StoneyTECH StoneyTECH.net

Evidence ledger · public receipt inventory

Proof Of Work

Receipts keep the Determinism Ladder honest. Each item names the trade, the artifact, the current checks, and the next proof needed.

The page is a public ledger, not a victory lap. It shows what shipped, what is partial, and where the evidence chain still needs another turn.

37

Receipts

34

Linked public items

7

Repository plans

Sanitized public panel record

Verification receipts

The site does not publish raw model votes or private run internals. It does publish the public shape of a run: status, panel date, confidence, satisfied count, architecture, and the article it judged.

9

Verified

0

Advisory

13

Pending

pending panel demystify

LLM construction stages, from pretraining to LoRA

Panel date
Pending
Confidence
—
Satisfied
—
Architecture
Pending first panel run

Initial public primer drafted from primary papers and public model-building concepts. Needs GVAR panel pass before promotion to verified status.

pending panel demystify

What is MCP? The USB-C port for AI context

Panel date
Pending
Confidence
—
Satisfied
—
Architecture
Pending first panel run

Initial public primer drafted for the StoneyTECH MCP page. Needs GVAR panel pass before promotion to verified status.

panel verified demystify

AI vs ML vs LLM vs agents — sorting out the words people keep mixing up

Panel date
2026-05-03
Confidence
0.98
Satisfied
5/ 6
Architecture
GVAR v3.3 — 6-verifier panel + Path A purity

Panel-verified clean under GVAR v3.3 execution 8003 on 2026-05-03 — 5 of 6 verifiers satisfied at average confidence 0.98, ZERO critical findings, 1 IMPORTANT (Security; same primer-register trade pattern as piece #1), 3 nice-to-haves. Cross-piece calibration data (Demystify #1 vs #2): IDENTICAL convergence pattern — both pieces 5/6 satisfied, 0 critical, 1 Security IMPORTANT, panel confidence 0.98, all cross-family voice verifiers >=0.97. The Security lens consistently finds "primer touches a security topic at audience depth without naming the threat at architect depth" — which is by design (the cross-link to the architect-register treatment is the deliberate handling). Architecture lens consistently finds "could extend into deployment-context territory" — also by design. The voice template is stable across two iterations. The panel response is predictable. The Demystify register can be wired into learning-agent (register=primer mode) with high confidence that generated drafts will pass the panel on first run with the same convergence pattern. Second piece of the Demystify AI series. Audience: technical generalists (IT, security, ops, project managers, anyone using AI tools daily but never given a working mental model).

panel verified demystify

LLMs work like word-query databases, but looser

Panel date
2026-05-03
Confidence
0.98
Satisfied
5/ 6
Architecture
GVAR v3.3 — 6-verifier panel + Path A purity

Panel-verified clean under GVAR v3.3 execution 7841 on 2026-05-03 — 5 of 6 verifiers satisfied at average confidence 0.98, ZERO critical findings. Calibration data for the rest of the Demystify AI series: cross-family voice verifiers all converged at >=0.97. Primer-register voice holds against the same gold-standard signal as architect-register. Security/Architecture lenses carry less load in a primer (no architectural decisions to threat-model), but still surface real gaps when an audience anecdote touches a known attack surface. The convergence rule handles primer pieces cleanly without re-tuning. First piece of the Demystify AI series. Audience: technical generalists — IT coworkers, sysadmins, helpdesk leads, project managers, security analysts who use AI tools daily but lack a working mental model for the mechanism under the hood. Voice: pragmatic, friendly, anti-hype. Structure: give an almost-correct metaphor, refine it just enough for use, then show why the imperfection creates the value.

panel verified learn

Deployment context first — when on-prem, sovereign-cloud, and public-cloud are different architectures

Panel date
2026-04-29
Confidence
0.98
Satisfied
5/ 6
Architecture
GVAR v3.3 — 6-verifier panel + Path A purity

Panel-verified clean under GVAR v3.3 (6-verifier panel + Path A purity) execution 2748 on 2026-04-29 — 5 of 6 verifiers satisfied at average confidence 0.98, ZERO critical findings, 1 IMPORTANT (Security lens on prompt-caching threat surface; recorded above for follow-up). Architecture lens summary: "fully centers deployment context as decision-zero and does not treat cloud-managed APIs as the universal architectural default. It explicitly distinguishes public cloud, sovereign/private cloud, restricted-network, and true air-gap architectures across models, APIs, RAG, agents." Sister piece to GVAR-36 (threat-surface-layer-by-layer); together they close the v3.2 panel's two gap classes (axioms #17 and #18) at the depth the panel asked for. 104-second panel run on the Path A self-verify path. Per the convergence rule (≥4/6 satisfied + no criticals), the Security IMPORTANT does not block panel-verified status; it is recorded as outstanding refinement for a future pass.

