# Changelog

All notable changes to [ProvSQL Studio](https://provsql.org/docs/user/studio.html)
are documented in this file. Studio's version stream is independent of
the ProvSQL extension's; the
[compatibility matrix](https://provsql.org/docs/user/studio.html#compatibility)
in the user manual records each Studio release's minimum required
extension version.

This file is **maintained by hand at release time** by the Studio
maintainer (mirroring how `release.sh` maintains the extension's
top-level `CHANGELOG.md`); pull requests should not modify it. The
release workflow (`.github/workflows/studio-release.yml`) extracts the
section matching the tag's version and embeds it under "What's
changed" in the GitHub release notes.

## [1.9.0] - 2026-10-17

Companion release for ProvSQL extension 1.13.0, which reports every
refusal and every untracked part of a query with a reason tag, and adds
resource limits, write-once probabilities, and recursion by equation
systems. Studio shows all of these where they occur, and its settings
panel is now built from the server's own list of parameters.
Requires extension **>= 1.13.0**.

### Highlights

- **Diagnostics you can act on.** A warning or an error from ProvSQL
  shows its reason tag as a badge (`provsql-reason`, and whether the
  limit is deliberate, a gap, or out of scope), and its `HINT`, which
  says what to write instead or which setting to raise, stays visible
  on its own line in the result area and the notebook cells.
- **Settings built from the server.** The configuration panel is
  generated from `pg_settings`, with a curated list in sections
  (Tracking, Limits, Evaluation, Messages, and a folded Advanced
  section). Each widget follows the parameter's type, PostgreSQL
  validates the value and its refusals appear on the status line, and
  parameters the user cannot change are greyed out. The new
  `provsql.implicit_freeze`, `provsql.max_memory`, `provsql.max_worlds`,
  and `provsql.gate_cache_size` are there. The tool settings
  (`provsql.tool_search_path`, `provsql.fallback_compiler`) move to a
  Settings section of the Tools panel.
- **Cancel and limits.** A query stopped by the cancel button, a
  statement timeout, or one of ProvSQL's resource limits is reported as
  such, with the limit's hint.
- **Probabilities are written once.** The circuit inspector shows an
  input gate's probability read-only, offering only a first write; the
  documentation explains how to give a row a new probability with
  `replace_input`.
- **Recursion and arithmetic in the circuit view.** The equation-system
  gates of a recursion over cyclic data are rendered, and every
  arithmetic operator ProvSQL records over aggregates (integer
  division, `floor`, `ceil`, `round`, `abs`, `power`…) has its glyph.
- **Subset, clearance, and consent semirings.** The evaluation strip
  offers `subset`, `clearance`, and `consent` in its User-enum group, for
  mappings whose value is an enum or an array of one; a row that some
  purposes return but the unrestricted query does not is flagged.
- **Paths in a graph.** The evaluation strip offers *Regular
  expression*, *Least cost with its paths*, and *Least cost with one
  path* (the extension's `sr_regex`, `sr_tropical_paths`, and
  `sr_best_path`), with a picker for the mapping that labels the
  tuples. Formula also renders circuits marked absorptive, and names
  the unknowns of a recursion's equations as the circuit view numbers
  them.
- **Evaluations in notebook cells.** A notebook cell evaluates random
  variables and aggregate results too, with the same distribution
  profile panel as Circuit mode, and lists the exact method and the
  guarantees first, as the evaluation strip does.
- **Surviving a restart.** Pooled connections are checked before use,
  so Studio carries on after PostgreSQL restarts; a notebook cell whose
  database session has ended is rerun on a new one, saying why; an
  unreachable database is reported at once, offering to pick another or
  edit the connection.
- **Build commit in the footer.** When Studio or ProvSQL is not on a
  release tag, the footer shows, on a line of its own, the commit each
  was built from.

### Fixes

- A saved setting for a parameter the server no longer has, or no longer
  accepts, no longer breaks every request: each override is applied on
  its own.
- Temporal mode offers plain tracked tables in its relation picker.
- An older `/api/temporal` response no longer overwrites a newer one on
  the timeline.
- Aggregate names are shown whole in circuit nodes.
- Opening a random variable or an aggregate result after a Boolean
  evaluation such as Probability selects Distribution profile in the
  evaluation strip, not Formula.
- Opening the panel that holds the tool settings no longer clears the
  evaluation result; an open popover closes when another opens.

### Notebooks

- New example notebook for case study 9 (a sales-forecast dashboard over
  deals that may not close).
- The other example notebooks are updated to the current versions of
  their case studies.

