Commit Graph
19 Commits
Author SHA1 Message Date
4d7fccbce6 docs(experiment): measure bounded native definition projections
Preserve a diagnostic-only patch and reproducible lossless encoding/projection probe. The real VM graph exceeds several public capacities; selected SSA neighborhoods are much smaller, with explicit frontiers. Record actual resource costs and keep public engine/API limits unchanged.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 10:30:19 +00:00
af53e0fbe5 fix(experiment): retain ownership after tagged cleanup failures
Check the native cleanup Result before clearing the lease and preserve cleanup resources through the MCP error adapter. Reproduce the old false success, then verify real SDK close/replacement ownership and both MCP retry paths with explicit fault injection.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 10:25:17 +00:00
f08c191ed3 docs(experiment): compare compact call counts in real model tasks
Retain the successful answer and source checks from both adapters and all measured task costs. Both avoid call-set paging; direct beats Code Mode in the new sample and before/after effects remain mixed.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 10:07:36 +00:00
da6f5470be perf(experiment): summarize retained calls before paging
Expose a compact call distribution in snapshot metadata through the shared validated owner. Preserve coverage, unknown targets, dead rows, immutable selected reads, and empty versus unavailable semantics. Retain the same boundary question and oracle for paired real-model comparison.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 10:00:04 +00:00
fb9adbf384 docs(experiment): record real-model call boundary results
Both lanes satisfy the pre-pinned SQLite task with one retained VM analysis. Record full answer/source review, actual model and resource counters, and mixed context costs without a general efficiency claim.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 09:53:53 +00:00
f9f9dc209d feat(experiment): retain call facts beyond full graph capacity
Separate native call and boundary-value ownership from the bounded recovered graph. Validate facets independently and preserve shared identities, source evidence, and explicit partial coverage. Pin the SQLite boundary task and oracle before paired real-model trials.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 09:46:42 +00:00
04c081fc61 experiment(re): verify model reference aggregation against machine calls
Record both completed model workflows, whole-task costs and source/identity checks. Compare every VM direct-call pair with independent disassembly, preserve indirect-call limits, verify unpatched stock through the shared CLI/API, and retain failed/partial research costs.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 09:01:54 +00:00
3aa685db39 feat(experiment): compose shallow outgoing references with explicit scope
Share one native parser for both directions, preserve occurrence identity and unknown function applicability, push kind filters to Kuna and reuse sources across fields/pages/scopes. Verify real SQLite machine references independently, including calls omitted from optimized IR and VM queries beyond graph capacity. Pin the aggregate model task before its paired runs.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 08:48:24 +00:00
33c4282b8a feat(experiment): expose structured native call and argument relations
Keep recovered callee identities separate from fspec annotations and SSA argument values separate from passing storage. Add selected calls and function-input reads through the shared fact owner, with real SQLite MCP coverage and a pre-pinned cross-function model task.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 07:58:01 +00:00
f34425dc24 fix(experiment): enforce Code Mode payload budgets on guest errors
Reject oversized error delivery with explicit omission metadata, account for JSON escaping, and retain source observations and their lookup after failure. Reproduce and verify the real MCP path without claiming model-token savings from byte counts.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 07:33:09 +00:00
73db4fad54 feat(experiment): compose bounded native byte windows with RE facts
Add shared SDK/CLI/MCP byte reads over the existing Kuna data protocol, preserve partial batches and exact u64 ranges, and retain complete source Evidence. Reuse named read lifecycles, bound aggregate hex expansion, and verify stock/native/direct MCP paths plus a shipped-file byte oracle before pinning a table-dispatch model task.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 07:17:18 +00:00
1346e97dca fix(experiment): preserve code when encoded native graphs exceed capacity
Bound compact native JSON with the existing writer, keep a valid code facet when the graph cannot fit, and pass the supervised byte budget explicitly. Verify real SQLite Parser/Pragma captures, exact preserved C/mappings/graph facts and public MCP reads. Keep string inventory parsing/projection under its own owner.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 06:50:20 +00:00
dc022bcd0f experiment(re): separate model rounds from MCP requests in fair trials
Validate per-response usage totals, retain the negative large-VM comparison and qualify the client executor, cache and investigation-depth differences. Add optional common envelope/stopping guidance without relabeling historical runs.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 06:29:26 +00:00
15d75ea1ea fix(experiment): preserve cached fact ownership for SDK callers
Freeze parsed inventories once at their owning boundary and share the rule with native IR and function views. Verify the real SDK mutation regression and retain corrected equal-depth SQLite model costs, including persistent-cache limitations.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 06:07:46 +00:00
f3f209d03f experiment(re)!: preserve partial captures and resolve function entries
Keep validated code available when native graph extraction is unavailable, bind output to the actual input/entry, preserve cancellation observations, and share batch source/error semantics. Add filtered function/alias reads and slice field selection across adapters.

Verified 38 focused tests, targeted lint, real stock/native MCP workflows, large SQLite graph-capacity fallback, and real cold/warm MCP reuse. Retain actual model costs and unequal-depth/selection limitations instead of claiming general superiority.

BREAKING CHANGE: Experimental native snapshots may succeed with readable code and an unavailable recovered_ir facet. Inspect facets before graph reads; unavailable graph counts are null/unknown.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 05:48:03 +00:00
4f13bd6733 experiment(re)!: separate IR origins from machine instruction facts
A real model conflated an optimized CFG edge with the bytes at its instruction origin. Make provenance explicit and add bounded, batched instruction windows from the existing Kuna CLI, preserving exact bytes, per-window failures and Evidence. Record the task-cost comparison without treating the incomplete machine-path answer as an equal-quality win.

Verified 33 focused tests, targeted lint, real stock/patched MCP decoding with mixed success, and a real 4096-row capacity run under the resource observer.

BREAKING CHANGE: Experimental operation.address/addresses and block.start become origin_address/origin_addresses and origin_start. Printed CFG target_address becomes target_origin; historical source captures are unchanged.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 04:51:55 +00:00
a7e9f8843b experiment(re): expose native CFG and value identities for composition
[AUTOMATED] Preserve the opt-in pinned Kuna exporter as a reproducible source patch. Share typed graph/value reads and derived control dependence through the SDK, CLI and MCP adapters, retaining unknown runtime and memory semantics.

Validated 29 focused tests, lint, actual diagnostic/native MCP workflows and a real zlib CLI composition. C and line mappings match the prior stock capture. Full upstream corpus/workspace gates and revised real-model costs remain unverified.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 04:28:22 +00:00
c5cefd622b experiment(re): refine composition contracts from a real model trace
Record genuine Code Mode use and full model usage rather than response-only estimates. Describe return shapes, permit byte-budgeted internal C reads, separate guest CPU from analysis deadlines, and move subsequent model profiles outside the source tree. Verify lease, admission, cleanup-retry and cancellation semantics.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 03:28:30 +00:00
5aa1c5c64f experiment(re): expose shared atomic facts through isolated Code Mode
Add a compact rea-codemode MCP server, bounded QuickJS execution and shared SDK/direct adapters. Reuse native target and process owners; retain evidence and lease-scoped snapshots. Extend the real-model trial runner with an isolated experimental MCP registration. Capability and evaluation work remains in progress.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-16 03:17:05 +00:00