panel verified learn

The threat surface, layer by layer — a security companion to the agentic stack

Panel date
2026-04-29
Confidence
0.99
Satisfied
6/ 6
Architecture
GVAR v3.3 — 6-verifier panel + Path A purity

Panel-verified clean under GVAR v3.3 (6-verifier panel + Path A purity) execution 2746 on 2026-04-29 — 6 of 6 verifiers satisfied at average confidence 0.99, ZERO critical findings, ZERO importants, 2 nice-to-have refinements (recorded above for a future pass). Security lens: "fully satisfies the Axiom #17 threat-surface lens. It names adversarial-input paths, prompt injection vectors, supply-chain assumptions, credential/token disclosure risks, data exfiltration paths, confused-deputy risks, memory poisoning, tool-output injection." Architecture lens: "satisfies the deployment-context-first requirement. It does not treat cloud-managed APIs as universal defaults." Structure pre-reviewed with GPT-5.5-pro via Oracle MCP (verdict: Approach A — seven sub-sections per lever, 4-part rhythm). 41-second panel run (Path A self-verify path).

panel verified learn

Cheaper alternatives to MCP — when gh, kubectl, and curl beat the protocol

Panel date
2026-04-29
Confidence
0.98
Satisfied
5/ 6
Architecture
GVAR v3.3 — 6-verifier panel + Path A purity (Self-Verify Check ifElse skips Generator when target_path == gold_standard_path)

Panel-verified under GVAR v3.2 (6-verifier panel) execution 2500 on 2026-04-28. Convergence rule satisfied: 5 of 6 verifiers satisfied at confidence ≥ 0.95, zero critical findings. The Security lens returned needs_changes with one IMPORTANT (not critical) finding — §2 (Single REST endpoint via plain function calling) inherits the same threat-model coverage gap that the original Cheaper-alternatives section closed for shell + CLI; the augmentation didn't extend the threat-model section to also cover the REST path. Recorded in outstanding_refinements; addressed in a follow-up pass (per the convergence rule, important-only findings don't block panel-verified status, but they're permanent record). Architecture lens satisfied cleanly with one nice-to-have (REST placement still depends on residency/regulatory context). Panel confidence 0.98.

panel verified learn

The agentic stack — 7 levers from foundation to autonomy

Panel date
2026-04-29
Confidence
0.99
Satisfied
6/ 6
Architecture
GVAR v3.3 — 6-verifier panel + Path A purity (Self-Verify Check ifElse skips Generator when target_path == gold_standard_path)

Panel-verified clean under GVAR v3.3 (6-verifier panel + Path A purity) execution 2719 on 2026-04-29. 6 of 6 verifiers satisfied at confidence >= 0.95, ZERO critical findings, zero importants. Average panel confidence 0.99. Both domain-specialized lenses (Security, Architecture) returned clean satisfied verdicts: Security at 0.99 with 0 critiques ("the article matches the gold standard and closes the prior security-lens gaps. It now names the threat surface for every lever, adds a matrix Threat-surface column, covers prompt injection, credential exposure, supply-chain risk, confused-deputy paths, excessive agency, memory poisoning, and deployment-context"); Architecture at 0.99 with 0 critiques ("the article now satisfies axiom #18: deployment context is explicitly chosen before model selection, and the decision tree correctly treats cloud, sovereign/private cloud, and on-prem/ air-gap as first-order architectural constraints"). The 16 critiques v3.2 surfaced on run 2201 are ALL closed. GVAR v3.3 architecture introduced 2026-04-29: a Self-Verify Check ifElse node after Fetch Gold Standard routes self-verify (target_path == gold_standard_path) directly to a Synthesize Self-Verify Inputs Code node that produces the same JSON shape as Parse Generator Output, then converges with the non-self- verify branch at Build Verifier Prompt. The Generator is no longer called in self-verify mode. Run time dropped from 7+ min (with frequent Anthropic API timeouts) to 60 seconds. The path here was: v3.2 run 2201 surfaced 16 critiques; commit 28c1248 augmented the article to close them; v3.2 run 2499 confirmed 15 closed but caught one residual CRITICAL at the MCP "Cheaper alternatives first" inline callout; commit 28c1248 added the inline caveat; three subsequent re-runs (2601, 2607, 2617) blocked by Generator API timeouts; GVAR-38 workflow patch bypassed Generator entirely in self-verify; v3.3 run 2719 converged 6/6 clean. The verification chain doing exactly what it exists for, end-to-end.

panel verified learn

The eighth lever — eval and observability, the rung the rest of the ladder rests on