## [1.8.0] - 2026-08-07

Companion release for ProvSQL extension 1.12.0. A focused release on the
circuit inspector's eval strip: the three **planner-time evaluation
routes** the extension now reports under their own names are surfaced as
methods on exactly the circuits they apply to, and **Formula** becomes
available on every gate kind rather than only on the ones with a leaf
mapping. Requires extension **>= 1.12.0**.

### Highlights

- **The planner-time routes are named methods.** The safe-query
  (read-once) rewriter, the joint-width UCQ compiler and the
  bounded-treewidth reachability compiler each hand the probability
  dispatcher a circuit of their own making; all three used to report as
  the generic `independent`. Studio now reads the route tag off the
  circuit -- `info1` of the elided `gate_assumed` wrapper, or `info2` of
  a materialised certified d-D root -- and offers `sq-rewrite`,
  `bounded-jw` or `reachability` in the method dropdown, each shown only
  on a token whose root actually carries that tag, with a tooltip
  explaining what makes it exact and linear-time.
- **Formula on every gate kind.** Formula serialises a circuit rather
  than evaluating it, so it is the one entry that renders the
  measure-carrier gates (`gate_rv`, `gate_arith`, `gate_mixture`,
  `gate_case`) instead of refusing them, and renders a conditioning
  marker as `(target | evidence)`. It accordingly joins PROV-XML as an
  option on scalar and conditioned targets, and its provenance mapping
  becomes optional -- a measure-carrier circuit has no input leaves to
  name, and an unnamed leaf renders as an abbreviated UUID rather than
  collapsing to the semiring's one. Backed by the optional-mapping
  `sr_formula` and the all-gate formula pseudo-semiring in extension
  1.12.0.

### Housekeeping

- The ruff rule set is pinned explicitly (`E4`, `E7`, `E9`, `F`) instead
  of inheriting ruff's default, which widened in 0.16 and turned an
  unpinned `pip install ruff` in CI into a lint contract that changed
  under us.

### Playground

- `circuit-vocab.js` is shipped in the assembled doc-root, so the
  circuit-vocabulary panel works in the browser build.

## [1.7.0] - 2026-07-17

Companion release for ProvSQL extension 1.11.0. The headline feature is
**Temporal mode**, a validity-timeline view of a relation or query with
as-of / during / full-history time operations. The second axis is keeping
the circuit inspector in step with the extension's grown random-variable
and Bayesian surface: RV rendering is now **registry-driven** (a family
added to the extension renders without a Studio upgrade), latent
parameters render their sub-circuit inline, the new `observe` / `CASE`
gates and arithmetic opcodes all display, and the distribution profile
gains entropy and quantile readouts. The release also pins the two
homomorphism-only modes to the Boolean scheme, moves the bundled fixtures
onto maintained provenance mappings, polishes Notebook mode (one-click
extension install, math-rendering fix), and refreshes the Playground
landing experience. Requires extension **>= 1.11.0** (enforced at
startup); Temporal mode additionally needs PostgreSQL **14+**.

