2026-06-12	active	medium	Verify paper/project artifacts and license before routing.	Product pages and research pages should not be mixed.
waymo-research	Waymo Research	industry-lab	T2	industry research source	perception, world-models, datasets-benchmarks, validation	paper, dataset, benchmark, technical report	publication page, date, topic	safety, simulation, datasets, planning, perception, E2E driving	page watch/manual	monthly	2026-05-12	2026-06-12	active	medium	Verify official paper or dataset page before routing.	Company claims should stay attributed.
researchgate	ResearchGate	weak-signal	T4	author follow-up source	perception, slam-mapping, world-models, vla-vlm, datasets-benchmarks, validation	author page, project update, paper pointer	author, topic, project	exact author or method follow-up after primary source discovery	manual	monthly	2026-05-12	2026-06-12	active	none	Use only to find a primary source elsewhere.	Not reliable as canonical evidence and not suitable for automation.

Active Frontier Query Bank

Query ID	Track	Source IDs	Query Pattern	Native Filters	Intent	Cadence
perception-occupancy	perception	arxiv-cs-cv, cvf-openaccess, openreview-venues, semantic-scholar	"occupancy" AND "autonomous driving" plus BEV, panoptic, semantic occupancy, open occupancy	cat:cs.CV, CVPR/ICCV/WACV, recent date	Find occupancy perception methods and benchmarks.	weekly
perception-open-world	perception	arxiv-cs-cv, cvf-openaccess, huggingface-papers, github-search	"open-vocabulary 3D" OR "open-world" OR "OOD" OR "anomaly segmentation" plus driving	cat:cs.CV, project/repo links	Find open-world, OOD, and rare-object perception methods.	weekly
perception-radar-sensor	perception	arxiv-eess-iv, ieee-xplore, icra, iros	"4D radar" OR "FMCW LiDAR" OR "event camera" OR "thermal" plus driving or robotics	cat:eess.IV, IEEE IV/ITSC/ICRA/IROS	Find sensor-frontier perception and fusion work.	weekly
slam-lio-radar	slam-mapping	arxiv-cs-ro, icra, iros, rss-proceedings, ieee-xplore	"LiDAR-inertial odometry" OR "radar odometry" OR "visual-inertial SLAM" plus benchmark	cat:cs.RO, ICRA/IROS/RSS, T-RO/RA-L	Find SLAM, odometry, and localization methods.	weekly
slam-map-change	slam-mapping	arxiv-cs-ro, rss-proceedings, semantic-scholar, github-search	"lifelong mapping" OR "map change detection" OR "dynamic object removal" OR "pose graph optimization"	cat:cs.RO, recent date, repo links	Find map hygiene, lifelong mapping, and backend gaps.	weekly
world-models-driving	world-models	arxiv-cs-cv, openreview-venues, cvf-openaccess, pmlr-proceedings, nvidia-research	"driving world model" OR "occupancy world model" OR "4D occupancy forecasting" OR "neural simulation"	cat:cs.CV, OpenReview venues, CVF	Find frontier driving world-model and simulation papers.	weekly
vla-vlm-driving	vla-vlm	arxiv-cs-cv, openreview-venues, huggingface-papers, github-search	"driving VLM" OR "vision-language-action" OR "spatial reasoning" OR "action head" plus driving or robot	cat:cs.CV, OpenReview, model cards	Find driving VLM/VLA methods and reliability benchmarks.	weekly
datasets-adverse-fod	datasets-benchmarks	waymo-open, nuscenes, argoverse, kitti, arxiv-cs-cv, semantic-scholar	"adverse weather dataset" OR "sensor corruption" OR "FOD" OR "V2X dataset" OR "4D radar dataset"	dataset release pages, benchmark pages	Find datasets and benchmarks that should feed audits.	monthly
validation-closed-loop	validation	carla-leaderboard, navsim, bench2drive, openreview-venues, pmlr-proceedings	"closed-loop evaluation" OR "scenario testing" OR "distribution shift" OR "fault injection"	benchmark task, metrics, recent date	Find validation protocols and closed-loop benchmarks.	monthly
validation-standards	validation	asam-openx, nhtsa-ads, ieee-xplore	"OpenSCENARIO" OR "OpenODD" OR "ADS safety" OR "runtime monitor" OR "safety case"	standard version, official guidance, publication date	Find scenario-evidence, safety, and regulatory updates.	monthly

Cross-cutting modifiers:

• autonomous driving
• self-driving
• ego vehicle
• driving scenes
• closed-loop
• ODD
• sim-to-real
• long-tail
• safety-critical

Negative filters:

• Add NOT medical when image or segmentation queries return medical imaging.
• Add NOT satellite when perception queries return remote-sensing papers.
• Add NOT generic LLM when VLM queries lose driving or robotics grounding.
• Add NOT manipulation when robotics queries are unrelated to vehicle autonomy.

Candidate Routing

This registry owns source discovery only. Candidate status is maintained in the canonical owner, not in this page.

