#!/usr/bin/env python3
"""Agent E (AI Research Specialist), WP44 Phase 5 — contributions.csv for the ai_ml landmark set.
Reads literature_synthesis/landmarks/ai_ml_landmarks.csv + ai_ml_abstracts.jsonl; the judgements below were made by the
specialist agent from the abstract (judged_from=abstract) or from title/venue/type only (metadata_only). data_quality_flag:
  ok | identifier_mismatch (title/abstract do not match the arXiv id or the citation count attached to the record)
  | abstract_mismatch (abstract belongs to another work) | year_implausible (OpenAlex year contradicts the known first
  publication) | normalisation_artefact (tiny c36 but very large NCS36: young cohort / rare document type)
  | off_scope (classified into an AI category but not an AI contribution) | no_abstract (metadata only).
Usage: python3 literature_synthesis/agent_E/make_contributions.py <repo_root>"""
import csv, json, sys, os, unicodedata
R = sys.argv[1] if len(sys.argv) > 1 else "."
L = {r["work_id"]: r for r in csv.DictReader(open(f"{R}/literature_synthesis/landmarks/ai_ml_landmarks.csv"))}
A = {str(json.loads(l)["work_id"]): json.loads(l).get("abstract") or "" for l in open(f"{R}/literature_synthesis/landmarks/ai_ml_abstracts.jsonl")}
def asc(s): return unicodedata.normalize("NFKD", s).encode("ascii", "ignore").decode()
# work_id: (contribution_type, description, data_quality_flag)
J = {
"7172491899": ("engineering","Text-to-MongoDB query translation (GAT + RAG + T5); record carries arXiv 2005.11401 (RAG, Lewis et al.) and its citations","identifier_mismatch"),
"7154011589": ("empirical","Portuguese-language essay on LLM pre-training data and copyright; record carries arXiv 2005.14165 (GPT-3) and its citations","identifier_mismatch"),
"4362515116": ("survey","Survey of LLM techniques (pre-training, adaptation, utilisation, evaluation); journal version of the 2023 arXiv survey","ok"),
"4288253152": ("engineering","Dissertation: fine-tuned multilingual sentence-transformer for Dewey Decimal classification; NCS36 implausible for a dissertation","normalisation_artefact"),
"4322718191": ("engineering","LLaMA: 7B-65B open-weight foundation language models trained on public data; the abstract reports LLaMA-13B outperforms GPT-3 on most benchmarks","ok"),
"4323655724": ("survey","Position paper on opportunities and challenges of LLMs for education (student and teacher perspectives)","ok"),
"4367369313": ("survey","Practitioner guide to prompt engineering for ChatGPT","ok"),
"4226278401": ("method","Proposal of a 'companion curriculum' for LM training; record carries arXiv 2203.02155 (InstructGPT) and its citations","identifier_mismatch"),
"4396531244": ("engineering","Dissertation: multilingual semantic search + LLMs for news retrieval; NCS36 implausible for c36=20","normalisation_artefact"),
"4221143046": ("theory","Essay on 'theatrical compliance' failure mode; record carries arXiv 2201.11903 (chain-of-thought) and its citations","identifier_mismatch"),
"4327810158": ("engineering","Title = GPT-4 Technical Report (arXiv 2303.08774) but the stored abstract describes an unrelated 'MFOUR Vibe Framework'","abstract_mismatch"),
"7135096569": ("empirical","Preregistered field experiment (758 BCG consultants) on the uneven ('jagged') effect of GPT-4 assistance on knowledge-work tasks","ok"),
"4414281281": ("method","DeepSeek-R1: reasoning incentivised by pure reinforcement learning without human-annotated reasoning traces (Nature version)","ok"),
"4389437528": ("survey","Review of potential benefits of ChatGPT for teaching and learning","ok"),
"4292779060": ("theory","Speculative 'Aion Framework' on AI consciousness; c36=13,333 cannot belong to this record","identifier_mismatch"),
"4360836968": ("empirical","'Sparks of AGI': qualitative experiments on an early GPT-4; contested interpretation; preprint","ok"),
"4406800520": ("survey","Survey of LLM-based agents (brain/perception/action framework); journal version of arXiv 2309.07864","no_abstract"),
"4385223296": ("survey","Book chapter on responsible use of ChatGPT in education","no_abstract"),