Panel date
2026-04-28
Confidence
0.99
Satisfied
6/ 6
Architecture
GVAR v3.3 — 6-verifier panel + Path A purity (Self-Verify Check ifElse skips Generator when target_path == gold_standard_path); confirmed by run 2723 on 2026-04-29 (6/6 satisfied, 0 critiques, 0.99 confidence)

Panel-verified clean under GVAR v3.2 (6-verifier panel) execution 2501 on 2026-04-28. 6 of 6 verifiers satisfied at confidence ≥ 0.95, ZERO critiques across all severities. The augmented essay (PII/PHI surface in the trace store, placement-by-classification table with self-host alternatives, "Threat model for LLM-as-judge" section 4a) closed both domain-specialized lenses on the first re-run. Security-lens summary: "It names the trace store as a PII/PHI and credential-adjacent exfiltration surface, addresses deployment context and data residency." Architecture-lens summary: "It explicitly rejects cloud-managed observability as the universal default, ties trace placement to data classification and residency obligations." Panel confidence 0.99.

panel verified learn

LoRA + RAG, composed — a worked example

Panel date
2026-04-28
Confidence
0.98
Satisfied
5/ 6
Architecture
GVAR v3.3 — 6-verifier panel + Path A purity; confirmed by run 2732 on 2026-04-29 (5/6 satisfied, 0 critical, 1 Security IMPORTANT on OWASP completeness; PASS). Predecessor 2724 returned 3/6 because 'zero added latency' phrase tripped 3 verifiers; commit 905d449 softened to 'negligible (when merged at load time)' and added a user-originated prompt-injection bullet.

Panel-verified under GVAR v3.2 (6-verifier panel) execution 2502 on 2026-04-28. Convergence rule satisfied: 5 of 6 verifiers satisfied at confidence ≥ 0.95, zero critical findings. The Security lens returned needs_changes with one IMPORTANT finding flagging that the OWASP LLM Top 10 (2025) numbering in the new threat-model section may not match the current OWASP registry — recorded in outstanding_refinements; verify and refine in a follow-up pass. Architecture lens satisfied cleanly. Cross-family verifiers all satisfied; their nice-to-haves (zero-latency phrasing, 17× cost framing, chunk-size measurement, injection-strip caveat) are pre-existing refinements carried in outstanding_refinements. Panel confidence 0.98.

panel verified learn

Model is portable — except when it isn't

Panel date
2026-04-28
Confidence
0.98
Satisfied
4/ 4
Architecture
GVAR v3.3 — 6-verifier panel + Path A purity; confirmed by run 2726 on 2026-04-29 (6/6 satisfied, 0 critiques, 0.97 confidence; tiered-system data-exfil fix at small→frontier escalation boundary closed prior Security IMPORTANT)

Panel-verified under GVAR v3.2 (6-verifier panel) execution 2503 on 2026-04-28. Convergence rule satisfied: 5 of 6 verifiers satisfied at confidence ≥ 0.95, zero critical findings. The Security lens returned needs_changes with one IMPORTANT finding — the latency-critical workaround "tiered system where a fast small model handles 90% of queries and only escalates to a frontier model on the long tail" doesn't name the data-exfiltration / classification- leak surface at the escalation boundary. Recorded in outstanding_refinements; addressed in a follow-up pass. Architecture lens summary: "It explicitly makes deployment context the first architectural decision when residency, latency, air-gap, regulatory, benchmark, or specialization constraints bind." Cross-family verifiers all satisfied at high confidence — Gemini and Grok perfect on the byte-identical augmented article. Panel confidence 0.98. History note (preserved from prior cycle): the v3.1A run executed 1787 only after the panel caught a critical bug in execution 1783 — the frontmatter axiom_outcomes[12].note had listed five exceptions that did not match the body's actual five sections. Fix in commit 6794f39, re-ran, panel converged 4/4. The verification chain doing exactly what it exists for; v3.2 confirmed it on the augmented artifact 2026-04-28 at 5/6.

Shipped

Public artifact exists and local checks pass.

25

Shipped governance

Narrator-free public voice contract

#

The style gate rejects first-person framing, second-person address, weak connector prose, passive drift, and employer-implied narration across published articles and public contract strings.

Ladder role
Public prose moves from personal narration into bounded publication constraints.
Evidence artifact
scripts/verify-public-style-contract.js
Current checks
  • npm run test:public-style
Next proof
Extend the gate as new public surfaces enter the contract.

Links

page / homepage / about Axiom #9Axiom #13Axiom #16

Shipped governance

Public identity contract

#

The identity gate keeps the site framed as anonymous learning synthesis, public-source study, and reference build notes with no employer representation or originality claim.