### Highlights

- **Temporal mode.** A new mode that places the rows of a relation or a
  query on a validity timeline, backed by the extension's
  `sr_temporal` interval-union kernel. A source × time-operation model
  drives it: pick a **Relation** or an arbitrary **Query** source, then an
  **as-of** (snapshot at an instant), **during** (rows valid across a
  window) or **full-history** operation. `datetime-local` pickers select
  the instant or window to the second; the **Send query** path activates
  the Query source and a carried run query flows straight into it.
- **A timeline that reads at any scale.** The axis adapts from seconds to
  years, captioning date / year rollovers inline at the ticks; the
  **During** window is framed (out-of-window time dimmed, bounds marked);
  `±infinity` validity renders as edge bars; degenerate and empty-union
  intervals get explicit markers and formatted-interval tooltips; the
  timezone is labelled UTC on both pickers and axis. An **Order** control
  sorts lanes chronologically, and lanes are labelled positionally so a
  query projecting duplicate column names still lays out correctly.
- **Boolean lock in the homomorphism modes.** Contributions and Temporal
  modes pin the provenance scheme to `Boolean`: both read the circuit
  purely as a Boolean function (Shapley/Banzhaf weights; the
  Boolean-to-temporal homomorphism), so the scheme cannot change their
  result. The scheme remains freely user-controllable in Circuit and
  Notebook modes.
- **Maintained mappings in the fixtures.** The bundled temporal fixtures
  and case studies now build their validity mappings with
  `create_provenance_mapping(..., maintained => true)`, so a deleted row
  keeps its original interval bounded at the deletion time – demonstrating
  the extension 1.11.0 feature end to end.
- **New arith-operator labels.** The circuit inspector labels the
  extension's new `gate_arith` opcodes: `^` (power), `ln`, `exp` – and a
  power gate whose exponent is the constant `0.5` displays as the square
  root it is (`√`, with `sqrt (^ 0.5)` in the operator row), matching
  the `sqrt()` SQL sugar. The `PERCENTILE` opcode backing the
  extension's `percentile_cont` aggregate over `random_variable` rows
  renders its fraction in the standard percentile notation (`p50` for
  the median, `p99`…), falling back to `pct` when the fraction is
  missing.
- **Entropy headline stat.** The distribution profile adds an `H` tile
  (entropy in nats – Shannon for a discrete root, differential for a
  continuous one, via the extension's new `entropy` readout) next to
  μ and σ. Shapes the entropy evaluator declines (composites without a
  Monte Carlo budget, aggregate roots) simply omit the tile – the rest
  of the profile is unaffected.
- **Quantile eval-strip entry.** A *Quantile* option under the
  Distribution group computes the inverse CDF at a chosen fraction
  (default `0.5`, the median) via `rv_quantile`, with the same
  conditioning support as Moment / Sample – exact for closed-form and
  categorical shapes, empirical Monte Carlo otherwise.
- **`gate_case` rendering.** The circuit inspector renders the
  `gate_case` guarded-selection gate (the `CASE`-over-`random_variable`
  lowering: abs / ReLU / clamp) with its own `⇢` glyph, a descriptive
  tooltip, and guard / value / default edge labels tracking the
  first-match wire layout.
- **`gate_observe` rendering.** The circuit inspector renders the
  `gate_observe` likelihood-weighting evidence gate (from
  `observe(leaf, datum)`, equivalently the `leaf = datum` conditioning
  form) as a binary `=` node – the observed random-variable leaf and a
  synthetic value node for the datum. This matches the equivalent
  `gate_cmp` equality, so the two spellings of the same evidence read
  alike; the true `observe` gate type still shows in the inspector.
- **Registry-driven RV rendering.** The circuit inspector now reads the
  extension's continuous-distribution family registry
  (`provsql.rv_families()`): gate glyphs and parameter symbols come from
  the registry, and the inline density preview is drawn from a
  server-computed pdf grid (`rv_analytical_curves`) attached per rv leaf
  to the circuit payload. Studio keeps no client-side family table or
  pdf math, so a family added to the extension – Gamma / Chi-squared in
  1.11.0 – renders correctly without a Studio upgrade. This makes
  extension 1.11.0 a hard floor; the startup version check now enforces
  it. A parametric (latent) leaf, whose distribution parameter is itself a
  random variable, renders that parameter's sub-circuit inline.

### Improvements

- **Notebook mode offers to install ProvSQL.** A notebook opened against a
  database without the extension shows a banner offering a one-click
  `CREATE EXTENSION provsql CASCADE`, instead of failing cell by cell, so
  a notebook is runnable on an arbitrary database.
- Markdown cells no longer carry a `[ ]` execution-count gutter (matching
  Jupyter, where the count is a code-cell affordance), and the **+ SQL**
  button documents the `y` convert-to-SQL command-mode shortcut.
- **Fix: the eval-strip condition is preset only from an event root.**
  Pinning a circuit whose root is a scalar distribution no longer
  pre-fills the conditioning field; only an event-typed root (a
  comparison) does.
- **Fix: math renders reliably in Markdown cells.** The vendored KaTeX
  auto-render chain is now awaited before math is rendered, so `$…$` /
  `$$…$$` no longer intermittently shows as raw source when the browser had
  not finished loading `katex.min.js` (a race present since math support
  was added).
- **Fix: a directly-loaded circuit survives a page reload.** A circuit
  opened by a token jump (from a notebook circuit cell or a where-mode
  result row) with no query to re-run is now restored after a reload,
  instead of the canvas coming up blank.
- **Fix: oversized circuit-node labels no longer overflow in notebook
  cells.** A node whose label carries wide payload (a distribution's
  parameters, a value scalar, an aggregate name) shrinks its font to fit
  the node, matching Circuit mode.
- The timeline clears when a query fails and surfaces the SQL error;
  a no-provenance result explains why instead of rendering blank.
- Source, query and time-operation selections are restored on page reload.
- Validity mapping is selectable in both the Relation and Query sources.

### Playground

- The default landing page is refreshed.

## [1.6.0] - 2026-06-18

Companion release for ProvSQL extension 1.10.0. The headline feature
is **Contributions mode**, a per-input Shapley / Banzhaf heat-map over
a query result; the release also renders every circuit construct the
extension 1.10.0 routes produce (absorptive class, d-D certificates,
conditioned gates, Möbius nodes), and brings a series of Playground
and notebook improvements driven by first-user feedback. Requires
extension **>= 1.10.0**.