"4287689466": ("empirical","LLM essay classification for writing placement (450 essays, four models); c36=722 implausible for an OASIcs paper","identifier_mismatch"),
"4366420437": ("survey","Rapid review of 50 early articles on ChatGPT in education (Dec 2022-Feb 2023)","ok"),
"4317910584": ("survey","Commentary on ChatGPT and assessment in higher education","ok"),
"4313564799": ("empirical","SSRN preprint on ChatGPT user experience in education","no_abstract"),
"4382618722": ("engineering","Chatlaw: multi-agent legal assistant with role-aligned mixture-of-experts (Chinese law)","ok"),
"4384464487": ("empirical","Survey of 399 Hong Kong students on generative AI perceptions","ok"),
"4392773210": ("survey","Discussion of AI literacy, prompt engineering and critical thinking in classrooms","ok"),
"4387355843": ("method","Time-LLM: reprogramming LLMs for time-series forecasting (book-chapter record of the ICLR 2024 paper)","ok"),
"4392947521": ("method","UNI: general-purpose self-supervised foundation model for computational pathology","no_abstract"),
"7124890304": ("method","Semi-supervised cross-domain fault diagnosis via contrastive pre-training","no_abstract"),
"4410551352": ("method","Aurora: foundation model for the Earth system trained on >1M hours of geophysical data; abstract reports outperforming operational forecasts","ok"),
"4322832366": ("method","TimeMAE: decoupled masked autoencoders for time-series representation","ok"),
"7133233472": ("method","QLoRA: 4-bit quantised base + low-rank adapters; 65B model finetuned on one 48 GB GPU","ok"),
"2990704537": ("benchmark","SuperGLUE: harder successor to GLUE after models exceeded non-expert human level","ok"),
"7125271461": ("engineering","ChemBERTa-3: open training framework for chemical foundation models","ok"),
"7119541039": ("theory","Position paper defining eleven desirable features of Earth foundation models","ok"),
"4393138529": ("survey","Survey of parameter-efficient fine-tuning (PEFT) for large models; preprint","ok"),
"4390690043": ("survey","Survey of data-centric foundation-model approaches in computational healthcare","ok"),
"7128029160": ("method","BrainIAC: self-supervised foundation model for brain MRI","ok"),
"4402220065": ("method","Pathology foundation model for cancer diagnosis and prognosis (Nature 2024)","no_abstract"),
"7118170538": ("survey","Review of AI applications in fisheries","ok"),
"4411932262": ("method","Centaur: language model fine-tuned on Psych-101 (10M choices) predicting human behaviour across experiments","ok"),
"4411099958": ("method","PanDerm: multimodal dermatology foundation model pretrained on 2M images, 28 benchmarks","ok"),
"4361229539": ("method","LLaMA-Adapter: 1.2M-parameter adaption prompts with zero-init attention for instruction following; preprint","ok"),
"4407135816": ("survey","Narrative review of foundation models for radiology","ok"),
"4385948838": ("method","PLIP: visual-language pathology foundation model trained on medical Twitter data","no_abstract"),
"4406643700": ("survey","Quantitative literature analysis and agenda for geospatial AI","ok"),
"4406472599": ("survey","Perspective on 'learning from models' beyond fine-tuning","no_abstract"),
"4393153153": ("method","TimesURL: self-supervised contrastive universal time-series representations","ok"),
"4402727669": ("method","mPLUG-Owl2: modality-collaborative multimodal LLM with shared modules","ok"),
"4401397240": ("method","BiomedGPT: generalist vision-language foundation model for biomedical tasks","no_abstract"),
"4385285835": ("method","Masked image modelling at scale for histopathology; medRxiv preprint","ok"),
"4318718936": ("method","BLIP-2: Q-Former bridging frozen image encoders and frozen LLMs; preprint","ok"),
"4393148714": ("method","T2I-Adapter: lightweight adapters adding structural/colour control to text-to-image diffusion","ok"),
"4396599905": ("survey","Survey of multimodal fusion under low-quality data (noise, missing modalities, imbalance, dynamics)","ok"),
