Condense the 2.0 changelog into curated release notes

Consume the 355 accumulated news fragments with towncrier and replace the
generated output with a hand-curated 2.0.0 section: a Highlights list
ordered by significance (intrinsic units, multi-valued units, closed-form
relations, matching, the 2026 formalism default, performance), then the
eight fragment categories with related entries merged and intermediate
development states collapsed into the final 1.2.0 -> 2.0 delta. Fixes for
behavior that never shipped are compressed into a pointer at git history.

Set towncrier's start_string to the marker comment in CHANGELOG.md; without
it, builds prepend new sections above the file header instead of inserting
at the marker. Update the P15 roadmap row.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PEAxNzhDCaTrntX3o1JqMV

Author: wmayner

CHANGELOG.md

@@ -44,7 +44,7 @@ P0 · P1 · P2 · P3 · P4 · P5 · P6 · **P6a** (free-threaded 3.14t CI lane +
 | P11 cluster backends | ✅ landed | 4 | Both halves shipped. **Dask backend** (2026-07-20): `DaskScheduler` distributes map-reduce over a user-connected `distributed.Client` (snapshot propagation, cost-balanced chunks, deterministic short-circuit prefixes; nested dispatch runs in-task); `cluster` extra; `docs/howto/chtc.md` covers CHTC (the Dask worker-pool pattern remains documented as a pilot pending CHTC port-access confirmation — an external facilitation question, not code work). **Campaign surface** (cycles 1+2, 2026-07-20/21): `pyphi.campaign` tasks-as-data batch campaigns — sweep cells (substrates axis on `sweep()`, cost-balanced packing, admission control, condor emitter, campaign ≡ local sweep proven end-to-end) and scoped CES sharding for one system: declarative `AxisScope`/`CESScope` (explicit lists, order bounds, unit containment; serialized into tasks and provenance), scope-aware `estimate_analysis` + per-mechanism `mechanism_workloads`, a three-rung planner (mechanism → purview-range → partition-stride, `bottleneck_first` ordering for sparse reducibility), exact tie-preserving merges (sharded ≡ unsharded verified under both IIT 4.0 presets, winner identity included — per-shard winners restored to global enumeration order), three SIA modes (merged strides / precomputed / intrinsic-state with no Φₛ), version+scheme merge guards, and a certified scope report (exact Σφ_r lower bound + measured upper bounds from `iit4/bounds.py`). MCP: `prepare_campaign`, `prepare_ces_campaign`, `campaign_status`, `collect_campaign`, scope-aware `estimate_cost`, `campaigns` topic; `docs/howto/campaigns.md`. Specs: `docs/superpowers/specs/2026-07-20-htcondor-campaign-design.md` + `2026-07-20-ces-sharding-design.md`. **Follow-ups (2026-07-21, merge `c2229d20`):** `prepare_ces` now sweeps substrates × states × subsets × formalisms under one scope (plan-once-per-group, multi-cell `SweepResult` collection, per-cell scope reports); per-shard `request_memory` estimated from the largest purview repertoire with memory-stratified packing (`request_memory` = floor); order-dependent purview caps (`CESScope.max_purview_order_by_mechanism_order` via the single `purview_axis` selection point); `limit=` threading with a single planning walk; SIA short-circuit handling on weakly-connected subsets; fat-node crossover criteria in the how-tos. Spec: `2026-07-21-scoped-ces-campaign-followups-design.md`. **Production-scale fixes (2026-07-22):** the first 21-unit campaign surfaced four gaps, all landed — purview order caps threaded into the enumeration itself, substrate files at raw-array size (msgpack + verified factor trim, `FORMAT_VERSION` 2), per-cell resolution states for multi-cell campaigns, and specified-state computation scaling past ~14 units; see the 2026-07-22 section below. |
 | Parallel engine unification | ✅ landed | 4 | Collapsed the two coexisting map-reduce implementations into one flat path: replaced the one-shot `MapReduce` class with a `pyphi.parallel.map_reduce()` function over the Scheduler Protocol, migrated all 10 call-sites, deleted the vestigial execution tree (`tree.py`/`test_tree.py` + dead `branch_factor`/`max_*` kwargs), and deduped the ~5 copied helpers. Backend selection (`config.parallel_backend`) is now honored from the public entry. Net ~900 lines removed; result-preserving (N2 + goldens + perf gate green). Prerequisite for B18. Spec/plan: `docs/superpowers/{specs,plans}/2026-06-23-unify-parallel-engine*` |
 | B18 adaptive chunking | ✅ landed | 4 | Activated `ChunkingPolicy.size_func`: a pure `cost_balanced_partition` LPT-packs items into cost-balanced chunks, and the chunk count is floored at `num_workers` (a general core-utilization win, de-risked empirically — 2.8–3× at every per-item cost, no overhead penalty). Cheap `size_func` proxies wired at the heterogeneous sites — relations (`overlap×degree`), purview eval (`2^|purview|`), concept eval (`2^|mechanism|`). Result-preserving (N2 incl. a forced-chunk concept test, goldens, perf gate). `pyphi/parallel/chunking.py`; spec/plan `docs/superpowers/{specs,plans}/2026-06-23-cost-balanced-chunking*`. **Dispatch-gate follow-up (2026-07-07)**: measured the below-chunksize regime (`benchmarks/b18_dispatch_gate.py` + `benchmarks/b18_dispatch_gate_results/`) and flipped the gate — `sequential_threshold` is now the sole dispatch gate; an explicit `chunksize` governs granularity only (a cost-sampled chunksize still gates, since it estimates ~1 s of work). Wins: purview eval 3–4× at 64–230 purviews, system partitions 2.5–4× at 64–2048, complexes 1.2–1.6× at 16–31 candidates; no regression on reducible short-circuiting systems. Per-level thresholds retuned to measured per-item costs (partition 1024→64, mech-partition & relations 1024→8192). The study also surfaced and fixed a per-item `hash(repr(snapshot))` (~1.3 ms) in the process-scheduler worker wrapper that dominated cheap-item workloads (~250× sequential relations speedup on its own) |