### Highlights

- **Contributions mode.** A fifth mode alongside Circuit, Where,
  Notebook and the query views: click a result cell to pin it as the
  target and get signed per-input contribution bars (Shapley or
  Banzhaf, backed by `shapley_all_vars`), with measure / method /
  mapping controls; source rows are labelled by their column values
  in table order, and the Method control exposes the extension's
  cost-based (`auto`) vs historical (`ladder`) d-DNNF construction
  routes. Covered in the user guide and a "Seeing it in Studio" coda
  to case study 2. With five modes, the mode switcher becomes a
  dropdown showing the current mode.
- **Absorptive provenance class.** The scheme selector becomes
  four-way (`where` / `semiring` / `absorptive` / `boolean`),
  including per-cell overrides in notebooks; absorptive-assumed
  circuits get an amber **A** badge and narrow the semiring menu to
  what is sound (strictly for cyclic-recursion truncation roots,
  leniently for load-time absorptive folds); a new "Tropical
  (min-plus, nonnegative)" entry in the eval strip makes min-cost
  reachability on cyclic data evaluable in one click.
- **Certificates and the new gates, rendered.** A green **D** badge
  marks gates carrying the persisted d-D certificate (deterministic /
  decomposable tooltips and an inspector row), joining the B / A / IF
  badges; surviving `assume()` wrappers get their own glyph;
  conditioned gates display with labelled evidence / query children;
  and a `gate_mobius` renders as a μ node with each child edge
  carrying its signed integer coefficient and the transparent lineage
  child labelled `lineage`.
- **Notebooks: math and more.** Markdown cells render LaTeX math via
  vendored KaTeX (auto-render, no CDN); notebooks gain a table of
  contents, internal links and a close-all action; the Load button
  also accepts a plain `.sql` file, appended as a ready-to-run cell.
- **Playground onboarding and data loading.** The landing page gains
  a call-to-action to the interactive tutorial plus per-case-study
  notebook links, and a first visit opens the tutorial notebook
  directly; boot, database switch and Reset show a modal busy overlay
  instead of a passive status bar. The in-browser psycopg shim
  implements `cursor.copy()`, so dump-style `COPY ... FROM stdin`
  loads (notably the bundled notebooks' data cells) actually populate
  tables; `?nb=` deep links open the notebook's bound database.

### Improvements

- SQL arriving via paste, drop, `?q=` links or a loaded file is
  cleaned of invisible Unicode (non-breaking spaces, zero-width and
  bidi characters, BOM) that PostgreSQL rejects with cryptic syntax
  errors, in both the query box and notebook cells.
- TID / BID pills are shared across views, approximation guarantees
  render uniformly, and aggregate moments display exactly where the
  extension computes them exactly.
- The absorptive badge placement is harmonised with the Boolean badge
  and the certificate frontier tag.

### Bug fixes

- Playground: the application `search_path` puts `public` first, so
  unqualified `CREATE TABLE` (notably the notebooks' setup cells) no
  longer lands in the `provsql` schema, which left the Schema panel
  empty after Run all.
- Playground: the case-study-8 deep link works (Markdown/KaTeX assets
  resolve under a sub-path, and the database is pre-created).
- The mode menu no longer renders white-on-white items.

### Demo and tests

- End-to-end suites walk case studies 2 (Contributions), 7 and 8
  through the live UI; the per-cell scheme cycle covers absorptive;
  navigation timeouts are sized for loaded CI runners.

## [1.5.0] - 2026-06-05

Companion release for ProvSQL extension 1.9.0. Two headline features:
**Notebook mode**, a Jupyter-style notebook over a ProvSQL database,
and the **ProvSQL Playground**, the full Studio running entirely in
the browser over PostgreSQL+ProvSQL compiled to WebAssembly. The
evaluation strip also gains the extension's guarantee-first
probability surface. Requires extension **>= 1.9.0**.

