docs: publish Chaski card from szl-forge@620f68b318a1757d8735f6a4d09cd1f42d68a136

Exact-source Chaski card publication. Named-N remains a measured failed gate, publication_eligible and autonomy_eligible remain false, and no model or provider state is promoted. No weights, adapter, configs, evals, visibility, hardware, collection, or runtime state changed.

README.md

@@ -1,5 +1,5 @@
 ---
-thumbnail: https://huggingface.co/SZLHOLDINGS/chaski/resolve/main/og-card.png
+thumbnail: https://huggingface.co/SZLHOLDINGS/chaski/resolve/main/holo-banner.svg
 license: apache-2.0
 language:
 - en
@@ -9,10 +9,10 @@ base_model: Qwen/Qwen3.5-0.8B
 base_model_relation: finetune
 tags:
 - szl-holdings
-- series-a
 - doctrine-v11
 - governed-ai
 - proposal-only
+- research
 szl:
   doctrine: v11-LOCKED
   lean: 749/14/163
@@ -20,10 +20,6 @@ szl:
   evidence_ceiling: 0.97
   artifact_class: MERGED_FINETUNE
   originality: FINETUNE_DISCLOSED_BASE
-  collection: SZL Fall 2026 — Original Cuts
-  jobs: COMPLETED
-  job_id: 6a91bf1045686a1580c12105
-  job_namespace: SZLHOLDINGS
   weights: AVAILABLE
   evals: MEASURED
   named_n:
@@ -37,16 +33,9 @@ szl:
     report_sha256: 4d057eb9867285e69b00222be110bbb660330a96fe7b284a4d7f488268a13e05
     report_bytes: 3996
     report_commit: db71c243d0176bccff1ff087cd4dd57663bd6502
-    method: in-process greedy generate, messages[:-1], transformers 5.16.1 bf16 CPU,
-      load_in_4bit=False; live Chaski merged shard 1c55df8; PR 63 gates
   publication_eligible: false
+  autonomy_eligible: false
   gpu: UNAVAILABLE
-  job_prior_error: 6a91bb7c984507d9db4ea0a4
-  job_completed_no_weights: 6a91ba00984507d9db4ea07f
-  job_prior_failed:
-  - 6a91b8ba984507d9db4ea071
-  - 6a91b990984507d9db4ea077
-  - 6a91bb7c984507d9db4ea0a4
 ---
 
 <p align="center">
@@ -57,132 +46,91 @@ szl:
 
 <p align="center"><em>The courier that cannot invent the dispatch.</em></p>
 
-<p align="center">
-  <img alt="License: Apache-2.0" src="https://img.shields.io/badge/License-Apache--2.0-8b5cf6?style=flat-square"/>
-  <img alt="Downloads" src="https://img.shields.io/huggingface/dt/SZLHOLDINGS/chaski?style=flat-square&color=22d3ee&label=downloads"/>
-  <img alt="Base: Qwen3.5-0.8B" src="https://img.shields.io/badge/base-Qwen3.5--0.8B-334155?style=flat-square"/>
-  <img alt="Doctrine v11-LOCKED" src="https://img.shields.io/badge/doctrine-v11--LOCKED-0f172a?style=flat-square"/>
-  <img alt="Named-N: MEASURED FAIL" src="https://img.shields.io/badge/Named--N-MEASURED%20FAIL-dc2626?style=flat-square"/>
-</p>
-
-<p align="center">
-  <code>KANCHAY</code> · Lean <code>749/14/163</code> · Λ = Conjecture 1 (advisory, never a theorem) · <a href="https://a-11-oy.com">a-11-oy.com</a>
-</p>
-
----
+> **RESEARCH / NEGATIVE EVIDENCE. Failed qualification. Not flagship.**
+> The current Named-N run failed both release gates. This card does not promote
+> the model, its later SKUs, or any live endpoint.
 