"4396952261": ("survey","Textbook on multimodal deep learning","no_abstract"),
"7133250524": ("benchmark","LAION-5B: open 5.85B image-text pair dataset for training CLIP-style models","no_abstract"),
"4392947532": ("method","CONCH: visual-language foundation model for computational pathology","no_abstract"),
"7138474371": ("method","OpenDriveVLA: vision-language-action model for end-to-end driving on open LLMs","ok"),
"7127319731": ("method","Med-R1: RL-enhanced vision-language model for medical reasoning","ok"),
"2619383789": ("survey","Taxonomy of multimodal machine learning (representation, translation, alignment, fusion, co-learning)","ok"),
"7148475777": ("empirical","Zero-shot evaluation of GPT-5 on multimodal medical QA/VQA","ok"),
"7160627408": ("method","Generative cross-domain knowledge synthesis framework (conference paper)","ok"),
"2995460200": ("method","VL-BERT: single-stream Transformer pre-trained on image regions and words","ok"),
"4415913530": ("method","TITAN: multimodal whole-slide pathology foundation model (335,645 WSIs)","ok"),
"4289785095": ("method","Textual inversion: personalising text-to-image diffusion with learned pseudo-words from 3-5 images; preprint","ok"),
"7125233009": ("method","SemCap: sentiment-aware captions as visual proxy for multimodal aspect-based sentiment analysis","ok"),
"7143512276": ("survey","Survey of multimodal ML in predictive healthcare analytics","ok"),
"4413447271": ("method","RadFM: generalist radiology foundation model on 2D/3D data (MedMD, 13M images)","ok"),
"7131830459": ("survey","Survey of multimodal recommender systems","ok"),
"7206206818": ("method","Lingshu: generalist medical multimodal LLM","ok"),
"7133220561": ("method","InstructBLIP: instruction tuning of vision-language models","no_abstract"),
"7164501397": ("empirical","Evaluation of 8 multimodal foundation models on 1,090 medical cases: image predictions largely driven by text","ok"),
"7165561001": ("method","Hierarchical graph-Mamba multi-agent clinical architecture","no_abstract"),
"7138470740": ("method","CoEvo: training-free test-time OOD detection with vision-language models","ok"),
"7138363754": ("method","AdaptCLIP: adapting CLIP for universal visual anomaly detection","ok"),
"4410882873": ("survey","Scoping review of 432 multimodal medical AI papers (2018-2024)","ok"),
"4400228804": ("empirical","Dissertation on generative AI in animation production; c36=2 with NCS36=540 is a normalisation artefact","normalisation_artefact"),
"7124262017": ("survey","OECD Digital Education Outlook 2026 (policy report on generative AI in education)","ok"),
"4406472463": ("method","MatterGen: diffusion model generating stable inorganic materials under property constraints","ok"),
"7123342962": ("method","Text-guided optimal transport for unpaired rain removal","ok"),
"4383312437": ("empirical","AI Ecological Education Policy Framework from 457 students and 180 staff (Hong Kong)","ok"),
"7133185348": ("method","Imagen: photorealistic text-to-image diffusion with large frozen text encoders","no_abstract"),
"3195296536": ("method","Dissertation on GAN-based adversarial defence","no_abstract"),
"4322617812": ("theory","Paradox-theory essay on generative AI in education","ok"),
"3096831136": ("survey","CACM review of generative adversarial networks","ok"),
"4400770292": ("empirical","Field experiment: GPT-4 tutoring access improved practice but harmed unassisted exam performance; SSRN preprint","no_abstract"),
"7127274868": ("empirical","Survey study of parental AI investment and informal English learning","ok"),
"7124974472": ("empirical","Nationally representative US surveys of generative-AI adoption (45% of adults 18-64 by late 2024)","ok"),
"7118187240": ("engineering","PatchTST-based photovoltaic forecasting with GAN features (application)","off_scope"),
"4406199489": ("survey","Review of generative AI and integrity in higher education","ok"),
"4398243313": ("survey","Scoping review of 32 studies on generative AI and assessment","ok"),
"4409269614": ("survey","Systematic review (41 studies) of generative AI and academic integrity","ok"),