### Highlights

- **Notebook mode.** A third mode alongside Circuit and Where: an
  ordered list of cells (SQL, Markdown, circuit snapshots, semiring /
  probability evaluations) executed against a pinned kernel session
  with Jupyter state semantics: temp tables and `SET`s persist across
  cells, each cell runs in its own transaction, restart gives a clean
  slate. Includes the Jupyter two-mode keymap (Esc / Enter, a/b, dd/z,
  m/y, Shift/Ctrl/Alt+Enter), Markdown rendering with `sql`-fenced
  highlighting (vendored marked + DOMPurify), per-cell
  provenance-scheme overrides, an outline + compact-relations sidebar,
  and round-trips to Circuit mode that restore selection and scroll.
- **Tabs as database bindings.** Each notebook tab is bound to the
  database it was authored against (stamped in the `.ipynb`
  metadata); a binding mismatch shows a banner offering switch /
  create / rebind, never switching silently. `POST /api/databases`
  backs the create action, and a nav-bar action empties the connected
  database (drop user schemas, reinstall provsql).
- **`.ipynb` save / load.** Notebooks are nbformat-v4 files: outputs
  carry standard MIME fallbacks (HTML tables, self-sufficient SVG
  circuit snapshots, plain-text eval results) so GitHub and nbviewer
  render them directly, while Studio re-renders from richer
  `application/vnd.provsql.*` payloads. Tabs autosave to browser
  storage between explicit saves.
- **Bundled example notebooks.** The user-guide tutorial and case
  studies ship as self-establishing notebooks (idempotent setup cells:
  create a fresh database, Run all, get the whole study), generated
  from the documentation sources by `studio/scripts/rst2nb.py` with a
  CI drift check, opened from the `Open example...` menu or a
  `/notebook?nb=<name>` deep link.
- **ProvSQL Playground.** The unmodified Studio Python (Flask +
  psycopg + sqlparse) runs client-side in Pyodide over PGlite
  (PostgreSQL+ProvSQL in WebAssembly), published as static files at
  [provsql.org/playground](https://provsql.org/playground/): no
  install, fully self-hosted (no CDN at run time), with the tutorial
  and case-study databases pre-seeded, a Reset button, shareable
  `?mode=&db=&q=&nb=` deep links, and notebook mode included (kernel
  restart maps to `DISCARD ALL` on the single shared session). A
  shell-page/iframe split keeps the WASM backend warm across mode and
  database switches. Browser support is gated on WebAssembly JSPI.
- **Guarantee-first probability evaluation.** The eval strip leads
  with the extension 1.9.0 exact / relative / additive guarantee
  paths, adds the `karp-luby` and `d-tree` methods (with an
  (eps,delta) vs samples toggle for the approximate methods), renders
  the returned guarantee as a bound or value interval with the actual
  sample count, surfaces the resolved method (down to
  `compilation:d4` / `wmc:ganak`), and includes karp-luby in the
  probability benchmark.

### Improvements

- Tool-backed compiler / counter pickers and the fallback-compiler
  config row hide tools that are not available on the backend instead
  of listing dead options; the probability-method dropdown has hover
  tips for every entry.
- A folder icon under the eraser loads a `.sql` file into the query
  box; dump-style `COPY ... FROM stdin` blocks run through the COPY
  sub-protocol, so `pg_dump` outputs paste and run as-is.
- Numeric `value` / `mulinput` gate labels compact to 4 significant
  figures; tuple counts pluralize (`1 tuple`).
- Result cells holding `agg_token` arithmetic results display
  `value (*)` like plain aggregate tokens (requires extension 1.9.0).

## [1.4.0] - 2026-05-27

Companion release for ProvSQL extension 1.8.0. Brings the extension's
two new surfaces into Studio: the **external-tool registry** (a panel
to manage compilers / counters / KCMCP servers, with every tool picker
now driven by the registry) and **inversion-free** d-DNNF compilation
(a certificate badge, per-input order keys, and a new compilation /
benchmark target). Requires extension **>= 1.8.0**.

### Highlights

- **Tools panel.** A new panel manages the extension's external-tool
  registry (`provsql.tools`): tools are grouped by operation, and can be
  enabled / disabled, re-prioritised, edited, registered and
  unregistered inline, covering both CLI tools (executable + command
  template) and KCMCP servers (endpoint or managed). Nav popovers are
  mutually exclusive, and a registry change refreshes the Evaluate
  strip's compiler dropdowns immediately.
- **Registry-driven tool selection.** The compiler / counter pickers and
  the probability benchmark are now populated from `provsql.tools`
  rather than hardcoded lists, and only tools that currently resolve on
  the backend are selectable.
- **Inversion-free integration.** Studio renders the inversion-free
  `gate_annotation` as an **IF** badge on the certificate root
  (coexisting with the safe-query **B** badge); the inspector shows the
  certificate order and each input's per-input order key and rank.
  When the root carries a certificate, inversion-free is offered both as
  a d-DNNF compilation target in the compiler picker and as a method in
  the probability benchmark (routed through the elided root's original
  token).
- **Compiled d-DNNF timing.** Compiling a circuit to a d-DNNF (`kc-ddnnf`)
  runs under the session's `statement_timeout`, and the canvas subtitle
  reports the server-side compile time.