Ladder role
Authorship moves from personal identity into StoneyTECH publication posture.
Evidence artifact
scripts/verify-public-identity-contract.js
Current checks
  • npm run test:public-identity
  • npm run build
Next proof
Keep disclosure language aligned across pages, feeds, and generated JSON.

Links

page / homepage / about Axiom #11Axiom #13Axiom #21

Shipped mcp

Public content static contract

#

The generator exports published pages, articles, axioms, builds, ladder placements, applied evidence, and proof receipts from one static contract.

Ladder role
Published site content moves into deterministic JSON for agents.
Evidence artifact
static/stoneytech-public-content.v1.json
Current checks
  • npm run test:public-content
  • npm run build
Next proof
Keep every new public artifact tied to receipt metadata before merge.

Links

page / mcppage / proof-of-workbuild / public-content-mcp Axiom #2Axiom #9Axiom #16Axiom #21 resource:public-content-contracttool:get_site_overview

Shipped mcp

Public site graph

#

The site now exports a public graph linking pages, articles, axioms, builds, repositories, proof receipts, and the public MCP surface itself.

Ladder role
Published relationships move from implied navigation into an explicit graph for agents.
Evidence artifact
static/stoneytech-public-graph.v1.json
Current checks
  • npm run test:public-content
  • npm --prefix stoneytech-site-mcp test
  • npm run build
Next proof
Add richer graph traversals and more visible human-facing proof chains on major pages.

Links

page / mcppage / proof-of-workpage / axioms Axiom #2Axiom #9Axiom #16Axiom #21 resource:public-content-graphtool:get_graph_neighbors

Shipped governance

Site self-judgment loop

#

The axioms page now names how StoneyTECH judges its own implementation: what holds, what is partial, and what proof still has to close.

Ladder role
Mission and axioms move from doctrine into a public scorecard judging the site itself.
Evidence artifact
/axioms#self-judgment
Current checks
  • npm run test:public-content
  • npm run test:public-style
  • npm run build
Next proof
Tighten the weakest partial areas until more mission checks can graduate from partial to held.

Links

page / axiomspage / proof-of-workpage / home Axiom #9Axiom #13Axiom #16Axiom #21

Shipped evals

Verification status gate

#

The build refuses article publication without a verification status matching the local verification log contract.

Ladder role
Publication claims move into code-enforced frontmatter and log checks.
Evidence artifact
scripts/validate-verification.js
Current checks
  • node scripts/validate-verification.js
  • npm run build
Next proof
Connect verification status directly to each public receipt record.

Links

page / learnpage / proof-of-workbuild / stoneytech-sitebuild / gvar-engine-v2 Axiom #4Axiom #5Axiom #9

Shipped governance

Axioms catalog

#

The axiom catalog gives recurring engineering judgment stable names, tiers, citations, and applied-evidence counts.

Ladder role
Repeated judgment moves from memory into explicit cited principles.
Evidence artifact
src/lib/data/axioms.js
Current checks
  • npm run test:public-content
  • npm run build
Next proof
Map more axioms to ladder rungs and receipt trails.

Links

page / axiomsbuild / stoneytech-site Axiom #1Axiom #2Axiom #4Axiom #9Axiom #11Axiom #13Axiom #16Axiom #21

Shipped skills

Glossary sidecars

#

Reusable sidecars keep loaded terms short in prose while giving human and agent readers enough local context.

Ladder role
Definitions move from reader inference into local linked explanations.
Evidence artifact
src/lib/data/glossary.js
Current checks
  • npm run test:public-content
Next proof
Add ladder-specific glossary links to proof and article sidecars.

Links

page / demystifydemystify / 2026-05-09-ai-ml-llm-agents-sorting-out-the-wordsdemystify / 2026-05-05-what-is-mcp Axiom #1Axiom #2Axiom #16 definition-sidecar-package

Shipped governance

Threat-surface companion essay

#

The companion article maps AI stack levers to threat surfaces, citations, and mitigations before higher agency enters the design.

Ladder role
Capability choices pair with explicit attack-surface review.
Evidence artifact
/learn/2026-05-04-threat-surface-layer-by-layer
Current checks
  • npm run test:public-content
  • npm run test:public-style
  • npm run build
Next proof
Add a public matrix generator or reusable checklist.

Links

learn / 2026-05-04-threat-surface-layer-by-layer Axiom #5Axiom #11Axiom #13Axiom #21 threat-surface-matrix-generator-repo

Shipped governance

Deployment-context companion essay

#

The deployment-context article places cloud, sovereign cloud, private cloud, and air-gap constraints before model selection.

Ladder role
Model selection moves behind deployment constraints.
Evidence artifact
/learn/2026-05-11-deployment-context-first
Current checks
  • npm run test:public-content
  • npm run test:public-style
  • npm run build
Next proof
Add a public deployment-context selector artifact.