-| P15 | 🟡 partial | 5 | Import-surface cleanup landed (eager submodule walk → explicit registry imports + PEP-562 lazy `__getattr__`, guarded by a registry-contents test). jsonify→msgspec landed (`pyphi.serialize` replaces `pyphi.jsonify`; the per-class `to_json`/`from_json` hooks removed) **plus its loading surface** (path-based `serialize.save`/`load`, top-level `pyphi.save`/`pyphi.load`, `.save()`/`.load()` on the result types via a delegation-only `Serializable` mixin, deferred type registration, `substrate.from_json` removed, transparent gzip on `.gz` paths). pandas-extend done (display → B21). Project-stage markers stripped from `pyphi/` + a docstring narrative-removal slice landed. PR triage done (audit against `main`: nothing left to absorb; closes pending a maintainer). Test suite reorganized to mirror the `pyphi/` package layout (`test/<subpackage>/` + `test/integration/`). Remaining: the comprehensive docstring/docs sweep (now its own [Documentation overhaul](#documentation-overhaul-20--iit-40) project), changelog condense |
+| P15 | 🟡 partial | 5 | Import-surface cleanup landed (eager submodule walk → explicit registry imports + PEP-562 lazy `__getattr__`, guarded by a registry-contents test). jsonify→msgspec landed (`pyphi.serialize` replaces `pyphi.jsonify`; the per-class `to_json`/`from_json` hooks removed) **plus its loading surface** (path-based `serialize.save`/`load`, top-level `pyphi.save`/`pyphi.load`, `.save()`/`.load()` on the result types via a delegation-only `Serializable` mixin, deferred type registration, `substrate.from_json` removed, transparent gzip on `.gz` paths). pandas-extend done (display → B21). Project-stage markers stripped from `pyphi/` + a docstring narrative-removal slice landed. PR triage done (audit against `main`: nothing left to absorb; closes pending a maintainer). Test suite reorganized to mirror the `pyphi/` package layout (`test/<subpackage>/` + `test/integration/`). Changelog condensed into curated 2.0.0 release notes (2026-07-23: 355 fragments consumed; hand-edited Highlights + category sections in `CHANGELOG.md`). Remaining: the comprehensive docstring/docs sweep (now its own [Documentation overhaul](#documentation-overhaul-20--iit-40) project) |
 | B17 drop dead deps | ✅ landed | 5 | Removed the declared-but-unused `tblib`; deleted the unused `OrderedSet`/`HashableOrderedSet` types (a sorted tuple is the better unit representation) and dropped `ordered-set`; migrated `toolz` to stdlib `itertools.chain.from_iterable`, `more_itertools.unique_everseen`, and `functools.partial`. Landed with the import-surface cleanup; spec/plan `docs/superpowers/{specs,plans}/2026-06-24-import-surface-cleanup*` |
 | MCP server | ✅ landed | — | Local Model Context Protocol server (`pip install pyphi[mcp]`, `pyphi-mcp` console script; `pyphi/mcp/`) exposing PyPhi to AI assistants: tools wrapping `pyphi.analyze`/examples/`Substrate`/`serialize` (build/load/describe substrates, `analyze` with a soft node-count guardrail + compact-vs-full detail, `inspect` into result slices, `plot` surfacing PyPhi's built-in visualizations — ces/repertoires/connectivity/tpm, plotly→interactive HTML + matplotlib→PNG), plus a bundled IIT teaching layer surfaced three ways (always-loaded `instructions`, `pyphi://theory/*` resources, and a `get_iit_reference` tool) with citation-checked reference content authored from `papers/`, and two workflow prompts. Built on the official `mcp` SDK's FastMCP; doctest collection guarded when the extra is absent. `test/mcp/test_server.py` (known-Φ regression; 2026 differentiation-cap and IIT-3.0 cases); how-to `docs/howto/mcp-server.md`. **Parallelization surfaces (2026-07-16):** a `parallelization` reference topic (the conditions under which work actually parallelizes, the seven levels, which levels pay for which workloads), per-call `analyze(parallel=…, workers=…)`, and a persistent `configure_parallel` tool; alongside, the `parallel_concept_evaluation` config option was renamed to `parallel_distinction_evaluation` (it governs the distinctions loop under every formalism) and the parallel how-to's two factual errors were fixed (the global flag is a gate, not a switch; no `pyphi[parallel]` extra exists) |
 | EMD backend → POT | ✅ landed | 5 | Swapped deprecated `pyemd` for POT (`ot.emd2`) behind `OptionalEMD`; `pyphi[emd]` extra now installs `pot`. Backends agree to machine epsilon (binary + k-ary, 3600 cases); the IIT 3.0 CES-distance EMD was reformulated to proper non-negative OT (the negative-null-mass / signed-φ case), which **reproduces the existing goldens exactly** — no regen needed. |