-> **RESEARCH / NEGATIVE EVIDENCE.** Failed qualification. Not flagship.
-> Later SKU `A11OY-MINI` inherits this failed parent. Not a product claim.
+## Contract
 
-> **Metadata correction 2026-08-30.** This repo declared
-> `base_model_relation: adapter` and `artifact_class: ADAPTER` while also declaring
-> `weights: AVAILABLE` and shipping a 1.75 GB merged shard. Relative to
-> `Qwen/Qwen3.5-0.8B` this is a **finetune**, not an adapter — corrected to
-> `base_model_relation: finetune` / `artifact_class: MERGED_FINETUNE`. The card's own
-> `originality: FINETUNE_DISCLOSED_BASE` already said so. No eval, weight or receipt
-> claim is touched; the MEASURED failing gate below stands exactly as it was.
+Chaski is an SZL fine-tune of the disclosed Apache-2.0 base
+[`Qwen/Qwen3.5-0.8B`](https://huggingface.co/Qwen/Qwen3.5-0.8B). It produces
+proposal-only drafts behind a validating controller. The model does not execute
+actions, authorize mutations, or convert uncertainty into a fact.
 
-## One line
+**No uniqueness claim is made.** The value of this artifact is its explicit
+controller boundary, disclosed lineage, and retained negative evidence.
 
-Messenger LLM. Proposal-only drafts and honest refusals for the SZL controller. Adapters are on this repo. Named-N evals MEASURED `json_draft` 0/5, `adversarial_refusal` 2/6 — not a pass. Not publication-eligible.
-
-## The cut
-
-Multimodal models narrate what they see. Chaski may only carry a payload the controller already signed. Vision without authorship.
-
-A courier that cannot invent the dispatch. The oldest job in the Andes, as a LoRA.
-
-| Leader | Take, then tweak |
+| Field | Verified statement |
 |---|---|
-| Anthropic | Claude vision, minus the right to conclude. |
-| NVIDIA | NVLM / NeMo multimodal, minus the right to act. |
-| Unsloth | Adapter on Qwen3.5-0.8B. |
-
-Nobody else ships this combination. That is the point of a one-of-one.
-
-Canonical GitHub: [`szl-holdings/szl-forge`](https://github.com/szl-holdings/szl-forge/blob/main/chaski/)
+| Artifact | Merged fine-tune plus LoRA artifacts are present on the Hub repository |
+| Parent/eval revision | `1c55df8652e9d0f7b84356b1e2d54849165ae884` |
+| Base relation | `finetune`; `artifact_class: MERGED_FINETUNE` |
+| Originality | `FINETUNE_DISCLOSED_BASE` |
+| Publication eligibility | `false` |
+| Autonomy eligibility | `false` |
+| Lambda | Conjecture 1, advisory and never a theorem |
+| Trust ceiling | `0.97` |
+
+Canonical source:
+[`szl-holdings/szl-forge/chaski/card`](https://github.com/szl-holdings/szl-forge/tree/main/chaski/card).
+Exact publication source and asset hashes are recorded in
+`szl-source-binding.json` on this model repository.
 
-## Load it right
-
-Three loadable artifacts sit in this repository at once — a bare `from_pretrained` on the repo id is ambiguous. Be explicit about which one you want.
-
-| Want | Load |
-|---|---|
-| the merged finetune | `AutoModelForCausalLM.from_pretrained("SZLHOLDINGS/chaski")` — resolves the merged shard via `model.safetensors.index.json` |
-| the LoRA adapter on the stock base | base `Qwen/Qwen3.5-0.8B` + `PeftModel.from_pretrained(base, "SZLHOLDINGS/chaski")` |
-| the Unsloth-base adapter | base `unsloth/...-bnb-4bit` + PEFT pointed at the `adapter-unsloth` subfolder |
-
-Three sharp edges worth knowing before you debug them:
-
-- **Both `config.json` and `adapter_config.json` live at the repo root.** With `peft` installed, some loader paths follow the adapter rather than the merged weights. Pass the path you mean rather than relying on precedence.
-- **The merged shard has a non-standard filename**, `model.safetensors-00001-of-00001.safetensors` instead of the conventional `model-00001-of-00001.safetensors`. It resolves through the index file, but tooling that globs the conventional pattern will not find it.
-- **`config.json` is `Qwen3_5ForConditionalGeneration` and carries a `vision_config`;** visual weights are present in the index. If you only want text, that tower still loads.
+## Evaluation
 
-## Specification
+**Named-N: MEASURED FAIL.** The measured run on 2026-08-28 is a failed gate, not a passing benchmark.
 