## [1.3.0] - 2026-05-24

Companion release for ProvSQL extension 1.7.0. Brings the extension's
new **knowledge-compilation surface** into Studio: a dedicated
inspector for compiled d-DNNFs, CNF, and tree decompositions, a
multi-compiler / multi-counter probability benchmark, and
tool-availability awareness throughout the UI. Requires extension
**>= 1.7.0**.

### Highlights

- **Knowledge-compilation inspector.** New `/api/kc/*` endpoints back
  a KC strip folded into the Evaluate strip: compile a circuit's
  Boolean provenance with any available compiler and inspect the
  result. Compiled d-DNNF circuits render directly in the circuit
  canvas (root at the top), with a result modal and a d-DNNF
  structural-statistics summary (the `ddnnf_stats` size column /
  canvas size summary). A `kc-nnf` panel shows the `.nnf` text export,
  and the CNF view is annotated with the `tseytin_cnf`
  variable-to-input mapping. The compiled circuit is labelled with its
  KC class, and the KC compilers are credited in the bibliography.
- **Tree-decomposition view.** Bags are coloured, clickable to
  inspect, and resolve back to their source rows; pinning a TD bag or
  a d-DNNF internal node hides the (inapplicable) evaluation strip.
- **Multi-backend probability benchmark.** The benchmark table runs
  every method the extension exposes, gains a compiled-size column,
  enforces a per-method `statement_timeout`, and runs under the
  session's `provsql.boolean_provenance` so its numbers match the
  marginal-probability path.
- **Tool-availability awareness.** Compiler / counter dropdowns and
  the benchmark are filtered to tools actually resolvable on the
  backend; `/api/kc/{ddnnf,td}` return `501 Not Implemented` (rather
  than `500`) when the required tool is absent, and missing-tool
  shortcuts surface as notices. WMC backends move to their own
  option group.
- **`provsql.fallback_compiler` in the Config panel.** A new dropdown
  with validation exposes the extension's fallback-compiler GUC.
- **Aggregated random variables.** The evaluation strip offers the
  distribution profile / moment / sample surface on `agg` and
  `semimod` gate targets, and hides the eval-strip options that do not
  apply to KC / agg / semimod targets.
- **UI polish.** Even 50/50 shell split so the dense left column
  matches the query / result pane; schema-panel key columns underlined
  (solid for primary keys, dotted for `repair_key` / BID grouping
  keys); circuit-canvas maximum zoom raised from 2.5 to 8 for dense
  circuits; result-table column sorting; trimmed Monte-Carlo seed input
  width.

## [1.2.0] - 2026-05-17

Companion release for ProvSQL extension 1.6.0. Surfaces the new
**safe-query rewriter** (Boolean-provenance mode) and the new
**TID / BID / OPAQUE classifier** through dedicated UI affordances:
a third selector value on the provenance-scheme widget, a
session-sticky toggle, per-result prov pills on the result-table
header, and per-relation prov sub-pills on the schema panel.

### Highlights

- **Boolean-provenance scheme integration**. A third value
  (`Boolean`) joins the existing provenance-scheme selector
  (Semiring / Where). When selected, Studio sets
  `provsql.boolean_provenance = on` on the backend for the
  duration of the session and the extension's safe-query
  rewriter fires on every accepted hierarchical-CQ query. The
  toggle is session-sticky: Studio retains the selected scheme
  across query executions, schema-panel clicks, and circuit-mode
  navigation, matching the existing Default / Where behaviour.
- **Result-pane classifier pill**. The `provsql` column header of
  every query result carries a per-query pill reflecting the classifier's
  verdict on the just-executed query, labelled `prov-tid` /
  `prov-bid` (TID and BID share the brand pill: the distinction matters
  for probabilistic evaluation but does not warrant a visual demotion) or
  bare `prov` with a muted tone for OPAQUE. The pill source is the
  `provsql.classify_top_level` NOTICE Studio captures from each query
  execution; the tooltip explains the kind in prose.
- **Schema-panel prov pills**. The per-relation PROV pill on each
  schema-panel row now carries a kind-aware label (`prov-tid` /
  `prov-bid` for tracked relations whose kind the classifier has
  certified, bare `prov` with a muted tone for OPAQUE, bare
  `prov` with the brand pill for relations on an older extension
  that does not surface a `prov_kind`). The kind is populated at
  schema-load time by probing the classifier on a
  `SELECT * FROM <relation>` query inside a SAVEPOINT-wrapped
  transaction; view bodies are probed identically (the classifier
  descends through `RTE_SUBQUERY`).
- **Schema-panel row-click**. Clicking a relation row in the
  schema panel now inserts `SELECT * FROM <rel>` into the
  query box (replacing the previous behaviour of typing the
  relation name without a SELECT). Bare-name click insertion is
  schema-qualified when the bare name does not resolve in the
  current search_path, so `public.foo` and `myschema.foo` both
  work after a single click.
- **Schema-panel column-count tooltip fix**. Views now report
  the actual number of columns of the underlying query in the
  hover tooltip, not the count of `pg_attribute` rows (which
  undercounted views with computed columns).

