docs: add XAI section and research grounding to README and concepts

Author: adrida

README.md

@@ -79,6 +79,31 @@ The surrogate is **not another LLM** - it is a classical ML or shallow DL model
 3. **Calibrate** - sweep the acceptor threshold to maximise coverage at your target parity (e.g. ≥ 95% teacher agreement)
 4. **Guard** - block deployment if the best candidate cannot clear the parity bar on held-out data
 
+## Explainability
+
+Every fit produces a structured audit report alongside the routing policy -- not just aggregate numbers.
+
+| Report component | What it answers |
+|-----------------|-----------------|
+| **Per-label slice stats** | Which intents are fully self-serve? Which always defer? |
+| **Contrastive boundary pairs** | For each label: one handled example, one deferred -- what makes an input "easy" vs "hard"? |
+| **Representative examples** | What does typical handled/deferred traffic look like? |
+| **Temporal deltas** | Which intents improved between the last refit and this one? |
+
+```python
+result = tracer.fit("traces.jsonl", embeddings=X)
+qr = result.qualitative_report
+
+print(qr.summary)
+# "Handled 9278/10003 (92.8%) by surrogate"
+
+for bp in qr.boundary_pairs[:3]:
+    print(f"{bp.teacher_label}: handled={bp.handled_preview!r} / deferred={bp.deferred_preview!r}")
+
+# Generate self-contained HTML report
+tracer.generate_html_report(".tracer")
+```
+
 ## Benchmark results (Banking77 - 77-class intent classification)
 
 | Metric | Value |
@@ -163,6 +188,18 @@ pip install tracer-llm[all]           # everything
 | [Artifacts](docs/artifacts.md) | `.tracer/` directory schema |
 | [AGENTS.md](AGENTS.md) | Integration guide for AI coding assistants |
 
+## Prior work
+
+TRACER builds on and connects three research threads:
+
+**Learn to Defer (L2D)** -- the algorithmic backbone. The surrogate + acceptor gate architecture is grounded in Mozannar & Sontag ([ICML 2020](https://proceedings.mlr.press/v119/mozannar20b.html)), Madras et al. ([NeurIPS 2018](https://papers.nips.cc/paper_files/paper/2018/hash/09d37c08f7b129e96277388757530c72-Abstract.html)), and Mao et al. ([NeurIPS 2023](https://proceedings.neurips.cc/paper_files/paper/2023/hash/7e1bf09d2e4e6e5c5a0b63b66b47ef42-Abstract.html)).
+
+**Post-hoc calibration** -- the threshold sweep that enforces the parity guarantee follows the post-hoc deferral estimator framework of Narasimhan et al. ([ICML 2022](https://proceedings.mlr.press/v162/narasimhan22a.html)).
+
+**LLM routing** -- the problem setting is shared with Hybrid LLM (Ding et al., [ICLR 2024](https://openreview.net/forum?id=02f3mUtqnM)) and RouteLLM (Ong et al., [ICLR 2025](https://openreview.net/forum?id=sSt9fROCn2)). TRACER's distinction: it targets classification specifically, uses the teacher's own trace distribution rather than preference data, and enforces a formal parity guarantee rather than a quality heuristic.
+
+**Interpretability** -- the qualitative audit design is informed by Slice Finder (Chung et al., [ICDE 2019](https://ieeexplore.ieee.org/document/8731353)) and the contrastive explanation literature (Miller, [AIJ 2019](https://www.sciencedirect.com/science/article/pii/S0004370217301126)).
+
 ## Paper
 
 A research paper detailing the approach, formal guarantees, ablation studies, limitations, and reproducible experiment tooling is in preparation. It will be linked here upon publication.

docs/concepts.md

@@ -38,13 +38,13 @@ A second binary classifier that predicts: "will the surrogate's prediction match
 
 ### Threshold calibration
 
-The acceptor threshold τ is calibrated on a held-out set (not used for training). The calibration objective: find the threshold that maximizes coverage subject to teacher agreement ≥ target. This is a conformal-style guarantee: on the calibration distribution, you get the promised agreement rate.
+The acceptor threshold τ is calibrated on a held-out set (not used for training). The calibration objective: find the threshold that maximizes coverage subject to teacher agreement ≥ target. This is a conformal-style guarantee: on the calibration distribution, you get the promised agreement rate. The approach follows the post-hoc deferral estimator framework of Narasimhan et al. ([ICML 2022](https://proceedings.mlr.press/v162/narasimhan22a.html)).
 
 ---
 
 ## Three pipeline families
 
-TRACER fits three candidate architectures and picks the one with highest coverage at your target teacher agreement.
+TRACER fits three candidate architectures and picks the one with highest coverage at your target teacher agreement. The L2D and RSB families are grounded in the Learn-to-Defer literature (Mozannar & Sontag, [ICML 2020](https://proceedings.mlr.press/v119/mozannar20b.html); Madras et al., [NeurIPS 2018](https://papers.nips.cc/paper_files/paper/2018/hash/09d37c08f7b129e96277388757530c72-Abstract.html); Mao et al., [NeurIPS 2023](https://proceedings.neurips.cc/paper_files/paper/2023/hash/7e1bf09d2e4e6e5c5a0b63b66b47ef42-Abstract.html)).
 
 ### Global
 
@@ -136,7 +136,7 @@ Coverage grows monotonically as more traces accumulate. Easy intents (tight embe
 
 ## Qualitative audit
 
-Every fit produces a structured explanation of the routing policy:
+Every fit produces a structured explanation of the routing policy. The slice and boundary-pair design is informed by Slice Finder (Chung et al., [ICDE 2019](https://ieeexplore.ieee.org/document/8731353)) and the contrastive explanation literature (Miller, [AIJ 2019](https://www.sciencedirect.com/science/article/pii/S0004370217301126)).
 
 **Slice summaries**: per-label and per-length-bucket handled/deferred rates. Answers: "which parts of my label space are well-covered?"