-| | |
-|---|---|
-| **Artifact** | `adapter_model.safetensors` + merged ~1.7GB shard **AVAILABLE** |
-| **Parent / eval revision** | `1c55df8652e9d0f7b84356b1e2d54849165ae884` |
-| **Originality** | SZL fine-tune of a disclosed Apache Qwen instruct base |
-| **Base** | `Qwen/Qwen3.5-0.8B` (Apache-2.0, 0.6B–2B lock). Not an Unsloth-default card. |
-| **License** | `apache-2.0` |
-| **HF Jobs** | Attempt 5 **COMPLETED** [`6a91bf1045686a1580c12105`](https://huggingface.co/jobs/SZLHOLDINGS/6a91bf1045686a1580c12105) (`report_to=none`). Tensors on Hub. |
-| **Named-N** | MEASURED fail. `json_draft` **0/5**. `adversarial_refusal` **2/6**. Not a pass. |
-| **Receipt** | `eval_report.json` 3996 bytes sha256 `4d057eb9867285e69b00222be110bbb660330a96fe7b284a4d7f488268a13e05` (commit `db71c24`) |
-| **Method** | in-process greedy generate, `messages[:-1]`, transformers 5.16.1 bf16 CPU, `load_in_4bit=False`. Date 2026-08-28. File `eval_report.json`. PR 63. |
-| **Quality** | ROADMAP (prose). Failed gate is not a publish. |
-| **Status** | CUTTING |
-| **Lab** | House CPU lab stays **Khipu GGUF**. Lab load forbidden for Chaski. |
-| **Later SKU** | [`A11OY-MINI`](https://huggingface.co/SZLHOLDINGS/A11OY-MINI) GGUFs exist on that repo. They do **not** inherit this Named-N. Mini stays evals none-this-run. |
-| **Sibling** | [`szl-receiptagent-qwen35-0.8b-v2`](https://huggingface.co/SZLHOLDINGS/szl-receiptagent-qwen35-0.8b-v2) |
+| Probe | Result | Interpretation |
+|---|---:|---|
+| JSON draft | **0/5** | failed |
+| Adversarial refusal | **2/6** | failed |
+| Hallucinated citations | **0 observed in this bounded run** | not a general guarantee |
 
-## Evaluation
+Receipt: `eval_report.json`, 3996 bytes, SHA-256
+`4d057eb9867285e69b00222be110bbb660330a96fe7b284a4d7f488268a13e05`,
+source commit `db71c243d0176bccff1ff087cd4dd57663bd6502`.
 
-**Named-N MEASURED fail** on live Chaski `1c55df8` (2026-08-28). Receipt: [`eval_report.json`](https://huggingface.co/SZLHOLDINGS/chaski/blob/main/eval_report.json) sha256 `4d057eb9867285e69b00222be110bbb660330a96fe7b284a4d7f488268a13e05`.
+Method: in-process greedy generation over `messages[:-1]`, Transformers 5.16.1,
+bf16 CPU, `load_in_4bit=False`. This evidence does not transfer to
+`A11OY-MINI`, another revision, another runtime, or another prompt set.
 
-| Probe | N | Score | Label |
-|---|---|---|---|
-| `json_draft` | 5 | **0/5** | MEASURED fail |
-| `adversarial_refusal` | 6 | **2/6** | MEASURED fail |
+## Loading
 
-Method: in-process greedy generate, `messages[:-1]`, transformers 5.16.1 bf16 CPU, `load_in_4bit=False` on live Chaski `1c55df8` (2026-08-28). File: `eval_report.json`. PR 63. What this is NOT: a passing eval gate, a published score, an A11OY-MINI eval, or a 5/5. `publication_eligible: false`.
+The repository contains multiple loadable forms. Select the intended artifact
+explicitly rather than relying on loader precedence.
 