### Internal

- **Catalog scan filter**: the tracked-input enumeration now
  filters temp / toast schemas, so the schema panel only lists
  user-visible tracked relations.
- **Psycopg pool teardown**: the e2e test fixture closes the
  psycopg connection pool cleanly on Playwright shutdown,
  avoiding a "FATAL: terminating connection due to
  administrator command" trail in the test output.

### Compatibility

Minimum required extension version: **1.6.0**. See the
[compatibility matrix](https://provsql.org/docs/user/studio.html#compatibility)
in the user guide for the full table.

The Boolean-provenance mode selector and the TID / BID schema /
result pills are inactive on extensions older than 1.6.0: the
provenance-scheme selector falls back to the Default / Where
two-way picker, and the result/schema pills omit the per-relation
classifier sub-pill.

## [1.1.1]

Patch release fixing a Circuit-mode crash on databases that contain
temp tables.

### Bug fixes

- **Crash on `/api/circuit` when another session has a temp table.**
  The tracked-input catalog scan was matching every relation with a
  `provsql uuid` column, including `pg_temp_NN.<name>` rows that
  transiently appear in another backend's `pg_class` snapshot before
  autovacuum cleans them up; the subsequent UNION ALL then dispatched
  against an unreachable temp relation and raised `UndefinedTable`.
  The catalog scan now filters `relpersistence <> 't'` and skips
  `pg_temp%` / `pg_toast%` namespaces, so Circuit mode is robust
  against any temp-table activity on the target database.

## [1.1.0]

Companion release for ProvSQL extension 1.5.0. Adds first-class
rendering and inspection for the continuous-distribution gate
family, conditional-inference workflow, simplified-circuit
rendering, and a new evaluator strip group for distribution
profile / sample / moment / support.

### Highlights

- **New gate-type renderers**: `gate_rv` leaves with
  distribution-kind glyphs (*N* / *U* / *E* / *Γ*); `gate_arith`
  with the operator glyph drawn from `info1`; `gate_mixture`
  with three labelled edges (`p` / `x` / `y` for Bernoulli
  mixtures, `key` plus one mulinput edge per outcome for
  categorical blocks) and the Bernoulli probability rendered
  inline in the parent circle.
- **Distribution profile evaluator**: new entry under the
  *Distribution* group of the eval strip. Header stats
  (`μ`, `σ²`), inline-SVG histogram, PDF/CDF toggle, per-bar
  tooltip with `σ`-markers, wheel zoom. Backed server-side by
  the new C entry point `rv_histogram`. When the gate has a
  closed-form, the panel also draws the analytical PDF (or CDF)
  on top of the histogram as a terracotta SVG path; Bernoulli
  mixtures, categoricals, and Diracs render as discs on stems
  in PDF mode and as a staircase in CDF mode. Universally-
  infeasible truncations short-circuit with an inline message
  instead of returning empty bars. Backed by the new C entry
  point `rv_analytical_curves`.
- **Sample evaluator**: second *Distribution* entry, drawing
  conditional samples via `rv_sample(token, n, prov)`. Renders
  as a `<details>` panel with a six-value inline preview and a
  "show full list" expander; when the MC acceptance rate
  truncates the run below the requested `n`, surfaces an
  actionable hint pointing at `provsql.rv_mc_samples`.
- **Moment evaluator**: third *Distribution* entry exposing
  `moment` / `central_moment` with `k` and raw/central
  selectors.
- **Support evaluator**: fourth *Distribution* entry showing the
  closed-form `support` interval (or its conservative
  all-real fallback).
- **Conditioning**: `Condition on` text input on the eval strip
  with row-provenance auto-preset (clicking a result-table cell
  stamps the row's provenance UUID). Toggleable
  `Conditioned by:` badge underneath; clicking the active badge
  clears the conditioning and reverts to the unconditional
  answer, clicking the muted badge restores the row prov.
  Manual edits stick within a row and reset on row navigation.
- **Simplified-circuit rendering**: Circuit mode honours the new
  `provsql.simplify_on_load` extension GUC and renders the
  in-memory simplified graph via the
  `simplified_circuit_subgraph` SRF; the Config-panel toggle
  switches between the raw, gate-creation view and the
  simplified, evaluation-time view.
- **Config-panel rows**: new sliders/toggles for
  `monte_carlo_seed`, `rv_mc_samples`, and `simplify_on_load`,
  matching the existing Provenance-section pattern.
- **Anonymous-input rendering**: anonymous `gate_input`
  probabilities render as a percentage in the node circle;
  tracked-table input gates continue to render `ι` per the
  existing iota convention.
- **Categorical and mixture inspector**: node-inspector entries
  for the two new gate shapes; single-outcome categoricals
  collapse to `as_random` in the simplifier.
- **Footer version readout**: the footer now displays the
  loaded ProvSQL extension version (read from `pg_extension`)
  and the Studio package version (`provsql_studio.__version__`)
  on the right edge, served by `/api/conn`. A new
  `provsql-studio --version` CLI flag prints the package
  version and exits.
- **Schema-panel column pills**: columns whose type is one of
  ProvSQL's circuit-bearing types now carry a terracotta
  column-level pill (`rv` for `random_variable`, `agg` for
  `agg_token`) next to the column name, mirroring the
  relation-level `prov` / `mapping` pills. The
  `create_provenance_mapping` click affordance is suppressed
  on these columns (their values are circuit references, not
  scalar tags).
- **Result-table column-type indicators**: each result-table
  `<th>` carries the column's SQL type name as a tooltip and,
  for ProvSQL-significant columns, the same `rv` / `agg`
  pills as the schema panel plus a purple `prov` pill on the
  row-provenance `provsql` column itself, so the affordance
  follows the data into the result without a round-trip to
  the schema panel.