Links

learn / 2026-05-11-deployment-context-firstlearn / 2026-04-27-model-portability-exceptions Axiom #1Axiom #2Axiom #13Axiom #21 deployment-context-selector-repo

Shipped mcp

Cheaper alternatives to MCP essay

#

The essay turns MCP adoption into a decision ladder: static files, APIs, CLI tools, and narrower contracts first.

Ladder role
Tool choice moves through lower-cost alternatives before protocol adoption.
Evidence artifact
/learn/2026-04-27-cheaper-alternatives-to-mcp
Current checks
  • npm run test:public-content
  • npm run test:public-style
Next proof
Add a public decision matrix artifact or repository.

Links

learn / 2026-04-27-cheaper-alternatives-to-mcppage / mcp Axiom #1Axiom #2Axiom #14Axiom #21

Shipped rag

LoRA plus RAG composition essay

#

The essay separates learned style, retrieved facts, and prompt behavior so each concern carries a smaller boundary.

Ladder role
Voice and facts move to separate rungs instead of prompt improvisation.
Evidence artifact
/learn/2026-04-27-lora-plus-rag-composition
Current checks
  • npm run test:public-content
  • npm run test:public-style
Next proof
Add a runnable demo or public synthetic example.

Links

learn / 2026-04-27-lora-plus-rag-compositiondemystify / 2026-05-03-why-llms-hallucinatedemystify / 2026-05-17-how-llms-are-built-and-where-lora-fits Axiom #1Axiom #2Axiom #11

Shipped governance

Prompt, context, fine-tune, gate placement essay

#

The essay closes the graph around existing Ladder pieces by mapping task framing, current facts, repeated behavior, external action, prevention, and proof to their proper system surfaces.

Ladder role
Prompt work, context work, fine-tuning, tools, gates, and evals become placement choices instead of competing slogans.
Evidence artifact
/learn/2026-05-17-prompt-context-fine-tune-gate
Current checks
  • npm run test:public-content
  • npm run test:public-style
  • npm run build
Next proof
Add a small interactive placement selector or graph-readable decision table.

Links

learn / 2026-05-17-prompt-context-fine-tune-gatelearn / 2026-04-26-the-stack-matrixlearn / 2026-04-27-lora-plus-rag-compositionlearn / 2026-04-27-eighth-lever-eval-and-observabilitydemystify / 2026-05-17-how-llms-are-built-and-where-lora-fits Axiom #1Axiom #2Axiom #5Axiom #11Axiom #14Axiom #16

Shipped graphs

Graph-constrained execution essay

#

The graph article explains why explicit nodes, edges, budgets, and gates make agent loops inspectable before autonomy grows.

Ladder role
Agent flow moves from implicit emergence into explicit topology.
Evidence artifact
/learn/2026-05-03-graph-constrained-execution
Current checks
  • npm run test:public-content
  • npm run test:public-style
Next proof
Add a clean public graph workflow reference implementation.

Links

learn / 2026-05-03-graph-constrained-execution Axiom #2Axiom #3Axiom #5Axiom #13 graph-workflow-convergence-repo

Shipped agents

Three SDKs, three jobs essay

#

The centerpiece comparison maps Anthropic TypeScript SDK, OpenAI Agents SDK, and LangGraph to three different agent jobs, with a matrix and selection tree.

Ladder role
SDK selection moves from fashion into job-shaped control-surface choice.
Evidence artifact
/learn/2026-05-05-three-sdks-three-jobs
Current checks
  • npm run test:public-content
  • npm run test:public-style
  • npm run build
Next proof
Wire companion README links and publish the clean-history public repository set.

Links

learn / 2026-05-05-three-sdks-three-jobs Axiom #1Axiom #2Axiom #11Axiom #13Axiom #14Axiom #18 sdk-comparison-centerpiece-repo

Shipped agents

Portable agent pattern kits essay

#

The follow-up essay explains why the Trinity repos ship runnable examples, repo-local MCP stubs, file-backed graphs, and provider-binding seams so a reader can bring a model without losing the pattern.

Ladder role
Portable public pattern repos move control into local graphs, local MCPs, and templates before heavier infrastructure appears.
Evidence artifact
/learn/2026-05-06-portable-agent-pattern-kits
Current checks
  • npm run test:public-content
  • npm run build
Next proof
Promote one repo-local MCP from stub into a stronger read-only server and keep the compliance scan in the release path.

Links

learn / 2026-05-06-portable-agent-pattern-kitslearn / 2026-05-05-three-repos-one-thesispage / mcp Axiom #1Axiom #2Axiom #11Axiom #14Axiom #16Axiom #21 sdk-comparison-centerpiece-repo

Shipped graphs

Local graphs first essay

#

The follow-up essay explains why the Trinity repos begin with file-backed graphs, repo-local MCP reads, and explicit upgrade triggers instead of starting with hosted graph gravity.