"3193917007": ("survey","Textbook chapter on GANs","ok"),
"4288099666": ("method","Classifier-free diffusion guidance: joint conditional/unconditional training; preprint","ok"),
"4394934695": ("empirical","Educator interviews on generative AI in higher education","ok"),
"4321648656": ("survey","Review of neural architectures for AI-generated art","ok"),
"4395957952": ("survey","UNESCO guidance for generative AI in education and research","no_abstract"),
"4394863715": ("empirical","SSRN preprint on generative AI and human creativity","no_abstract"),
"4407778399": ("empirical","Analysis of generative-AI guidelines at 50 US universities","ok"),
"7124432017": ("theory","Intelligent-TPACK pedagogy framework","ok"),
"7140170686": ("empirical","Mixed-methods study of human-GenAI pedagogical partnerships","ok"),
"4406728221": ("survey","Survey of agentic AI concepts and applications","ok"),
"7161734976": ("engineering","Robin: multi-agent system automating hypothesis generation and data analysis for experimental biology","ok"),
"7135409568": ("survey","Scoping review of agentic AI in healthcare","ok"),
"7134187809": ("theory","Event-triggered optimal consensus control for nonlinear multi-agent systems (control theory)","off_scope"),
"4390730779": ("survey","Survey of LLM-based agents: definitions, methods, prospects","ok"),
"4225355126": ("survey","Textbook introduction to multi-agent systems","no_abstract"),
"4407570917": ("engineering","AgentSociety: large-scale simulation of LLM-driven generative social agents","ok"),
"4410197190": ("survey","Narrative review of agentic AI in organisations","ok"),
"4413427262": ("survey","Conceptual taxonomy distinguishing AI agents from agentic AI","ok"),
"4414991289": ("engineering","DIVE: multi-agent workflow extracting experimental data from figures for hydrogen-storage materials","ok"),
"7122857583": ("theory","Conceptual agentic-AI framework (literature synthesis)","ok"),
"4401023556": ("survey","Survey of LLM-based multi-agent systems","ok"),
"4414023998": ("survey","Review of 143 primary studies on agentic AI systems","ok"),
"7118083939": ("theory","Conceptual overview of agentic information systems","ok"),
"4412505619": ("survey","Duplicate record of the AI agents vs agentic AI taxonomy (different venue)","ok"),
"4416209823": ("survey","PRISMA-based review (90 studies) with symbolic vs neural dual-paradigm framework for agentic AI","ok"),
"7155930820": ("theory","Prescribed-performance consensus control for multi-agent systems (control theory)","off_scope"),
"7162257503": ("theory","Predefined-time sliding-mode formation control (control theory)","off_scope"),
"4412709685": ("engineering","Virtual Lab: LLM agents designing SARS-CoV-2 nanobodies with experimental validation","no_abstract"),
"7128165235": ("theory","Five-component framework for multi-agent AI in knowledge work","ok"),
"7118190454": ("survey","Perspective on agentic AI for elderly care","ok"),
"7130481397": ("benchmark","Benchmarking two agent systems (OpenManus, Manus) on clinical decision tasks","ok"),
"4390638955": ("survey","Scoping review of AI learning tools in K-12 education","off_scope"),
"4412203333": ("survey","Survey of memory mechanisms of LLM-based agents","ok"),
"4399982941": ("theory","Dissertation on value engineering for autonomous agents (normative MAS)","normalisation_artefact"),
"2911283634": ("survey","Textbook chapter on reinforcement learning","no_abstract"),
"4389473504": ("method","Dissertation on trust-region policy optimisation","ok"),
"2964083594": ("empirical","Large-scale study of curiosity-driven learning across 54 environments (ICLR 2019; OpenAlex year 2025 is a repository copy)","year_implausible"),
"2098774185": ("method","Maximum-entropy inverse reinforcement learning (AAAI 2008; OpenAlex year 2018)","year_implausible"),
"1591675293": ("method","Bayesian inverse reinforcement learning (IJCAI 2007; OpenAlex year 2025 book chapter)","year_implausible"),
"7123365826": ("theory","Event-triggered zero-sum game control for saturated nonlinear systems (control theory)","off_scope"),
