test(viewer): session-ledger dedup compiler proptest surface (WBS-6.2 #468)

CHANGELOG.md

@@ -70,7 +70,7 @@ Follows [Keep a Changelog](https://keepachangelog.com/); versioning is [SemVer](
 - session-ledger heuristic intent extractor property surface (WBS-6.2 #463): `crates/sl-viewer/tests/properties_session_ledger_intent.rs` adds 10 proptest properties pinning `session_ledger::distill::extractor::HeuristicIntentExtractor::extract_intent` (the P1 SSOT for what the user wants — drives resume prompt, search index, and wiki/docs view). `user_turn_count` always equals the count of `Role::User` messages and ignores `Assistant` / `Subagent` / `Tool` / `System` messages. Empty sessions produce an empty `Intent`. Repeated acceptance / constraint patterns are deduplicated. Every documented acceptance pattern (16 of them: `looks good`, `works`, `that's correct`, `correct`, `done`, `fixed`, `passes`, `approved`, `looks right`, `looks great`, `all good`, `that works`, `nice`, `perfect`, `exactly`, `confirmed`) and every documented constraint pattern (19 of them: `don't change`, `do not change`, `must not`, `should not`, `keep`, `maintain`, `preserve`, `never`, `don't touch`, `do not touch`, `don't modify`, `do not modify`, `only`, `but don't`, `but do not`, `without changing`, `without modifying`, `leave alone`, `leave as is`) is recognized in any user message. Labeled `Goal:` / `Objective:` / `Task:` lines win over surrounding preamble. Labeled `Constraint:` / `Requirement:` / `Boundary:` lines contribute their full text. The extractor is deterministic across calls.
 - session-ledger heuristic context extractor property surface (WBS-6.2 #464): `crates/sl-viewer/tests/properties_session_ledger_context.rs` adds 20 proptest properties pinning `session_ledger::distill::context_extractor::HeuristicContextExtractor::extract_context` (the P2 adapter for [`ContextExtractor`] — drives the working-context slice of any `ContinuationBundle`). `cwd` and `title` are always copied verbatim from the session. Empty sessions and whitespace-only messages yield an empty `Context`. Every detected file path is non-empty, ≥3 chars, and either contains a `/` or ends with a documented extension. Every documented file extension (28 of them: `.rs`, `.ts`, `.tsx`, `.js`, `.jsx`, `.py`, `.go`, `.java`, `.kt`, `.rb`, `.c`, `.h`, `.cpp`, `.hpp`, `.cs`, `.swift`, `.toml`, `.json`, `.yaml`, `.yml`, `.md`, `.sql`, `.css`, `.scss`, `.html`, `.sh`, `.tf`, `.lock`) is detected. Files / symbols / decisions / environment notes are deduplicated across messages. Decision summaries all read `Session contains '<pattern>' language` and rationale equals the source message. Every documented decision pattern (15 of them) and every documented environment pattern (19 of them) is detected, both case-insensitively. Symbols are extracted only from tokens containing `::` or `()`. `extract_context` is deterministic across calls, and the `ContextExtractor` trait path returns the same `Context` as the associated function.
 - session-ledger heuristic contract extractor property surface (WBS-6.2 #465): `crates/sl-viewer/tests/properties_session_ledger_contract.rs` adds 18 proptest properties pinning `session_ledger::distill::contract_extractor::HeuristicContractExtractor::extract_contract` (the P2 adapter for [`ContractExtractor`] — drives the acceptance-contract slice of any `ContinuationBundle`). Empty sessions and whitespace-only messages yield an empty `Contract`. Every section string has a documented prefix and suffix shape (`Goal/requirement: '...'`, `Verification: '...'`, `Constraint: '...'`, `Do-not-touch: '...'`). Every documented criteria pattern (12 of them), test pattern (21 of them), constraint pattern (15 of them), and do-not-touch pattern (14 of them) is detected, and all patterns match case-insensitively. Success criteria / tests / constraints / do-not-touch lists are deduplicated across messages. `extract_contract` is deterministic across calls, and the `ContractExtractor` trait path returns the same `Contract` as the associated function. `is_empty()` is true iff every collection is empty.
-- session-ledger token estimator property surface (WBS-6.2 #467): `crates/sl-viewer/tests/properties_session_ledger_token_estimator.rs` adds 24 proptest properties pinning `session_ledger::distill::token_estimator::CharCountTokenEstimator` (the bounded token-budget accountant for the distill pipeline — backs every `token_estimate()` query on a `Bundle` slice). `estimate_text` always equals `(chars + 3) / 4` (ceiling-division rounding up at every 4-char boundary), the documented fixed points (`n=1..=9` → `1, 1, 1, 1, 2, 2, 2, 2, 3`) hold, and the estimator counts unicode scalar values (chars) — not UTF-8 bytes (4 × `🦀` = 4 chars = 1 token, even though those are 16 UTF-8 bytes). `estimate_text` is deterministic, finite (u32), never panics on any unicode input (CJK / RTL / emoji / combining marks), and is monotonic for prefixes (`est(text) ≥ est(prefix)`). `estimate_text(s.repeat(n))` follows the exact `(chars(s) * n + 3) / 4` contract. `estimate_json(&v)` always equals `estimate_text(&v.to_string())` — i.e. it uses compact serialisation with no whitespace — and is total over `Null / Bool / Number / String / Array / Object`. `CharCountTokenEstimator` is `Default + Copy + Clone`, and the trait-default `estimate_json` delegates to `estimate_text` (verified with a dummy estimator that overrides `estimate_text`).
+- session-ledger dedup compiler property surface (WBS-6.2 #468): `crates/sl-viewer/tests/properties_session_ledger_dedup.rs` adds 15 proptest properties pinning `session_ledger::distill::dedup_compiler::DedupCompiler::compile` (the deterministic scope-keyed manifest pipeline that compiles same-scope sessions into a typed, token-sized Dedup slice). Empty sessions always return `Err(EmptySessions)`. Whitespace-only or empty topic returns `Err(EmptyTopic)`. Topic is trimmed and lowercased before use. Output bundle always has `kind = BundleKind::Dedup` and a positive `token_estimate`. Manifest body always deserializes to `DedupManifest`. `dedup_key` is exactly 64 hex chars (sha256 fingerprint), independent of input order, and differs across topics and cwds. Duplicate members (same id + corpus) appear once preserving first occurrence order. Members preserve input order. Different corpora for same cwd produce different members. Mixed-scope sessions return `Err(ScopeMismatch)` with the offending session_id. Compilation is deterministic across calls and `token_estimate` is monotonic non-decreasing in session count.
 - Commit signing header scan (C04 L34): `commit-signing-check.ps1` reads bounded commit headers via line-scanner (no unbounded `git cat-file` buffers or `(?ms)` regex); `-SelfCheck` + `tests/commit_signing_check.rs`.
 
 - Loom permutation CI timeout (P0 stability): split blocking `loom-permutation.yml` into core + per-daemon `loom_model` jobs with `LOOM_MAX_PREEMPTIONS` on broadcast/pipeline/shutdown; mirror in soft `loom-smoke.yml` so Wave-40 tokio-shaped daemon graph tests no longer exceed single-job ceilings.