📝 p4: honesty cleanup — AI claims stay unverified, release narrative = web framework (item #5)
swadhinbiswas committed 2 months ago
5426d7e
1 parent f05b19c
commit 5426d7e
1 file changed
+3 -1
Showing 1 changed file+3-1
▼PLAN.md
+3-1
| @@ -76,7 +76,7 @@ cloud-native readiness. | ||
76 | 76 | | 37 | OTel Distributed Tracing | ✅ verified | OTLP, contextvars, 225 tests | |
77 | 77 | | 38 | Hardware-Accelerated JWT & Security | ✅ verified | RS256/ES256, CORS builder, IpRateLimiter | |
78 | 78 | | 39 | SAST, Fuzzing & Memory Safety | ✅ verified | 6 fuzz targets, miri, SAFETY comments | |
79 | -| 40 | The JustAPI 2.0 "Singularity" Release | 🟡 in progress | **PyPI upload still blocked (no token)** — see Next actions | | |
79 | +| 40 | The JustAPI 2.0 Release | 🟡 in progress | **PyPI upload still blocked (no token)** — see Next actions. AI/inference phases 41-52 are 🟡/🔴 (no real GPU run, ADR-067): the release narrative is the **fastest, most complete web framework** (2.4x Granian hello-world, 5.4x JSON echo, native SSE/CRUD/DB-awaits, HTTP/3, free-threaded wheels) — NOT unproven LLM-serving claims. | | |
80 | 80 | | 41 | Native Inference Engine Foundation | 🟡 implemented / unverified | Candle CPU; `Engine`/`Model` traits, 24 tests. **No real-weights run (MockModel only)** | |
81 | 81 | | 42 | KV-Cache Manager (PagedAttention, Rust) | 🟡 implemented / unverified | `KvBlockPool`, `PrefixCache`; unit tests only, no real model | |
82 | 82 | | 43 | Continuous-Batching Scheduler | 🟡 implemented / unverified | `Scheduler` prefill/decode interleave; **CPU MockModel only, 0.4% of naive on instant-forward (scheduler sleep dominates)** | |
| @@ -613,6 +613,8 @@ Don't duplicate that reasoning here — just reference the entry. | ||
613 | 613 | |
614 | 614 | ## Next actions |
615 | 615 | |
616 | +0. **[DONE — 2026-08-07] Honesty cleanup (item #5):** AI phases 41-52 remain 🟡/🔴 (no real GPU). The release narrative is the web-framework story (benchmarked); LLM claims stay unverified and are NOT part of the v2.0.8 public claims. New verified wins recorded: multi-worker prefork scaling (1.88x at 4 workers), native async DB awaits query_async/execute_async (53x on slow queries, ADR-093), mypy-clean type stubs (py.typed), scaffold demos the differentiators. | |
617 | + | |
616 | 618 | 1. **Framework Release:** |
617 | 619 | - Package all crates and Python distributions for PyPI/crates.io. |
618 | 620 | - Publish documentation and guides. |
619 | 621 | |