Proof — verifiable claims across the SUM Equities portfolio
Trust dashboard
Proof
Every verifiable claim across the portfolio, in one place. Each entry carries an explicit epistemic status — proved, empirically benchmarked, shipped, or in-progress — and names its source artifact. Where the repository is public, the source links directly; private references are named but not linked. Status is per-claim, not per-project: a claim can be shipped (its artifact exists and runs) while the parent project is still in development. Each row shows the parent project's status next to the claim so the relationship is unambiguous. Filter below. Every load-bearing claim we make is listed here; project pages carry the implementation detail behind each one.
50 claims across 14 projects · Aggregated at build time from each project's frontmatter.
Proved (2)
Mechanically proven. Enforced by deterministic code and verified by tests, including cross-runtime witnesses (epistemic status: `provable`) — not a formal mathematical proof.
- SUM — Verifiable Bidirectional Knowledge Distillation Project: Beta
Three-runtime byte-symmetric Ed25519 over JCS bytes (Python ↔ Node ↔ browser)
- SUM — Verifiable Bidirectional Knowledge Distillation Project: Beta
Canonical round-trip reconstruct(parse(canonical_tome(S))) == S — 0.00% drift on every CI run
Empirical benchmark (5)
Measured under a reproducible harness; numbers cite the corpus and command.
- Memory Mind Mesh — Living Memory for AI Project: In development
After corrections, accuracy on held-out phrasings improves: mean embedding cosine-to-gold 0.60 → 0.99 across 10 single-fact memories with 3 corrections each, 10 topics improved and 0 regressed (gpt-4o-mini, temperature 0)
memory-mind-mesh eval/receipt.json (private repo; reproduced byte-identical at temperature 0) private repository
- SUM — Verifiable Bidirectional Knowledge Distillation Project: Beta
Slider fact preservation median 1.000 (p10 0.769 long n=16 / 0.818 short n=8) — measured; same-commit replay receipt pending (bench-hardening T2/T3)
- SUM — Verifiable Bidirectional Knowledge Distillation Project: Beta
Extraction F1 1.000 on seed_v1 (50 hand-authored single-fact SVO sentences); precision 1.000, F1 0.762 on seed_v2 (20 hand-authored harder prose patterns — apposition, passive, relative clause, conjunction). Synthetic author-written corpora; no clinical, customer or third-party text has been measured. Reproduce: python -m scripts.bench.run_bench --corpus scripts/bench/corpora/seed_v1.json --no-llm
- SUM — Verifiable Bidirectional Knowledge Distillation Project: Beta
T1 iterated round-trip K=10 stability — STABLE on all three corpora (seed_v1, seed_v2, seed_long_paragraphs)
- SUM — Verifiable Bidirectional Knowledge Distillation Project: Beta
First real certified chain, over the first 32 BillSum test bills (CC0-1.0): per-hop 95%-confidence strict-NLI meaning-loss upper bounds of 0.865768 (hop 1, abstractive summarization performed by the dataset itself, not by SUM) and 0.488860 (hop 2, deterministic lead-N extractive compression), composed to a Bonferroni budget ≤ 1.354628 at joint confidence 0.90; separately measured direct end-to-end leg ≤ 0.874216. The bounds are wide at n=32 (Hoeffding), and the budget is not the end-to-end loss.
Shipped (33)
The named artifact (test, doc, code path) exists in the repo today and is runnable. Per-claim, not per-project — a claim can be shipped while the parent project is still in development.
