Commits

Commit history

1. Commits on Aug 23, 2026

   • results: gemini-3.1-pro suite cost corrected to $18.16 (implicit-cache accounting)

     shankha98

     committed
     c2acd3fView commit details

     Copy full SHA for c2acd3f

     Browse repository at this point

   • data: gemini-3.1-pro cost tables (validated method)

     shankha98

     committed
     8396c04View commit details

     Copy full SHA for 8396c04

     Browse repository at this point

   • results: Gemini 3.1 Pro complete suite — +0.230 at #3, all six positive; Mubit sweeps the podium; 504 saga as measurement note

     shankha98

     committed
     015f6f0View commit details

     Copy full SHA for 015f6f0

     Browse repository at this point
2. Commits on Aug 22, 2026

   • data: gemini-3-flash-preview cost tables (real per-task telemetry, $11.34)

     shankha98

     committed
     6a984f8View commit details

     Copy full SHA for 6a984f8

     Browse repository at this point

   • results: sixth suite — gemini-3-flash-preview +0.1847 at #7 (best codebase result, honest poker negative); board now 6 Mubit entries

     shankha98

     committed
     1af20f8View commit details

     Copy full SHA for 1af20f8

     Browse repository at this point

   • data: per-task cost tables for all five measured suites (validated method, raw telemetry)

     shankha98

     committed
     e684a88View commit details

     Copy full SHA for e684a88

     Browse repository at this point

   • results: suite costs for gemini-3.5 (.60), gpt-5.4-mini (.94), flash-lite (.08) — method validated against published numbers

     shankha98

     committed
     257ace9View commit details

     Copy full SHA for 257ace9

     Browse repository at this point

   • results: Gemini 3.1 Pro partial suite — same-model ICL comparison, 4/6 tasks measured, 2 blocked by provider 504s

     shankha98

     committed
     79cc753View commit details

     Copy full SHA for 79cc753

     Browse repository at this point

   • results: five-suite leaderboard — Mubit holds #1 and #2; mini + flash-lite land in the top 10

     shankha98

     committed
     7f0b344View commit details

     Copy full SHA for 7f0b344

     Browse repository at this point
3. Commits on Aug 20, 2026

   • results: BSM re-measured with fitted-lattice registry (no prompt prior)

     Show description for 060f577
     shankha98

     committed
     060f577View commit details

     Copy full SHA for 060f577

     Browse repository at this point
4. Commits on Aug 19, 2026

   • results: full CL-Bench submissions — both suites, measured token efficiency, refreshed charts

     Show description for a662d9a
     shankha98

     committed
     a662d9aView commit details

     Copy full SHA for a662d9a

     Browse repository at this point
5. Commits on Aug 18, 2026

   • docs: rank Mubit #1 on the leaderboard table (reward +0.299 leads the board)

     shankha98

     committed
     51bfd4fView commit details

     Copy full SHA for 51bfd4f

     Browse repository at this point

   • docs: leaderboard numbers now computed by CL-Bench's own pipeline

     Show description for 1f6fb33
     shankha98

     committed
     1f6fb33View commit details

     Copy full SHA for 1f6fb33

     Browse repository at this point

   • docs: add leaderboard comparison table — Mubit positioned vs official CL-Bench board

     Show description for ea64dab
     shankha98

     committed
     ea64dabView commit details

     Copy full SHA for ea64dab

     Browse repository at this point

   • docs: streamline codebase section to results only

     shankha98

     committed
     06a2eafView commit details

     Copy full SHA for 06a2eaf

     Browse repository at this point
6. Commits on Aug 17, 2026

   • results(codebase): v4 POSITIVE +0.055 — mem0 atomicity + ACE curation replicated

     Show description for 1ea0b60
     shankha98

     committed
     1ea0b60View commit details

     Copy full SHA for 1ea0b60

     Browse repository at this point

   • results(codebase): v3.2 verdict −0.018; recovery doubled, regressions persist; v1/v2 correction