"2991046523": ("survey","Selective overview of multi-agent RL theory and algorithms","no_abstract"),
"2973229164": ("method","Dissertation on deep RL for continuous robot control","ok"),
"2982316857": ("engineering","AlphaStar: grandmaster level in StarCraft II with multi-agent RL","no_abstract"),
"2902907165": ("method","AlphaZero: one self-play RL algorithm for chess, shogi and Go","ok"),
"7133481537": ("theory","Predefined-time fault-tolerant consensus control (control theory)","off_scope"),
"7133226216": ("empirical","MAPPO: PPO is surprisingly strong in cooperative multi-agent benchmarks","ok"),
"3041807988": ("theory","Critical essay on ed-tech solutionism after Covid-19 (not RL)","off_scope"),
"2781726626": ("method","Soft actor-critic: off-policy maximum-entropy RL; preprint record","ok"),
"2963923407": ("method","TD3: clipped double-Q and delayed policy updates against overestimation","ok"),
"7122543576": ("theory","ADP-based event-triggered fault-tolerant control (control theory)","off_scope"),
"2899138465": ("method","Dissertation on gradient-based meta-learning (MAML lineage)","ok"),
"4391631327": ("empirical","Pre-registered toy study of two-stage post-training; record carries arXiv 2402.03300 (DeepSeekMath) and its citations","identifier_mismatch"),
"4409147637": ("method","DreamerV3: world-model RL mastering 150+ tasks with one configuration (Nature)","ok"),
"2895974005": ("empirical","Master's thesis on exploration/exploitation parameters","normalisation_artefact"),
"7134057781": ("engineering","Fuzzy-TD3 hybrid control for a 5-DOF manipulator (application)","ok"),
"2754517384": ("empirical","Deep RL That Matters: reproducibility and variance of deep RL results","ok"),
"2761873684": ("method","Rainbow: combining six DQN extensions, with ablations","ok"),
"7133208539": ("method","Direct preference optimisation: closed-form RLHF objective without a separate reward model","no_abstract"),
"7128436456": ("survey","Survey classifying combinations of MPC and RL","ok"),
"2962862931": ("method","Interpretability measure for degradation diagnosis (HAL); c36=12,586 cannot belong to this record","identifier_mismatch"),
"1698155719": ("survey","Textbook chapter on support vector machines","ok"),
"4288283362": ("method","'Spiral Time' geometric time encoding with a small LSTM ablation; no venue; NCS36 implausible","normalisation_artefact"),
"46659105": ("theory","Title = Powers (2011) evaluation-measures paper (arXiv 2010.16061) but the stored abstract is an LLM-as-judge study of program repair","abstract_mismatch"),
"3145506661": ("survey","Textbook: Introduction to Machine Learning (Springer Briefs)","no_abstract"),
"2291445757": ("survey","Textbook chapter on support vector regression","ok"),
"2100415632": ("theory","Bandits with side observations (2014 paper carrying a 2026 arXiv id and year)","year_implausible"),
"1568555062": ("theory","Verma & Pearl: equivalence and synthesis of causal models (reprint chapter, 2022)","ok"),
"2751318774": ("survey","Textbook: Introduction to Machine Learning (Alpaydin)","ok"),
"3208688400": ("survey","Anomaly detection survey (book chapter)","no_abstract"),
"3105488314": ("theory","Dissertation on risk bounds for statistical learning","no_abstract"),
"57706852": ("theory","StoSOO stochastic optimistic optimisation (2013 paper carrying a 2026 arXiv id and year)","year_implausible"),
"2946444988": ("method","Deep Embedded Validation for UDA model selection (2019 paper carrying a 2026 arXiv id; NCS36 implausible)","year_implausible"),
"2945976633": ("theory","Rudin: argument for interpretable models over post-hoc explanation in high-stakes decisions","no_abstract"),
"4400823820": ("survey","Overview of random forests (non-indexed journal)","ok"),
"2999309192": ("theory","Advantages of MCC over F1 and accuracy for imbalanced binary classification","ok"),
"2999615587": ("method","TreeExplainer: exact SHAP values for tree ensembles (NMI)","no_abstract"),
"2439568532": ("theory","Lipton: the mythos of model interpretability","ok"),