- AgentXAgent — Agent Team Arena Project: In development
Bradley-Terry team ratings on the Elo scale (anchor 1000) with deterministic bootstrap 95% confidence intervals, exposed as a getRatings RPC, plus order-swapped LLM judging to cancel position bias — backend only, not yet surfaced in the arena UI
agentx-orchestrator/src/scoring/rating.rs, gateway/router.rs (getRatings), persistence/db.rs — PR #13, merged 2026-07-04 private repository
- To your Zenith Project: Production
107 curated figures in the Archive
OtotaO/2enith data.ts (107 profile entries) private repository
- To your Zenith Project: Production
Defining achievements span ages 2 to 109
OtotaO/2enith data.ts ids 1 (Wang Yani, age 2) and 107 (Whang-od, age 109) private repository
- To your Zenith Project: Production
Persona Atlas and the One Real Step loop run locally in the browser with no account
OtotaO/2enith docs/PERSONA_ATLAS.md and docs/ZENITH_LOOP.md; live chunks PersonaAtlas-*.js and ZenithLoop-*.js private repository
- Chefrson — Smart Recipe & Inventory Management Project: Beta
Public MCP server card at /.well-known/mcp/server-card.json, declaring an HTTP transport at /api/mcp/v1
.well-known/mcp/server-card.json — served live, 200 application/json ↗
- Chefrson — Smart Recipe & Inventory Management Project: Beta
The MCP endpoint answers an unauthenticated JSON-RPC tools/list with 8 tools (all prefixed chefrson_): whats_for_dinner, save_recipe_from_url, list_my_cookbooks, share_cookbook, list_pantry, search_my_recipes, add_to_shopping_list, plan_my_week
JSON-RPC tools/list POSTed to /api/mcp/v1 with no credentials (the endpoint answers POST only; GET returns 405) unlinked
- Chefrson — Smart Recipe & Inventory Management Project: Beta
Public OpenAPI 3.0 description of the HTTP API at /api/docs
- Chefrson — Smart Recipe & Inventory Management Project: Beta
Declared allergens are enforced server-side on AI output: meals matching a declared allergen are dropped from generated meal plans and from AI recipe suggestions, and a meal plan emptied by the filter raises an error rather than returning an unsafe plan
src/lib/calendar/meal-plan-service.ts and src/app/api/recipes/suggest/route.ts (private repo) private repository
- Chefrson — Smart Recipe & Inventory Management Project: Beta
Plan pricing is canonical in code — Starter $5/mo or $49/yr, Pro $12.99/mo or $129.99/yr
src/lib/pricing.ts — PLAN_TIERS (private repo) private repository
- Comicke — AI Comic & Book Maker Project: In development
The apex comicke.com is live and serves the app (HTTP 200)
- Cornucopia Map — Interactive Farm Discovery Platform Project: Production
Dataset is 7,367 rows — 292 Cornucopia scorecard, 9 hand-curated, 7,066 MyHealthForward import. 298 rows carry a certification; 272 of those carry a Cornucopia 1-5 rating; no imported row carries either.
- Cornucopia Map — Interactive Farm Discovery Platform Project: Production
1,765 static URLs in the sitemap — 1,433 farm pages, 265 state×category pages, 52 state pages, 13 category pages, the directory hub and the map
- Cornucopia Map — Interactive Farm Discovery Platform Project: Production
An unknown state slug returns a real HTTP 404 rather than a soft 200
- HFAO — Hugging Face Agent Observatory Project: In development
Storage plane (ClickHouse + DuckDB) with parity tests — the tests that need a live ClickHouse, including the DuckDB↔ClickHouse parity test, skip unless HFAO_TEST_CLICKHOUSE_DSN is set
- HFAO — Hugging Face Agent Observatory Project: In development
Ingest plane (OTel GenAI + OpenInference normalizer, OTLP server, PII redaction, bounded buffer, body offload at 64 KiB)
- HFAO — Hugging Face Agent Observatory Project: In development
SPEC.md v1.0.0 locked; every build-plan commit cites a spec section (§15.3 rule 1)
- HFAO — Hugging Face Agent Observatory Project: In development
MCP-native queryability across observability primitives (FastMCP Streamable HTTP, §9.2 read surface + gated score_observation write)
packages/hfao/mcp_server/ + tests/acceptance/test_ac_9_mcp.py (v0.5.0, SPEC §9) ↗
- HFAO — Hugging Face Agent Observatory Project: In development
Closed eval-trace loop (traces ↔ datasets ↔ scores ↔ monitors), with causal attribution, cost rollups, and retention
- HFAO — Hugging Face Agent Observatory Project: In development
Stage 2 counterfactual replay — driver-pluggable, Tier-1 frameworks (LangGraph / OpenAI Agents SDK / Claude Agent SDK); tests use deterministic injected callables, not live framework or model calls
- HFAO — Hugging Face Agent Observatory Project: In development
Proactive anomaly surfacing (Insight schema + AnomalyEngine) and rule-based insight routing (subscriptions)
- InfiniteContext — Tiered Memory for AI Project: In development
Exact flat vector store with persisted, rebuildable index artifacts (save/load/merge/split), plus an approximate HNSW backend for retrieval at scale
- InfiniteContext — Tiered Memory for AI Project: In development
Hierarchical bucket memory with a multi-strategy retriever
- InfiniteContext — Tiered Memory for AI Project: In development
Automatic prompt/output categorization with keyword, vector-similarity, and adaptive strategies
- InfiniteContext — Tiered Memory for AI Project: In development