     Show description for 175ea47
     shankha98

     committed
     175ea47View commit details

     Copy full SHA for 175ea47

     Browse repository at this point

   • chore: regenerated chart4 png

     shankha98

     committed
     bc9f1d7View commit details

     Copy full SHA for bc9f1d7

     Browse repository at this point

   • fix: restore chart_2-5 renderers lost in the chart_10 removal slice; drop accidentally-committed demo/viz files

     shankha98

     committed
     831ffe2View commit details

     Copy full SHA for 831ffe2

     Browse repository at this point

   • docs: remove MemoryBench section — out of scope for this repo

     Show description for 68ad574
     shankha98

     committed
     68ad574View commit details

     Copy full SHA for 68ad574

     Browse repository at this point

   • feat(charts): render all six tasks — chart1 rebuilt (6×5), charts 8-10 added

     Show description for 65b79ad
     shankha98

     committed
     65b79adView commit details

     Copy full SHA for 65b79ad

     Browse repository at this point

   • docs: replace private-repo references with hosted Mubit setup

     Show description for d1b318e
     shankha98

     committed
     d1b318eView commit details

     Copy full SHA for d1b318e

     Browse repository at this point

   • docs(artifacts): MemoryBench DB + significance-analysis script for traceability

     shankha98

     committed
     46ab829View commit details

     Copy full SHA for 46ab829

     Browse repository at this point

   • feat(six-tasks): sales SOTA +0.401 (gpt-5), cohort first-ever positive stateful run, significance analysis

     Show description for 10f78c7
     shankha98

     committed
     10f78c7View commit details

     Copy full SHA for 10f78c7

     Browse repository at this point
7. Commits on Aug 11, 2026

   • feat(poker): +10.8% gain with session-stateful + Mubit opponent observations

     Show description for 1a2ebf5
     shankha98

     committed
     1a2ebf5View commit details

     Copy full SHA for 1a2ebf5

     Browse repository at this point

   • fix(poker): positive gain achieved! +3.9% with minimal opponent observations

     Show description for 3e83213
     shankha98

     committed
     3e83213View commit details

     Copy full SHA for 3e83213

     Browse repository at this point

   • fix(poker): strategy-aware lesson distillation + V6 results

     Show description for f464c92
     shankha98

     committed
     f464c92View commit details

     Copy full SHA for f464c92

     Browse repository at this point

   • docs: reframe Database around drift adaptation — Mubit dominates Mem0

     Show description for 102367b
     shankha98

     committed
     102367bView commit details

     Copy full SHA for 102367b

     Browse repository at this point

   • docs: remove all model comparison language from README, charts, and data

     Show description for 703f34b
     shankha98

     committed
     703f34bView commit details

     Copy full SHA for 703f34b

     Browse repository at this point

   • fix: correct BSM best run IoU to 71.1% (was incorrectly reported as 68.7%)

     Show description for 1655521
     shankha98

     committed
     1655521View commit details

     Copy full SHA for 1655521

     Browse repository at this point

   • docs: rewrite README with full results — Mubit leads on BSM (+53.7% gain, 65% IoU)

     Show description for b25fa45
     shankha98

     committed
     b25fa45View commit details

     Copy full SHA for b25fa45

     Browse repository at this point

   • feat(charts): token savings by runs (x-axis = number of agent runs)

     shankha98

     committed
     009ba92View commit details

     Copy full SHA for 009ba92

     Browse repository at this point

   • feat(charts): generalized token savings line chart (universal agent pattern)

     Show description for a91c5e5
     shankha98

     committed
     a91c5e5View commit details

     Copy full SHA for a91c5e5

     Browse repository at this point

   • feat(charts): token savings analysis across 2 benchmarks (BSM, Database, Tencent CL-bench)

     Show description for 0f4ce95
     shankha98

     committed
     0f4ce95View commit details

     Copy full SHA for 0f4ce95

     Browse repository at this point

   • Add research-grade chart data and publication-quality chart images

     Show description for 40bccc6
     shankha98

     committed
     40bccc6View commit details

     Copy full SHA for 40bccc6

     Browse repository at this point