Ladder role
Relationship knowledge becomes portable and inspectable in repo-local graph files before it graduates into larger infrastructure.
Evidence artifact
/learn/2026-05-06-local-graphs-first
Current checks
  • npm run test:public-content
  • npm run test:public-style
  • npm run build
Next proof
Promote one repo graph from file-only doctrine into a richer query surface once the repo-local MCP has real traversal pressure.

Links

learn / 2026-05-06-local-graphs-firstlearn / 2026-05-06-portable-agent-pattern-kitslearn / 2026-05-03-graph-constrained-execution Axiom #1Axiom #2Axiom #5Axiom #14Axiom #16Axiom #21 graph-workflow-convergence-repo

Shipped agents

Shadow tribunals essay

#

The follow-up essay explains why the Trinity repos expose shadow tribunal seams, why second opinions should begin as non-blocking sentinels, and how weekly comparisons turn disagreement into evidence.

Ladder role
Second opinions move from intuition into named shadow roles, retained receipts, and explicit disagreement policy.
Evidence artifact
/learn/2026-05-06-shadow-tribunals
Current checks
  • npm run test:public-content
  • npm run test:public-style
  • npm run build
Next proof
Turn one shadow role into a real runnable comparison path with receipt-level disagreement output.

Links

learn / 2026-05-06-shadow-tribunalslearn / 2026-05-06-portable-agent-pattern-kitslearn / 2026-05-05-three-repos-one-thesis Axiom #1Axiom #2Axiom #5Axiom #13Axiom #14Axiom #16 sdk-comparison-centerpiece-repo

Shipped governance

Determinism Ladder source corpus

#

The source corpus names the ladder, its proof needs, page integration pattern, and maintenance rules before public presentation.

Ladder role
Site authorship moves into a single constraint system.
Evidence artifact
corpus/determinism-ladder.md
Current checks
  • npm run test:public-content
  • npm run build
Next proof
Keep corpus notes aligned with the public evidence ledger.

Links

page / determinism-ladderpage / proof-of-worklearn / 2026-04-26-the-stack-matrix Axiom #2Axiom #9Axiom #16

Shipped governance

Determinism Ladder public hub

#

The ladder hub gives each AI system layer a rung, autonomy pattern, determinism purchase, failure mode, and receipt trail.

Ladder role
The core frame becomes public navigation.
Evidence artifact
/determinism-ladder
Current checks
  • npm run test:public-content
  • npm run test:public-style
  • npm run build
Next proof
Keep the proof ledger linked from the hub and public contract.

Links

page / determinism-ladderpage / proof-of-worklearn / 2026-04-26-the-stack-matrixbuild / stoneytech-site Axiom #2Axiom #9Axiom #16

Shipped governance

Article ladder sidecars

#

Each published article carries ladder metadata and a sidecar showing rung, trade, failure mode, and receipt links.

Ladder role
Page meaning moves from standalone prose into graph placement.
Evidence artifact
LadderSidecar.svelte and article frontmatter
Current checks
  • npm run test:public-content
  • npm run test:public-style
Next proof
Add per-article proof ledger backlinks on article templates.

Links

page / learnpage / demystifydemystify / 2026-05-02-llms-as-a-loose-databasedemystify / 2026-05-03-tokens-context-attention-no-mathlearn / 2026-04-26-the-stack-matrix Axiom #2Axiom #16

Shipped evals

Builds ladder placement

#

Every build note carries ladder placement, public influences, axiom outcomes, and receipt references for agent-readable evidence.

Ladder role
Builds become receipts rather than a catalog.
Evidence artifact
src/lib/data/builds.js
Current checks
  • npm run test:public-content
  • npm run build
Next proof
Add public repository release links as clean-history repos go live.

Links

page / buildsbuild / stoneytech-sitebuild / gvar-engine-v2build / public-content-mcp Axiom #9Axiom #11Axiom #16

Shipped evals

Public proof-of-work ledger

#

The ledger turns the planning inventory into a public surface with status, ladder role, evidence artifact, checks, links, and next-proof gaps.

Ladder role
Receipts become browsable and agent-readable.
Evidence artifact
/proof-of-work
Current checks
  • npm run test:public-content
  • npm run test:public-style
  • npm run build
Next proof
Add live drift checks comparing deployed JSON, MCP responses, and page output.