-`train_loss` MEASURED `1.783925924450159` is a train metric, not an eval.
+```python
+from transformers import AutoModelForCausalLM
 
-## Training
+model = AutoModelForCausalLM.from_pretrained(
+    "SZLHOLDINGS/chaski",
+    trust_remote_code=False,
+)
+```
 
-- **Recipe:** Unsloth QLoRA SFT. Script: `train_chaski.py`. Loads **only** `szl_dataset.jsonl`.
-- **1–2 FAILED:** `6a91b8ba984507d9db4ea071` CastError; `6a91b990984507d9db4ea077` pyyaml 30s timeout.
-- **3 COMPLETED receipt-only:** [`6a91ba00984507d9db4ea07f`](https://huggingface.co/jobs/SZLHOLDINGS/6a91ba00984507d9db4ea07f).
-- **4 ERROR, not RUNNING:** [`6a91bb7c984507d9db4ea0a4`](https://huggingface.co/jobs/SZLHOLDINGS/6a91bb7c984507d9db4ea0a4) Trackio 404. Files persisted.
-- **5 COMPLETED `report_to=none`:** [`6a91bf1045686a1580c12105`](https://huggingface.co/jobs/SZLHOLDINGS/6a91bf1045686a1580c12105). Tensors now on Hub.
-- **Trackio:** 404 `betterwithage/trackio-bucket`. No dashboard URL.
-- **publication_eligible:** false
+For the LoRA path, load `Qwen/Qwen3.5-0.8B` first and attach the adapter with
+PEFT. The root contains both `config.json` and `adapter_config.json`; callers
+must verify which path their library selects. The merged shard uses a
+non-standard filename resolved through its index.
 
 ## Intended use
 
-- **Who:** a11oy / Alloy controller, not an end-user chatbot
-- **What:** JSON drafts (`decision=DRAFT`, `approvalRequired=true`, `executed=false`) and doctrine-faithful UNKNOWN — carry signed payloads across organs
-- **Where:** behind a validating controller. The weights propose. The controller gates.
+- Proposal-only JSON drafts with `approvalRequired=true` and `executed=false`.
+- Doctrine-faithful abstention and routing experiments.
+- Research behind an independent validator and human approval boundary.
 
-## What it is NOT
+## Prohibited interpretation
 
-- Not an autonomous agent, executor, factual oracle, or weapon.
-- Not a Qwen rehost.
-- Not the live lab model. Not a tokens/s claim.
-- Not publication-eligible on this MEASURED run.
-- Not a 5/5 or 6/6. Do not read 2/6 as a pass.
-- Not a VLM product card without a signed vision eval.
+- Not an autonomous agent, executor, factual oracle, or weapons system.
+- Not a passing 5/5 or 6/6 release.
+- Not evidence of a deployed Alloy endpoint.
+- Not evidence that a later SKU inherits this evaluation.
+- **Lab load forbidden for Chaski.**
 
 ## Limitations
 
-- Proposal-only. Narrow curriculum. Controller required.
-- Λ = Conjecture 1. Trust ceiling 0.97. CUTTING.
-- This MEASURED run failed Named-N. Do not ship as a pass.
-
----
+Narrow curriculum, small bounded evaluation, CPU-only measured run, and failed
+release gates. A new model or runtime revision requires a new immutable
+evaluation receipt; this card cannot confer approval.
 
-<p align="center">
-  <sub><strong>Fashion rule.</strong> Silhouette from Qwen3 / Qwen3.5 instruct. Cut is original SZL. We do not republish someone else's tensors.</sub>
-</p>
+Doctrine v11 LOCKED. Λ = Conjecture 1. Owner: Stephen Lutar / SZL Holdings.

szl-source-binding.json

@@ -21,8 +21,8 @@
   "source": {
     "assets": {
       "README.md": {
-        "bytes": 9287,
-        "sha256": "f2a2e434fb0c628214881f9ce0c6a86f9c570d34dd69a727597ce14a3c403e43"
+        "bytes": 4695,
+        "sha256": "fbd3f10904e5066adda55054ec88aca9002d79669c2d138935de2008a8db51d6"
       },
       "holo-banner.svg": {
         "bytes": 1374,
@@ -30,7 +30,7 @@
       }
     },
     "repository": "szl-holdings/szl-forge",
-    "revision": "a64a35969218bf0735b24249e6580ed9ec954b43"
+    "revision": "620f68b318a1757d8735f6a4d09cd1f42d68a136"
   },
   "target": {
     "controlled_paths": [