Multi-level summarization via OpenAI-compatible chat completions, with a deterministic extractive fallback when no LLM client is configured
- InfiniteContext — Tiered Memory for AI Project: In development
Negation-aware retrieval (Widdows orthogonal negation)
src/core/NegationAwareRetrieval.ts; tests/core/NegationAwareRetrieval.test.ts ↗
- InfiniteContext — Tiered Memory for AI Project: In development
Pluggable storage tiers — local disk and Google Drive behind a common provider interface
src/providers/{StorageProvider,LocalStorageProvider,GoogleDriveProvider}.ts ↗
- InfiniteContext — Tiered Memory for AI Project: In development
One governance contract across all stored memory classes — redaction, deletion with tombstones, retention, and export — with tamper-evident Ed25519 signed receipts chained over governed operations, signing-key rotation retaining retired keys for verification, and erasure that survives restart
- InfiniteContext — Tiered Memory for AI Project: In development
Continuous integration on Node 20 and 22 (build + test)
- SUM — Verifiable Bidirectional Knowledge Distillation Project: Beta
Render receipt format sum.render_receipt.v1 (Ed25519 / JCS / detached JWS) — verifier in three runtimes
- SUM — Verifiable Bidirectional Knowledge Distillation Project: Beta
Transform substrate (sum.transform_receipt.v1 + registry: slider / extract / compose) — 20-fixture cross-runtime K-matrix locks accept + reject across Python ↔ Node ↔ browser; T4 source-chain binding, T5 ShareableRender, T6 multi-school extract shipped in v0.7.0
- SUM — Verifiable Bidirectional Knowledge Distillation Project: Beta
sum verify --explain layered output (sum.verify_explained.v1) — seven per-dimension checks each tagged with epistemic_status; truth-of-content always not_asserted (test-locked invariant)
- SUM — Verifiable Bidirectional Knowledge Distillation Project: Beta
Compliance validators across six regimes (EU AI Act Art. 12 / GDPR Art. 30 / HIPAA §164.312(b) / ISO 27001 A.8.15 / SOC 2 CC 7.2 / PCI DSS v4.0 Req 10) — sum compliance check emits sum.compliance_report.v1
- SUM — Verifiable Bidirectional Knowledge Distillation Project: Beta
Certified multi-hop chain receipt sum.chain_receipt.v1 (Ed25519 / JCS / detached JWS over an ordered sequence of meaning-risk hops) — issued by `sum mint-chain`, replayed offline by a third party via `python -m sum_verify`; the mandatory budget_scope field is fail-closed. Research tier for issuance and dependency-light verification.
In progress (10)
Stated as a target. Not yet a verifiable claim — included for transparency.
- AgentXAgent — Agent Team Arena Project: In development
Public deployment URL for the arena UI + live-match viewer, with match history surface reachable to non-author visitors
- Comicke — AI Comic & Book Maker Project: In development
A public, inspectable artifact — the source is private, so nothing in 'What's Built' can be checked from outside
- HIPAA Voice Agent — Medical AI Scribe Project: In development
Independently auditable compliance posture (HIPAA / SOC 2 / equivalent), or a referenceable pilot deployment with a real clinical partner
- ICP-AR — Autonomous B2B Acquisition Engine Project: In development
Deployed behind Cloudflare Access; worker test suite (33 K-matrix + 4 A-matrix files) and engine math/improvement tests pass on local runs. CI exists (.github/workflows/ci.yml, added 2026-06-16) but every run on main failed at startup on then-blocked GitHub Actions billing; the last run was 2026-06-23 and the last green run was 2026-06-06 on branch ship-prep-2026-06-06, so the suite is still not independently verifiable
README / PROOF_BOUNDARY.md; workers/api/test; engine/test_*.py; .github/workflows/ci.yml private repository
- ICP-AR — Autonomous B2B Acquisition Engine Project: In development
Public (unauthenticated) demo URL, or read-access source, so the implementation can be inspected directly
- Simplytics — Lean Web-Vitals Dashboard Project: In development
Reproducible automated test suite enforcing privacy claims (no high-entropy identifiers, no full URLs / referrers / UA strings in tracked events) end-to-end on the deployed worker, with linkable results — or a referenceable third-party privacy audit. A privacy-attestation test file (test/privacy.attestation.test.js) exists in the private backing repo, but its CI has not had a green run since 2026-06-05 and the repo is not publicly readable, so the results are not linkable
- Memory Mind Mesh — Living Memory for AI Project: In development
Public source repository linked via githubUrl, or a downloadable usable public artifact (release binary, package, or runnable distribution)
- Medicoder Platform — AI Medical Coding Assistant Project: In development
Independent code/infrastructure audit confirming the no-PHI-persistence guarantee end-to-end against a deployed artifact (the guarantee is currently enforced by a hermetic CI test against the source, but nothing is deployed and no third-party audit exists yet)
- Medicoder Platform — AI Medical Coding Assistant Project: In development
Published evaluation of CPT E/M coding accuracy against a reference set, methodology and dataset class disclosed (non-patient-identifying)
- MTE — Meta-Template Engine Project: In development
Public source repository linked via githubUrl, or a public demo URL exercising the template-adaptation flow