### Demo and tests

- `studio/scripts/demo_continuous.{sql,py}` fixture loader for
  the sensors / air-quality narrative used in
  `doc/source/user/casestudy6.rst`; a standalone copy is also
  shipped as `doc/casestudy6/setup.sql` for the rendered docs.
- Playwright e2e at `studio/tests/e2e/test_continuous.py`
  covering `gate_rv` / `gate_arith` rendering, the
  *Distribution profile* evaluator (including the closed-form
  PDF/CDF overlay over the histogram bars), the *Sample* panel,
  the conditioning auto-preset, and a Monte-Carlo
  `p ∈ (0, 1)` smoke test on the sensors fixture.

### Compatibility

Minimum required extension version: **1.5.0**. See the
[compatibility matrix](https://provsql.org/docs/user/studio.html#compatibility)
in the user guide for the full table.

## [1.0.0]

First public release. Requires the ProvSQL extension at version 1.4.0
or later.

### Highlights

- **Circuit mode** renders the provenance DAG behind any result UUID or
  `agg_token` cell. Drag-to-reposition, wheel-to-zoom, frontier
  expansion via gold-`+` badges, plus fullscreen, fit-to-screen and
  reset-positions toolbar buttons. Pinning a node opens an inspector
  with gate-specific metadata; `input` and `update` gates expose the
  stored probability as click-to-edit (sends `set_prob` server-side).

- **Semiring evaluation strip** runs any compiled semiring against the
  pinned node (or the root by default). Optgroups for Boolean, Lineage
  (`formula` / `how` / `why` / `which`), Numeric (`counting` /
  `tropical` / `viterbi` / `lukasiewicz`), Intervals (`interval-union`),
  User-enum (`minmax` / `maxmin`), plus probability methods, PROV-XML
  export, and any user-defined wrappers over `provenance_evaluate`. The
  mapping picker filters by the selected semiring's expected value
  type.

- **Where mode** turns on `provsql.where_provenance` and wraps every
  `SELECT` so each output cell becomes hover-aware: the source rows
  that contributed to it light up in the per-relation sidebar. Each
  result row carries a `→ Circuit` button to switch modes with the
  same provenance UUID.

- **Schema panel** lists every selectable relation with `PROV` /
  `MAPPING` pills. Click-to-prefill `add_provenance` /
  `remove_provenance` / `create_provenance_mapping` calls into the
  query box.

- **Config panel** with Provenance, Session and Display-Limits
  sections. Values persist to `$XDG_CONFIG_HOME/provsql-studio/`
  (Linux), `~/Library/Application Support/provsql-studio/` (macOS) or
  `%APPDATA%\provsql-studio\` (Windows); same options exposed as CLI
  flags.

- **In-page connection layer**: DSN editor probes new endpoints with
  `SELECT 1` before swapping pools; database switcher; the per-batch
  `search_path` always pins `provsql` at the end (lock chip in the
  header).

- **Query history**: in-session, with `Alt+↑` / `Alt+↓` to step in
  place and a `History` listbox. Mode switching carries the current
  SQL forward via `sessionStorage`; auto-replay only after an explicit
  run.