Links

page / proof-of-workpage / determinism-ladderbuild / stoneytech-site Axiom #2Axiom #5Axiom #9Axiom #16 tool:list_proof_receiptstool:get_proof_receipts_for_published_item

Shipped graphs

Graph data fabric doctrine

#

The article and diagrams separate semantic graph meaning from hybrid persistence categories: relational, document, object, event, analytical, vector, search, cache, and ledger-style storage.

Ladder role
Graph-first doctrine moves from slogan into a vendor-neutral persistence placement model.
Evidence artifact
/learn/2026-05-17-graph-data-fabric
Current checks
  • npm run test:public-content
  • npm run test:public-style
  • npm run build
Next proof
Add a build note once a concrete public pattern repo implements the placement matrix.

Links

learn / 2026-05-17-graph-data-fabriclearn / 2026-05-03-graph-constrained-executionlearn / 2026-05-06-local-graphs-first Axiom #1Axiom #2Axiom #5Axiom #11Axiom #14Axiom #16Axiom #19

Shipped mcp

AI demystified MCP explainer

#

The MCP primer explains the protocol in plain language and links the public StoneyTECH endpoint boundary.

Ladder role
MCP comprehension moves from assumed knowledge into a primer.
Evidence artifact
/demystify/2026-05-05-what-is-mcp
Current checks
  • npm run test:public-content
  • npm run test:public-style
Next proof
Keep endpoint docs aligned with deployed Cloudflare URL and receipt tools.

Links

demystify / 2026-05-05-what-is-mcppage / mcp Axiom #1Axiom #2Axiom #21

Partial

Public artifact exists while repo packaging, metadata, or live drift checks remain open.

5

Partial mcp

Published-content MCP

#

The MCP reads the generated public contract and exposes published content only: pages, essays, axioms, builds, repository notes, applied evidence, ladder placements, and receipts.

Ladder role
Published content moves into typed read-only MCP tools.
Evidence artifact
stoneytech-site-mcp/
Current checks
  • npm --prefix stoneytech-site-mcp test
  • npm --prefix stoneytech-site-mcp run build
Next proof
Finish clean-history public repo packaging and add live endpoint drift checks.

Links

page / mcplearn / 2026-05-04-published-content-mcpsbuild / public-content-mcp Axiom #1Axiom #2Axiom #9Axiom #13Axiom #14Axiom #16Axiom #21 public-content-mcp-clean-history-repo tool:list_published_contenttool:get_published_itemtool:search_published_content

Partial evals

GVAR verifier loop

#

The public build note frames GVAR as citation-first learning from Google DeepMind Aletheia and Gemini Deep Think work, with no originality claim.

Ladder role
Editorial review moves from subjective trust into generator, verifier, refiner evidence.
Evidence artifact
/builds#gvar-engine-v2
Current checks
  • npm run test:public-content
  • npm run test:public-style
  • npm run build
Next proof
Publish a sanitized learning repo or paper-style explainer tied to public citations.

Links

build / gvar-engine-v2learn / 2026-04-27-eighth-lever-eval-and-observability Axiom #2Axiom #3Axiom #4Axiom #5Axiom #9Axiom #11Axiom #13Axiom #14 gvar-learning-repo

Partial graphs

Path A self-verify patch

#

The private GVAR workflow patch proved a topology gain: self-verify mode bypasses generation and reuses the same verifier input shape.

Ladder role
Workflow topology moves from generator dependence into explicit branch purity.
Evidence artifact
GVAR-38 workflow patch notes
Current checks
  • prior panel receipt
  • npm run build
Next proof
Add a public diagram and postmortem explaining the determinism gain.

Links

build / gvar-engine-v2learn / 2026-05-03-graph-constrained-execution Axiom #2Axiom #3Axiom #5Axiom #13 graph-workflow-convergence-repo

Partial agents

Three repos, one thesis essay

#

The follow-up essay ties learning-agent, evidence-agent, and gvar-engine into one architectural claim: determinism moves into loop boundaries, evidence contracts, and explicit graphs as job shape changes.

Ladder role
A repeated thesis moves from selection advice into runtime-shaped proof.
Evidence artifact
/learn/2026-05-05-three-repos-one-thesis
Current checks
  • npm run test:public-content
  • npm run test:public-style
  • npm run build
Next proof
Teach the portable agent pattern contract directly from the site and keep the repo-local MCP surfaces compliance-scannable.

Links

learn / 2026-05-05-three-repos-one-thesislearn / 2026-05-05-three-sdks-three-jobs Axiom #1Axiom #2Axiom #3Axiom #13Axiom #14Axiom #16 sdk-comparison-centerpiece-repo

Partial mcp

MCP ladder and evidence query

#

The public MCP now exposes receipt lists and per-item receipt lookups while existing content tools retain ladder placement metadata.

