Add files using upload-large-folder tool

DATASHEET.md

@@ -0,0 +1,42 @@
+# Datasheet — data accompanying the TempGraphRAG methods paper
+
+Anonymous double-blind artifact. Covers the data shipped to reproduce the methods
+paper. The benchmark itself is described in concurrent anonymous work
+(`anon-bench`); here it is a read-only input.
+
+## Provenance
+- **Source KG:** `tkgl-smallpedia` from TGB 2.0, derived from **Wikidata**
+  (public, **CC0**). 550,376 timestamped edges, 47,433 entities, 283 relations,
+  span 1900–2024.
+- **Benchmark questions** (`benchmark_labelled.jsonl`): 8,710 multi-hop
+  temporal-KGQA questions, each with a gold supporting subgraph ($S^{*}$),
+  same-$(s,r)$ distractors ($S_{\mathrm{dist}}$), and stale-fact set
+  ($S_{\mathrm{stale}}$), stratified over a 4×3 operator × complexity matrix.
+  70/10/20 train/dev/test split (6,096 / 871 / 1,743). Generated from a
+  TimelineKGQA-derived pipeline (see `anon-bench`).
+- **SFT corpora** (`sft_data_5k_cot.jsonl`, `*_terse_lever.jsonl`,
+  `*_multitq_5k.jsonl`): retrieval-conditioned traces derived from the training
+  split; used to fine-tune the generator.
+- **Evaluation outputs** (`outputs/eval/*.json`): per-question prediction + gold
+  for every reported run (3-seed headline, ablations, judge runs, MultiTQ).
+
+## Composition / splits
+Test results in the paper are on the 1,743-question test split. The train/test
+splits share the source KG; the paper measures and reports contamination (0.40%
+verbatim-question overlap, 5.1% anchor overlap) and treats absolute TempBench EM
+as an in-distribution upper bound, with MultiTQ as the cross-distribution check.
+
+## Intended use
+Research reproduction of retrieval-augmented temporal multi-hop KGQA. Not
+intended as a production correctness oracle. The triple-level grounding metric
+(appendix) is a population-level diagnostic, not a per-instance guarantee.
+
+## Personal / sensitive data
+Questions concern historical and public-figure facts as exposed by Wikidata; no
+personal data beyond Wikidata's public content. No human-subject data in this
+archive. (Human inter-annotator validation is reported with `anon-bench`, not
+here.)
+
+## License
+CC0-derived (Wikidata) via TGB 2.0; redistribution under permissive terms
+consistent with TGB 2.0. Trained LoRA adapters are released for research use.

PROVENANCE.txt

@@ -0,0 +1,6 @@
+The temporal-KGQA benchmark used here (outputs/benchmark/benchmark_labelled.jsonl,
+labels.tsv) is described in concurrent anonymous work cited in the paper as
+[anon-bench]. It is included in this archive as a read-only INPUT required to
+reproduce the methods paper's results, and is NOT a contribution of this paper.
+
+Underlying KG: tkgl-smallpedia (TGB 2.0), derived from Wikidata (CC0).

README.md

@@ -0,0 +1,42 @@
+# Temporal-Aware GraphRAG — anonymous data archive
+
+Anonymous double-blind artifact accompanying the EMNLP submission *"The Lever Is
+the Prompt: Retrieval-Conditioned Supervision in Temporal-Aware GraphRAG."* This
+is the **data + trained-adapter** half; the **code** is in the companion anonymous
+repository linked from the paper. See `DATASHEET.md` for provenance and license.
+
+> No author, affiliation, or identifying information is included. Please do not
+> attempt to de-anonymize.
+
+## Contents
+```
+adapters/      LoRA adapters (inference-ready: adapter_model.safetensors +
+               adapter_config.json + tokenizer + chat_template). 14 policies:
+               sft-v3{,-seed1337,-seed7}        Qwen3-8B headline (3 seeds)
+               sft-llama31{,-seed1337,-seed7}    Llama-3.1-8B cross-arch (3 seeds)
+               sft-mistral{,-seed1337,-seed7}    Mistral-7B cross-arch (3 seeds)
+               sft-multitq{,-seed1337,-seed7}    MultiTQ cross-benchmark (3 seeds)
+               sft-multitq-{llama,mistral}       MultiTQ cross-arch
+eval/          Per-question predictions + gold for every reported run (74 JSONs):
+               TempBench 3-seed, 3-hop ablation, empty/shuffled-evidence,
+               data-scale (1k/2k), full-test LLM-judge runs, MultiTQ, baselines.
+benchmark/     benchmark_labelled.jsonl + labels.tsv (the benchmark of [anon-bench],
+               provided as a read-only input; see PROVENANCE.txt).
+sft_data/      Retrieval-conditioned SFT corpora (CoT, terse, MultiTQ).
+logs/          Training / evaluation logs (negative-result evidence).
+```
+
+## Mapping to the code repository
+Place these so the code repo's scripts find them:
+```
+adapters/<name>/        ->  checkpoints/<name>/final/
+eval/*.json             ->  outputs/eval/
+benchmark/*             ->  outputs/benchmark/
+sft_data/*.jsonl        ->  outputs/
+```
+
+## Notes
+- Adapters are **inference-only** (training state / optimizer checkpoints removed).
+- Three contaminated judge runs (errored API calls) are **excluded**; all reported
+  judge-EM numbers come from the clean full-test runs included here.
+- License: derived from Wikidata (CC0) via TGB 2.0 — permissive (see `DATASHEET.md`).