Add adaptive context compilation study

.github/workflows/ci.yml

@@ -37,6 +37,8 @@ jobs:
         run: scripts/eceb-conformance.sh
       - name: Replay public ECEB study
         run: scripts/eceb-study-check.sh
+      - name: Replay adaptive ECEB study
+        run: scripts/eceb-adaptive-study-check.sh
       - name: Build documentation
         run: cargo doc --all-features --no-deps --locked
       - name: Package

.github/workflows/pages.yml

@@ -30,6 +30,7 @@ jobs:
           cp -a site/. _site/
           cp spec/symgliph-context-glyph-v0.1.md _site/spec/SCGP-0.1.md
           cp spec/eceb-v0.1.md _site/spec/ECEB-0.1.md
+          cp spec/adaptive-context-v0.2.md _site/spec/SACP-0.2.md
       - name: Configure Pages
         uses: actions/configure-pages@v5
       - name: Upload Pages artifact

CHANGELOG.md

@@ -16,6 +16,10 @@ the project intends to use Semantic Versioning after its first stable release.
   conformance seed, executable evaluator, and negative policy/provenance gates.
 - Deterministic 30-task Multi-SWE-bench Rust study suite, four-arm runner,
   comparative study schemas, raw golden run records, and offline replay gate.
+- Experimental SACP-0.2 adaptive context profile, deterministic expansion
+  engine, CLI and JSON-RPC entry points, auditable stopping trace, schema, and
+  optional Multi-SWE-bench study arm. The replay evaluator binds every trace to
+  its declared ladder and sufficiency policy and rejects premature limit claims.
 
 ### Changed
 
@@ -34,5 +38,12 @@ the project intends to use Semantic Versioning after its first stable release.
 - The Multi-SWE-bench Rust v0.1 study reduced paired prompt tokens by 97.58% and
   localization-model spend by 98.54% with Symgliph compiled context, while exact
   fix-path recall declined from 53.32% to 41.76%; lexical BM25 retained 46.87%.
+- A retrieval-only adaptive threshold sweep selected 50% query-term coverage as
+  the lowest tested setting that retained the fixed compiled arm's 51.85%
+  evidence recall. This is in-sample calibration, not model-quality evidence.
+- The adaptive 30-task follow-up retained 51.85% evidence recall, reduced mean
+  compiled prompt tokens from 5,138 to 4,345, improved exact fix-path recall
+  from 41.76% to 45.09%, and reduced recorded localization spend from $0.0668
+  to $0.0573. It remains below BM25 and full-context recall.
 
 [Unreleased]: https://github.com/copyleftdev/symgliph/commits/main

README.md

@@ -71,6 +71,39 @@ this run, but it did **not** preserve quality parity: exact fix-path recall was
 lexical BM25. The study supports the efficiency hypothesis and identifies
 retrieval quality as the next research problem.
 
+## Adaptive context experiment
+
+The experimental SACP-0.2 profile addresses that gap by walking an ascending
+token and semantic-node ladder. It stops at the first packet that passes a
+declared evidence-sufficiency heuristic, when candidates are exhausted, or when
+hard limits are reached. Every attempted context digest and stopping condition
+is retained for evaluation.
+
+```text
+query -> retrieve -> pack -> estimate sufficiency
+                           ├── pass: emit packet
+                           └── fail: expand budget and evidence
+```
+
+The built-in estimator measures query-term coverage, evidence count, and
+optional excerpt completeness. It is deliberately named `estimated_sufficient`:
+passing it does not establish retrieval recall or answer quality. A follow-up
+Multi-SWE-bench run tested the policy against full context, BM25, and the fixed
+compiled arm.
+
+The provisional 50% query-term threshold was selected by an exploratory replay
+of the existing 30-task traces. It preserved the fixed compiled arm's evidence
+recall while reducing selected context, but it is not independent validation.
+See the [calibration record](docs/adaptive-calibration.md).
+
+In that follow-up, adaptive context retained the fixed arm's 51.85% evidence
+recall while reducing mean prompt tokens from 5,138 to 4,345 and improving
+exact fix-path recall from 41.76% to 45.09%. It still trailed BM25 and full
+context. Read the [adaptive study and claim boundary](docs/eceb-adaptive-study.md).
+
+Read the experimental
+[`Adaptive Context Compilation 0.2`](spec/adaptive-context-v0.2.md) profile.
+
 Read the [methodology and claim boundary](docs/eceb-multiswe-study.md), inspect
 the [raw report](benchmarks/results/eceb-multiswe-rust-v0.1/report.json), or
 recompute it without API access:
@@ -90,6 +123,13 @@ cargo run -- pack ./my-corpus \
   --query "where is authorization checked?" \
   --max-tokens 4000
 
+# Start small and expand until the declared evidence estimator passes.
+cargo run -- adaptive-pack ./my-corpus \
+  --query "where is authorization checked?" \
+  --min-tokens 1000 \
+  --max-tokens 8000 \
+  --require-sufficient
+
 # Recompute the corpus and semantic roots against current source.
 cargo run -- verify ./my-corpus
 ```
@@ -126,9 +166,13 @@ Every evidence record identifies its semantic node, source path, byte range,
 full-file digest, and excerpt digest. If a selected file changed after
 compilation, packing fails rather than emitting stale evidence.
 
+`ContextEngine::pack_adaptive` accepts an `AdaptiveContextRequest` and returns
+the selected packet plus the complete expansion trace. A failed estimate still
+returns a valid verified packet unless the caller elects fail-closed behavior.
+
 For local model and agent bridges, `symgliph serve` exposes
-`context.describe`, `context.pack`, and `context.proof` over newline-delimited
-JSON-RPC:
+`context.describe`, `context.pack`, `context.adaptive`, and `context.proof` over
+newline-delimited JSON-RPC:
 
 ```json
 {"jsonrpc":"2.0","id":1,"method":"context.pack","params":{"query":"authorization checks","max_tokens":2000,"max_nodes":8}}
@@ -160,6 +204,7 @@ Run the independent gates:
 spec/validate.sh                 # SCGP artifact and packet conformance
 scripts/eceb-conformance.sh      # ECEB policy, provenance, and budget gates
 scripts/eceb-study-check.sh      # replay the public comparative study
+scripts/eceb-adaptive-study-check.sh # replay the adaptive follow-up
 ```
 
 ## Artifact model
@@ -202,6 +247,12 @@ project code or build scripts. Additional analyzers can implement the
   A/B runs, Fabric instruction recovery, and blind semantic discovery.
 - [ECEB-0.1](spec/eceb-v0.1.md): a quality-, policy-, provenance-, and
   budget-aware measurement protocol.
+- [SACP-0.2](spec/adaptive-context-v0.2.md): deterministic expansion, explicit
+  stopping semantics, and auditable evidence-sufficiency estimates.
+- [Adaptive calibration](docs/adaptive-calibration.md): threshold sweep,
+  negative findings, and the preregistered next-run setting.
+- [Adaptive ECEB study](docs/eceb-adaptive-study.md): the prospective model
+  run, complete traces, measured improvement, and remaining quality gap.
 - [Multi-SWE-bench Rust study](docs/eceb-multiswe-study.md): the current
   four-arm public result and raw observations.
 - [Semantic-routing golden set](docs/golden-set.md): linked corpora, queries,