Ladder role
Agent readers navigate by rung, trade, and proof receipts.
Evidence artifact
stoneytech-site-mcp/src/mcp-server.ts
Current checks
  • npm --prefix stoneytech-site-mcp test
  • npm --prefix stoneytech-site-mcp run build
Next proof
Add live endpoint drift checks against deployed Cloudflare output.

Links

page / mcppage / proof-of-workbuild / public-content-mcp Axiom #2Axiom #5Axiom #9Axiom #21 tool:list_proof_receiptstool:get_proof_receipts_for_published_item

Planned

Backlog or repository concept exists before public release.

7

Planned mcp

Public content MCP clean-history repo

#

A future StoneyTECH public repo should package the read-only MCP, release manifest, negative-data contract, and smoke test from clean history.

Ladder role
Private implementation becomes a shareable learning package.
Evidence artifact
StoneyTECH org public repo plan
Current checks
  • release manifest pending
  • negative data contract pending
  • smoke test pending
Next proof
Create clean public repository after private history scrub.

Links

build / public-content-mcplearn / 2026-05-04-published-content-mcps Axiom #1Axiom #13Axiom #21 public-content-mcp-clean-history-repo

Planned evals

GVAR learning repo

#

A future repo should show a synthetic generate, verify, adjudicate, refine loop with citations to Aletheia and no private workflow leakage.

Ladder role
Verifier-loop learning becomes a clean public reference.
Evidence artifact
StoneyTECH org public repo plan
Current checks
  • README citations pending
  • synthetic fixtures pending
  • negative-data scrub pending
Next proof
Create clean public repo from synthetic fixtures and public citations.

Links

build / gvar-engine-v2 Axiom #4Axiom #11Axiom #13 gvar-learning-repo

Planned graphs

Graph workflow convergence repo

#

A future repo should show deterministic graph state, replay fixtures, branch gates, and convergence receipts using synthetic examples.

Ladder role
Workflow convergence becomes a public reference pattern.
Evidence artifact
StoneyTECH org public repo plan
Current checks
  • state-machine tests pending
  • replay fixture pending
  • convergence receipt pending
Next proof
Create the public graph workflow reference implementation.

Links

learn / 2026-05-03-graph-constrained-execution Axiom #2Axiom #3Axiom #5Axiom #13 graph-workflow-convergence-repo

Planned governance

Threat-surface matrix generator repo

#

A future CLI or small app should generate a layer-by-layer threat table with public OWASP and MITRE citation fixtures.

Ladder role
Threat modeling moves into a repeatable layer matrix.
Evidence artifact
StoneyTECH org public repo plan
Current checks
  • citation fixtures pending
  • snapshot tests pending
Next proof
Create a public generator with sample inputs and snapshots.

Links

learn / 2026-05-04-threat-surface-layer-by-layer Axiom #5Axiom #11Axiom #13Axiom #21 threat-surface-matrix-generator-repo

Planned governance

Deployment-context selector repo

#

A future selector should map residency, security, latency, cost, and operating constraints before model or provider choice.

Ladder role
Architecture context moves into an explicit decision tree.
Evidence artifact
StoneyTECH org public repo plan
Current checks
  • scenario fixtures pending
  • residency checks pending
  • cost checks pending
Next proof
Create a public selector with scenario fixtures.

Links

learn / 2026-05-11-deployment-context-firstlearn / 2026-04-27-model-portability-exceptions Axiom #1Axiom #13Axiom #21 deployment-context-selector-repo

Planned skills

Definition sidecar package

#

A future package or recipe should publish the glossary data shape, sidecar component pattern, accessibility rules, and public export checks.

Ladder role
Vocabulary support becomes a reusable site pattern.
Evidence artifact
StoneyTECH org public repo plan
Current checks
  • component tests pending
  • accessibility check pending
  • public glossary export pending
Next proof
Extract a clean package or documentation recipe.

Links

demystify / 2026-05-09-ai-ml-llm-agents-sorting-out-the-wordsdemystify / 2026-05-02-llms-as-a-loose-databasedemystify / 2026-05-03-tokens-context-attention-no-math Axiom #1Axiom #2Axiom #16 definition-sidecar-package

Planned governance

D1 graph maintenance receipt

#

A future ops receipt should cover D1 write reliability, compaction or rotation, and reconcile persistence for work claims.

Ladder role
Coordination reliability moves from implicit storage to monitored capacity.
Evidence artifact
coordination backlog plan
Current checks
  • work-claim write pending
  • reconcile persistence pending
Next proof
Close the D1 uniqueness drift and persist reconciliation evidence cleanly.

Links

page / proof-of-work Axiom #5Axiom #9Axiom #13

Maintenance Rule

Every merged public artifact updates the receipt inventory, adds a ladder role, links public evidence, names the current checks, and records the next proof gap.