Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 14 additions & 2 deletions Sources/AngouriMath/Docs/Contributing/EMatching.md
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,15 @@ using the reversibility work to avoid firing a rule right next to its own invers
whichever matcher is underneath — not a consequence of this document. This document is scoped to what
e-matching actually buys: not materialising a term to find out whether a rule's *shape* matches.

> **Answered since, by measurement rather than by a policy** (#1193, #1194). The one runaway at the
> widest ceiling is a single inverse pair split across `ExpandMultipleAngle` and `Trigonometric`,
> bounded to the trigonometric sets; at `RulesUpTo(Rearranges)` every member of that family
> saturates, because `ExpandMultipleAngle` declares `Expands`. **The growth ceiling is the
> scheduling policy**: it withholds the expanding direction of every pair whose two directions are
> declared, and can only protect a pair whose directions *are* declared — which is what declaring a
> growth for a code-built rule is for (#1195–#1197). `Expands` rules do not join `SafeRules`, and no
> second mechanism is planned. See `InversePairTable.md` and `RunawayBreadthTest`.

## Where the patterns actually are

`RewriteRules.All` — the public registry `Transformation.EqualitySaturation` drew `SafeRules` from
Expand Down Expand Up @@ -168,8 +177,11 @@ bound are extracted — narrower than falling back on the whole rule.

## What is deliberately not decided here

**Whether `Growth.Expands` rules ever join `SafeRules`**, and under what scheduling policy — a
separate document's question, once one exists to answer it.
**Whether `Growth.Expands` rules ever join `SafeRules`** — decided since, and by measurement: they
do not, and the growth ceiling is the scheduling policy (see the note under *What this is not
trying to fix*, above). What the graph does need is a rational folded on insertion (#1198) and an
extraction that is a fixed point from the leaves up rather than a recursive walk (#1202), both of
which were found by running the safe ceiling over the corpus rather than over sixteen expressions.

**Whether the local fallback (extract-then-`TryApply`, kept for `Gathered`-containing rules) is worth
removing later** by also solving e-matching for `GatheredPattern`, or stays permanently as the honest
Expand Down
6 changes: 5 additions & 1 deletion Sources/AngouriMath/Docs/Contributing/Transformations.md
Original file line number Diff line number Diff line change
Expand Up @@ -244,7 +244,11 @@ So what is left, in dependency order:
consumer #746 tier 2 names is the inverse-pair table equality saturation needs, since a saturation
engine keeps both results where this pipeline keeps one and so has to be told which rewrites undo
each other. `RewriteRuleGrowth` does not answer that: it says which way a rule moves, not which
rule is the inverse of which.
rule is the inverse of which — though for a pair whose two directions are *declared*, the growth
ceiling already withholds the expanding one (#1194), which is as much of the table as saturation
has needed so far. The graph itself has its first production caller: the perfect-square collapse
asks `Saturation.ProvesEqual` before it asks `Simplify` (#1201), measured as a caller that meets
the standing performance condition rather than as a speed-up.
- **Rendering a step as a sentence** — #746's v5.0. Everything a sentence needs is now on a step
except one thing: *why the rewrite is allowed*. `Soundness` is declared, not checked, and a rule
that holds only under an assumption does not say which. That is the same gap
Expand Down
Loading