Candidate type	Canonical owner after triage
Perception methods, perception robustness, perception datasets	Perception Coverage Audit
SLAM, odometry, localization, mapping methods, SLAM benchmarks	SLAM Coverage Audit
World models, VLA/VLM, end-to-end driving, simulation frontier	Knowledge Gap Backlog until a dedicated audit exists
Cross-track datasets and benchmarks	Owning domain audit if method-specific; otherwise Knowledge Gap Backlog
Validation, safety evidence, evaluation protocols	Knowledge Gap Backlog until a dedicated validation audit exists

Promotion rule:

A candidate can be promoted only after primary-source verification: canonical paper/DOI/arXiv/OpenReview ID, official repo or project page where available, dataset/benchmark page if relevant, artifact/license/access notes, validation or benchmark claim, and at least one caveat.

Candidate flow:

source scan -> primary-source verification -> dedupe -> canonical backlog/audit -> atomic page or watchlist

Workflow And Automation

Manual-First Boundary

The registry is a manual-first intake surface. Automation may validate structure, report diagnostics, and produce candidate digests. It must not edit canonical backlogs, route candidates, delete sources, promote work, or replace primary-source evidence with aggregator metadata.

Review Cadence And Gates

• Weekly: review new candidates and automation notes, dedupe them, and assign a routing decision.
• Monthly: prune stale watch traces, confirm canonical links still resolve, and close entries that have been routed.
• Quarterly: review whether source categories, scope boundaries, or automation checks need adjustment in Continuous Research Loop.

Human review is required before any source changes canonical backlog, audit, or loop content. A source may be routed only after primary-source verification and dedupe are complete.

Automation Phases

• Phase 0: Markdown-only registry.
• Phase 1: offline Node checker validates IDs, enums, required fields, ISO dates, duplicate IDs/URLs, source-health values, stale review dates, and canonical links.
• Phase 2: opt-in Node scanner creates review artifacts only: dated Markdown digest, raw-result manifest/cache, and skipped/error summary.
• Phase 3: scheduled GitHub Action may open a review issue or upload artifacts. It must not commit edits or modify canonical files.

CI And Fetch Policy

Normal PR checks stay offline and deterministic: schema checks, internal links, tests, and VitePress build.

Live source fetching runs only through manual commands or scheduled jobs. Scheduled fetches must use provider allowlists, short timeouts, bounded retries, low concurrency, per-provider budgets, stable user-agent strings, cached responses where permitted, and secrets only from GitHub Actions or local environment variables.

Missing API keys skip that provider rather than failing the build.

Failure, Security, And Terms

External failures produce diagnostics, not content changes. API outages, 403/429 responses, DNS/TLS failures, malformed data, and schema drift should be recorded as unknown or unreachable until manually reviewed.

No API keys, cookies, signed URLs, or tokens belong in Markdown, VitePress output, logs, or PR jobs. Scheduled jobs should use least permissions, such as contents: read and issues: write.

Use official APIs where possible. Do not store copied abstracts, paper bodies, README bodies, fetched PDFs, dataset contents, or large scraped text. Store metadata, URLs, license/access notes, and human-written review notes.

Testing Split

Run on PRs:

• Registry schema validation.
• Internal link checks.
• VitePress build.
• Fixture-based parser tests.
• Deterministic diagnostic snapshots.
• Secret-pattern checks.

Run only manually or on schedule:

• Live provider smoke tests.
• Freshness scans.
• Source-health reports.
• Candidate discovery digests.

Later Automation Feasibility

Initial high-feasibility automation sources:

• arXiv API/RSS
• GitHub Search API
• OpenAlex API

Medium-feasibility sources:

• Semantic Scholar API
• OpenReview API
• CVF static proceedings pages
• stable dataset/benchmark pages

Manual-only or weak automation sources:

• vendor blogs
• benchmark leaderboards with changing HTML
• standards pages behind access controls
• ResearchGate
• social/news/community sources

Provider limits must be configuration, not hidden constants. arXiv expects polite delays and bounded slices; GitHub search uses separate rate buckets and secondary limits; Semantic Scholar and OpenAlex have key-dependent throughput; OpenReview has venue-specific fields and API-version differences.

Sources

• arXiv API user manual: https://info.arxiv.org/help/api/user-manual.html
• arXiv RSS help: https://info.arxiv.org/help/rss.html
• arXiv category taxonomy: https://arxiv.org/category_taxonomy
• OpenReview API: https://docs.openreview.net/getting-started/using-the-api
• IEEE developer API documentation: https://developer.ieee.org/docs/read/Home
• GitHub Search API: https://docs.github.com/en/rest/search/search
• OpenAlex API: https://docs.openalex.org/
• Semantic Scholar API: https://api.semanticscholar.org/api-docs/
• CVF Open Access: https://openaccess.thecvf.com/menu
• ASAM OpenSCENARIO: https://publications.pages.asam.net/standards/ASAM_OpenSCENARIO/ASAM_OpenSCENARIO_XML/latest/00_preface/01_introduction.html
• ASAM OpenODD: https://www.asam.net/standards/detail/openodd/
• NHTSA automated vehicles safety: https://www.nhtsa.gov/vehicle-safety/automated-vehicles-safety
• ISO 21448: https://www.iso.org/standard/77490.html