"3194464139": ("engineering","Indonesian thesis applying kNN to credit risk; NCS36 implausible for c36=21","normalisation_artefact"),
"3041133507": ("survey","Comprehensive survey of transfer learning (Proc. IEEE)","ok"),
"2436219157": ("survey","Bottou, Curtis, Nocedal: optimisation methods for large-scale ML (SIAM Review)","ok"),
"3135028703": ("survey","Overview of ML algorithms and applications (SN Computer Science)","no_abstract"),
"7124426637": ("survey","Review of machine learning with small and limited data","ok"),
"2165856466": ("theory","Spectral bandits for smooth graph functions (2014 paper carrying a 2026 arXiv id and year)","year_implausible"),
"4402843978": ("survey","UK POSTnote on interpretable ML","ok"),
"2964121744": ("engineering","'AutoResearch Lite' Adam reproduction runner; c36=45,775 and arXiv 2108.04578 cannot belong to this record","identifier_mismatch"),
"1710476689": ("survey","Textbook chapter on GANs (The Science of Deep Learning)","ok"),
"4322614756": ("method","Path-based embedding for road-network community detection; c36=1,542 implausible for a LIPIcs paper","identifier_mismatch"),
"2963399829": ("method","mixup: training on convex combinations of examples (ICLR 2018; OpenAlex year 2024)","year_implausible"),
"4382203079": ("empirical","Zeng et al.: simple linear models beat Transformers on long-term time-series forecasting benchmarks","ok"),
"4406170795": ("method","TabPFN: tabular foundation model (prior-data fitted network) for small data (Nature)","ok"),
"146900863": ("survey","NeurIPS proceedings volume record (MIT Press)","no_abstract"),
"4396821501": ("method","KAN: Kolmogorov-Arnold networks with learnable spline edge functions; preprint","ok"),
"2950067852": ("method","Variational RNN (NeurIPS 2015) recorded as a 2026 preprint; NCS36 implausible","year_implausible"),
"4221150249": ("survey","Survey of GNNs for heterophilic graphs","ok"),
"7148890333": ("method","Multi-scale graph transfer learning for rotating-machinery diagnosis","no_abstract"),
"3177318507": ("method","Informer: ProbSparse attention for long-sequence forecasting","ok"),
"4389326242": ("method","Mamba: selective state-space model with linear-time sequence modelling; preprint","ok"),
"7118302872": ("method","GNN + ant colony optimisation for influence maximisation","ok"),
"3137147200": ("survey","Survey of quantisation methods for efficient inference (book chapter)","ok"),
"3140854437": ("survey","Review of deep learning concepts and CNN architectures","ok"),
"4210257598": ("survey","Comprehensive survey on GNNs (repository duplicate of the TNNLS paper)","ok"),
"2907492528": ("survey","Comprehensive survey on GNNs (TNNLS)","ok"),
"3152893301": ("survey","GNNs: review of methods and applications (AI Open)","ok"),
"7110333790": ("method","CaLLiPer+: POI-name-enriched location embeddings; c36=496 implausible for a LIPIcs paper","identifier_mismatch"),
"4385763767": ("survey","Survey of Transformers in time series","ok"),
"3124006607": ("method","Simple spectral graph convolution (ICLR 2021; OpenAlex year 2026)","year_implausible"),
"4285723986": ("survey","Synthesis lecture on graph attention networks","no_abstract"),
"2892035503": ("survey","Review of deep learning for time-series classification","no_abstract"),
"2962776659": ("empirical","Auxiliary-task analysis of sentence embeddings (ICLR 2017; OpenAlex year 2024)","year_implausible"),
"4406063062": ("survey","Review and research framework for responsible AI governance (IS)","ok"),
"7130596601": ("survey","Review of human-in-the-loop AI in healthcare","ok"),
"3084163594": ("survey","Book review of Zuboff's Surveillance Capitalism for educators","off_scope"),
"2963808661": ("engineering","Governance risk-assessment tooling (OASIcs practitioner track); c36=390 implausible","identifier_mismatch"),
"7133236347": ("survey","TRiSM (trust, risk, security management) for LLM-based multi-agent systems","ok"),
"7128782475": ("empirical","Physician-led red-teaming of four chatbots on 222 patient questions (HealthAdvice)","ok"),
