nsaspy/symbolic-memory

Fork 0

RAGE 002: durable symbolic projections and symbolic-first recall #3

Open

nsaspy wants to merge 23 commits from rage/002-symbolic-projection-recall into main

nsaspy commented

2026-09-04 22:26:34 +00:00

Owner

What this slice does

Adds the first durable symbolic projection + recall vertical slice on top of the existing exact-source memory boundary.

preserve exact source text unchanged
persist structured symbolic projections separately from source text
keep predicates/arguments as non-executable data
add stable projection IDs
canonicalize native Prolog atom / MCP string arguments into one symbolic representation
add memory_recall/4 with exact predicate matching and positional JSON null wildcards
require read authority before recall candidate generation
enforce namespace visibility before recall
rank session → project → global matches
render compact Markdown-like recall packets with visible trust class
automatically attach exact source prose for lossy / context_required projections
allow explicit exact-source expansion with include_source
keep source storage successful when projection parsing fails (projection_error)
keep external_untrusted / unknown sources as evidence-only rather than semantic projections (blocked_untrusted)
expose projections + memory_recall through MCP
migrate v1 snapshots forward to v2 without losing existing memories

Safety / authority boundary

A supplied projection is structured data, never an arbitrary Prolog goal. Projection failure cannot roll back the exact source memory. Low-trust external evidence can be retained for audit/history without being promoted into semantic recall.

Example

A source memory can carry a projection like:

{
  "predicate": "prefers",
  "arguments": ["user", "prolog", "constraint_solving"],
  "statement": "User prefers Prolog for constraint solving.",
  "quality": "exact"
}

and be recalled with:

{
  "predicate": "prefers",
  "arguments": ["user", null, "constraint_solving"]
}

No model-provided predicate is executed as a Prolog goal.

Tests authored

Adds PlUnit coverage for projection round-trip, canonical argument representation, JSON-null wildcard matching, non-destructive projection failure, untrusted evidence-only admission, read authority, source fallback/expansion, scope ordering, project isolation, restart durability, v1 snapshot migration, and MCP recall output.

Verification state

Draft intentionally. The ChatGPT implementation environment does not have swipl or nix, so executable verification has not been run here.

Before marking ready:

nix flake check

or at minimum:

nix develop --command swipl -q -s test/run_tests.pl

Deferred

Automatic prose → symbolic extraction, natural-language search/recall, contradiction/supersession, temporal truth, generated rules, embeddings, and natural-language forgetting remain later slices.

## What this slice does Adds the first durable symbolic projection + recall vertical slice on top of the existing exact-source memory boundary. - preserve exact source text unchanged - persist structured symbolic projections separately from source text - keep predicates/arguments as non-executable data - add stable projection IDs - canonicalize native Prolog atom / MCP string arguments into one symbolic representation - add `memory_recall/4` with exact predicate matching and positional JSON `null` wildcards - require read authority before recall candidate generation - enforce namespace visibility before recall - rank session → project → global matches - render compact Markdown-like recall packets with visible trust class - automatically attach exact source prose for `lossy` / `context_required` projections - allow explicit exact-source expansion with `include_source` - keep source storage successful when projection parsing fails (`projection_error`) - keep `external_untrusted` / `unknown` sources as evidence-only rather than semantic projections (`blocked_untrusted`) - expose projections + `memory_recall` through MCP - migrate v1 snapshots forward to v2 without losing existing memories ## Safety / authority boundary A supplied projection is structured data, never an arbitrary Prolog goal. Projection failure cannot roll back the exact source memory. Low-trust external evidence can be retained for audit/history without being promoted into semantic recall. ## Example A source memory can carry a projection like: ```json { "predicate": "prefers", "arguments": ["user", "prolog", "constraint_solving"], "statement": "User prefers Prolog for constraint solving.", "quality": "exact" } ``` and be recalled with: ```json { "predicate": "prefers", "arguments": ["user", null, "constraint_solving"] } ``` No model-provided predicate is executed as a Prolog goal. ## Tests authored Adds PlUnit coverage for projection round-trip, canonical argument representation, JSON-null wildcard matching, non-destructive projection failure, untrusted evidence-only admission, read authority, source fallback/expansion, scope ordering, project isolation, restart durability, v1 snapshot migration, and MCP recall output. ## Verification state **Draft intentionally.** The ChatGPT implementation environment does not have `swipl` or `nix`, so executable verification has not been run here. Before marking ready: ```sh nix flake check ``` or at minimum: ```sh nix develop --command swipl -q -s test/run_tests.pl ``` ## Deferred Automatic prose → symbolic extraction, natural-language search/recall, contradiction/supersession, temporal truth, generated rules, embeddings, and natural-language forgetting remain later slices.

nsaspy added 23 commits

2026-09-04 22:26:34 +00:00

Add structured symbolic projection model 11109cdb53

Persist symbolic projections with v1 migration 94c46d04fd

Add symbolic-first projection recall 0d164e6893

Expose symbolic projections and recall over MCP dd80737744

Add symbolic projection and recall tests 3a5abfdb39

Run symbolic recall tests f47c10847a

Make projection quality validation deterministic 776fec53a1

Document symbolic projection recall slice 7d027a099f

Record RAGE 002 symbolic recall status 0009a20b1e

Handle json_read_dict null wildcards correctly a05b128c70

Test JSON null wildcard semantics 814f4d359f

Gate semantic projections by source trust aec69c5c5a

Keep source durable when projection admission fails e93d13a8f1

Verify source preservation and trust-gated projections f4363e6cb6

Expose projection creation metadata 0dc587a51e

Include projection creation time in memory_get d81e84501d

Canonicalize symbolic projection arguments 39915484a1

Cover canonical arguments and recall precedence a2375a586c

Document trust-gated symbolic recall semantics efc58467c5

Record projection trust and source-preservation invariants 7fc9fa7fa9

Remove unused projection storage helper 942a168157

Keep projection predicates as non-interned data acdf4a3da5

Store projection predicate names as strings d4a0deecf6

This pull request can be merged automatically.

You are not authorized to merge this pull request.

View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.

git fetch -u origin rage/002-symbolic-projection-recall:rage/002-symbolic-projection-recall

git switch rage/002-symbolic-projection-recall

Merge

Merge the changes and update on Forgejo.

Warning: The "Autodetect manual merge" setting is not enabled for this repository, you will have to mark this pull request as manually merged afterwards.

git switch main

git merge --no-ff rage/002-symbolic-projection-recall

git switch rage/002-symbolic-projection-recall

git rebase main

git switch main

git merge --ff-only rage/002-symbolic-projection-recall

git switch rage/002-symbolic-projection-recall

git rebase main

git switch main

git merge --no-ff rage/002-symbolic-projection-recall

git switch main

git merge --squash rage/002-symbolic-projection-recall