From 7caedd59f156c2defc3d1d2c097be2a7be079403 Mon Sep 17 00:00:00 2001
From: Diogo Ribeiro <diogo_dj@hotmail.com>
Date: Tue, 11 Aug 2026 12:30:52 +0100
Subject: [PATCH] docs: refresh roadmap

Records 0.1 and 0.2 as released with their DOIs and what actually shipped,
rather than as intentions. The performance section is done, so it becomes
an unreleased entry.

Reorders what is next: simulation before visualisation, since a plot of an
unvalidated result is a confident-looking wrong answer.
---
 docs/ROADMAP.md | 79 +++++++++++++++++++++++++++++++++++--------------
 1 file changed, 57 insertions(+), 22 deletions(-)

diff --git a/docs/ROADMAP.md b/docs/ROADMAP.md
index d4be7a7..95e5b4a 100644
--- a/docs/ROADMAP.md
+++ b/docs/ROADMAP.md
@@ -1,43 +1,78 @@
 # Roadmap
 
-## 0.1 — foundation
+Released versions record what shipped. Everything below the line is intent, not
+commitment, and the ordering matters more than the version numbers attached to
+it.
+
+## 0.1 — foundation (released 2026-08-10)
+
+[10.5281/zenodo.21879360](https://doi.org/10.5281/zenodo.21879360)
 
 - direct fuzzy calibration, logistic and piecewise
 - crisp calibration
 - typed fuzzy-set expressions
 - necessity/sufficiency parameters of fit
 - complete binary truth tables
-- exact classical QMC
+- exact classical Quine-McCluskey
 - conservative and parsimonious csQCA/fsQCA solutions
-- directional intermediate solutions
 - pandas-native result objects
 - parity harness against R `QCA`
 
-## 0.2 — parity and robustness
+## 0.2 — parity and robustness (released 2026-08-11)
+
+[10.5281/zenodo.21887472](https://doi.org/10.5281/zenodo.21887472)
 
-- standard intermediate-solution simplifying-assumption algorithm
-- enhanced necessity analysis and supersets/subsets
-- contradictory simplifying assumptions
+- correct intermediate solutions, with easy/difficult counterfactual
+  classification matching R
+- multi-value QCA
+- enhanced necessity analysis: supersets, trivialness, relevance of necessity,
+  SUIN conditions
 - solution-specific unique coverage
+- robustness sweeps over cutoffs and case removal
+- Schneider–Rohlfing case typology
 - calibration diagnostics
-- XY plots
-- extend the R-QCA golden parity suite (calibration, truth tables, fit measures
-  and conservative/parsimonious solutions are already covered as of 0.1;
-  remaining: intermediate solutions, necessity supersets, multi-outcome models)
-- optional R-compatible calibration snapping, so extreme memberships can be
-  reported exactly as R does when replicating an existing analysis
+- prime-implicant chart inspection, so a solution can explain itself
+- R parity extended to intermediate solutions, necessity screens, per-term fits
+  and multi-value models, with the two known divergences pinned by tests
+
+## Unreleased
+
+- exact minimisation roughly two orders of magnitude faster, via a bitmask cube
+  representation
+- phase-level benchmark harness
+- a complexity warning raised before the exponential search, not after
+
+---
+
+## Next
+
+**A second minimisation engine.** A CCubes/eQMC-style backend, selected
+explicitly rather than substituted silently, so a fast approximate answer is
+never mistaken for the exact one the current engine guarantees.
+
+**Cross-language validation as a running check.** The R fixtures are committed
+golden values today; the generator should run on a schedule so divergences
+surface when the reference implementation moves, rather than when someone next
+looks.
+
+**Simulation.** Generating data with a known causal structure is what makes it
+possible to ask whether the method recovers what is actually there — coverage of
+the true solution, behaviour under noise and limited diversity.
 
-## 0.3 — performance
+**Visualisation.** XY plots, truth-table and chart rendering. Deliberately after
+simulation: a plot of an unvalidated result is a confident-looking wrong answer.
 
-- faster bitset/cube minimiser
-- prime-implicant consistency filters
-- row dominance
-- optional Rust acceleration
+**Provenance.** Recording the calibration anchors, cutoffs and version that
+produced a result, so a published analysis can be reproduced from the artefact
+rather than from a description of it.
 
 ## 1.0
 
-- tQCA
-- stable public API
+- a public API settled deliberately rather than by accretion, and then frozen
+- temporal QCA (tQCA)
+- multi-outcome models
+- optional R-compatible calibration snapping, for replicating an existing
+  analysis exactly
 - benchmark corpus
-- exhaustive cross-software validation
-- published algorithm and software paper
+- documentation rebuilt around tasks rather than modules
+- published software paper