"4406571986": ("empirical","Analysis of generative-AI policies at 116 US R1 universities","ok"),
"2981731882": ("survey","Arrieta et al.: XAI concepts, taxonomies and responsible AI","no_abstract"),
"4402698709": ("survey","AI Risk Repository: 1,725 risks from 74 taxonomies with causal and domain taxonomies","ok"),
"4408314709": ("survey","Review of bias recognition and mitigation in healthcare AI","ok"),
"4407243994": ("theory","Comparative ethical framework (EU/US/China policies)","ok"),
"4330336443": ("empirical","Title = 'GPTs are GPTs' (arXiv 2303.10130) but the stored abstract describes a TRIPS/Anthropic Economic Index comparison","abstract_mismatch"),
"7128600296": ("survey","Systematic review of 35 healthcare-AI governance frameworks and a maturity model","ok"),
"4399450035": ("empirical","Power Hungry Processing: measured inference energy of task-specific vs general-purpose models","ok"),
"4318586239": ("survey","Policy book chapter defining AI","ok"),
"4360615722": ("theory","Qualitative examination of ethical implications of chatbots in education","ok"),
"2670253439": ("survey","Miller: explanation in AI through the lens of the social sciences","no_abstract"),
"7118631510": ("method","Ante-hoc interpretability for Industry 5.0","no_abstract"),
"7127143535": ("empirical","Autonomous AI agents in industrial information integration (paradoxical effects)","no_abstract"),
"4409249263": ("survey","Ethical challenges of AI in clinical practice","ok"),
"4366262984": ("survey","XAI: what is known and what remains for trustworthy AI (Information Fusion)","ok"),
"2969896603": ("survey","Survey on bias and fairness in machine learning (CSUR)","ok"),
"4406152291": ("empirical","Data-poisoning threat assessment: 0.001% of tokens with medical misinformation yields harmful LLMs","ok"),
"4410609100": ("method","PAIR: black-box semantic jailbreaks in about twenty queries","ok"),
"4411542009": ("empirical","Experiments showing LLMs express stigma and respond inappropriately as replacement therapists","ok"),
"4360620450": ("survey","Dwivedi et al.: 43 expert contributions on ChatGPT opportunities and challenges","ok"),
"3185341429": ("survey","Liu et al.: systematic survey of prompt-based learning","ok"),
"4384071683": ("benchmark","Med-PaLM: MultiMedQA benchmark and human evaluation framework for clinical LLM answers","ok"),
"4384918448": ("empirical","Longitudinal ICU simulation of 23 open-weight models; record carries arXiv 2307.09288 (Llama 2) and its citations","identifier_mismatch"),
"4393065402": ("survey","Wang et al.: survey of LLM-based autonomous agents","ok"),
"4319083882": ("theory","van Dis et al.: five priorities for research on ChatGPT (Nature comment)","no_abstract"),
"4399790131": ("survey","Systematic review of over-reliance on AI dialogue systems and cognition","ok"),
"3133702157": ("theory","Bender et al.: risks of ever-larger language models ('stochastic parrots')","ok"),
"4321499901": ("empirical","Qualitative case study of ChatGPT among early adopters in education","ok"),
}
rows = []
for wid, r in L.items():
    ct, desc, flag = J[wid]
    has = bool(A.get(wid, "").strip())
    if flag == "ok" and not has: flag = "no_abstract"
    rows.append({"openalex_id": wid, "title": asc(r["title"])[:140], "year": r["year"], "type": r["type"], "primary_category": r["primary_category"], "doi": r["doi"], "arxiv_id": r["arxiv_id"], "c36": r["c36"], "ncs36": round(float(r["ncs36"]), 1),
                 "contribution_type": ct, "description": desc, "judged_from": "abstract" if has else "metadata_only", "data_quality_flag": flag})
missing = set(L) - set(J); assert not missing, missing
out = f"{R}/literature_synthesis/agent_E/contributions.csv"
with open(out, "w", newline="") as f:
    w = csv.DictWriter(f, fieldnames=list(rows[0].keys())); w.writeheader(); w.writerows(rows)
from collections import Counter
print(len(rows), "rows ->", out); print(Counter(r["data_quality_flag"] for r in rows)); print(Counter(r["contribution_type"] for r in rows)); print(Counter(r["judged_from"] for r in rows))