Public surface scrub: personal framing only

LAUNCH_AUG3.md

@@ -1,7 +1,7 @@
-# Laguna Spark HF launch lock — 2026-08-03 (birthday)
+# Laguna Spark HF launch lock — 2026-08-03
 
 Status: **LOCKED**  
-Owner: Manwë (personal; not Ainfera / not Neptune)  
+Owner: personal (hizrianraz)  
 Repo: https://huggingface.co/hizrianraz/laguna-s-2.1-spark  
 Timezone: Asia/Jakarta (WIB)
 
@@ -13,7 +13,7 @@ Timezone: Asia/Jakarta (WIB)
 | Aug 2 18:00 WIB | Content freeze (no new claims without re-measure) |
 | Aug 3 09:00 WIB | Preflight |
 | Aug 3 12:00 WIB | **GO LIVE** (card + announce ready state) |
-| Aug 3 18:00 WIB | Post-launch verify + founder brief |
+| Aug 3 18:00 WIB | Post-launch verify + brief |
 
 ## Launch definition of done (all required)
 
@@ -26,17 +26,17 @@ Timezone: Asia/Jakarta (WIB)
    - Target: **≥38/40 (95%)** minimum ship
    - Stretch: **40/40** if fixed before freeze
 7. Gen throughput quote: ~21 tok/s @ 128 completion (or fresher re-measure)
-8. Clear disclaimer: personal · not Ainfera · not first quant · not Poolside/Nous affiliate
+8. Clear disclaimer: personal · not first quant · not Poolside/Nous affiliate
 9. No DIY GGUF unless it beats a scoreboard cell (default: **bind official only**)
 10. Launch receipt JSON written + card updated same day
 
 ## Non-goals (hard)
 
-- Ainfera / Neptune / Fin branding
+- Org / company product branding on this surface
 - Re-hosting official GGUF without measured win
-- K3 / other model side quests under this tag
+- Side-quest models under this tag
 - Fake 100% smoke or unverified tok/s
-- X/public brand push beyond HF card (rename gate still holds for X)
+- X/public brand push beyond HF card
 
 ## Daily clock (WIB)
 
@@ -54,14 +54,14 @@ Timezone: Asia/Jakarta (WIB)
 
 If smoke cannot reach 40/40 by freeze: ship **38/40 with named fails**.  
 If Spark down at launch: freeze last green MEASURED row; do not invent numbers.  
-If HF token dies: restore from Doppler `ainfera-ai/prd` `HUGGINGFACE_TOKEN` only; never paste token in chat.
+If HF token dies: restore the personal Hugging Face write token from the local secret store only; never paste token in chat.
 
-## Announce text (draft — founder edits OK)
+## Announce text (draft — edits OK)
 
 > Personal DGX Spark agent-runtime pack for Poolside Laguna-S-2.1: measured OpenAI-compatible serve, fixed agent smoke, official Q4_K_M digests (no bare re-upload).  
 > https://huggingface.co/hizrianraz/laguna-s-2.1-spark
 
 ## Authority
 
-Personal founder surface only. Manwë may push docs/results to this HF repo.  
-No Tier-1 actions. No company vault binding for this pack.
+Personal founder surface only. Docs/results may be pushed to this HF repo.  
+No company vault binding for this pack.

NOTICE

@@ -9,7 +9,7 @@ Upstream GGUF: poolside/Laguna-S-2.1-GGUF
 
 Original materials in this repository (SPARK.md measurement scaffolding,
 agent_smoke cases/runner, hermes sample client, scripts): personal work by
-hizrianraz for reproducibility. Not Ainfera, not Neptune, not Fin, not Gate0.
+hizrianraz for reproducibility. Independent personal project.
 
 Weight binaries are not re-hosted here unless a measured Spark-unique delta
 exists (policy S5). Prefer official GGUF download + sha256 verify via SHA256SUMS.

README.md

@@ -15,14 +15,13 @@ library_name: gguf
 pipeline_tag: text-generation
 ---
 
-# personal · not Ainfera · not Neptune · not Fin · not Gate0
+# Laguna-S-2.1 · DGX Spark pack (personal)
 
 **Launch target: 2026-08-03 (WIB)** — see [`LAUNCH_AUG3.md`](./LAUNCH_AUG3.md)
 
 Personal DGX Spark agent-runtime pack for [poolside/Laguna-S-2.1](https://huggingface.co/poolside/Laguna-S-2.1) (118B-A8B MoE, agentic coding).
 
-This is **not** an Ainfera product, not Neptune Core, not Fin-70B, and not Gate0.
-Unrelated to company org ships. Solo founder personal HF surface only.
+Solo personal HF surface by **hizrianraz**. Independent measurements on one DGX Spark. Not affiliated with Poolside or Nous Research.
 
 ## What this is
 
@@ -140,10 +139,10 @@ Snapshot JSONs: [`results/measured.json`](./results/measured.json) · [`results/
 
 ## What is **not** in this pack
 
-- No Ainfera routing, no Neptune staged gates, no Fin weights
-- No Verda/cloud CREATE for this workstream
+- No bare GGUF re-upload when the official file is the stand-behind artifact
+- No affiliate / org / company product claims
 - No fake benches — only fixed `agent_smoke` + `llama-bench` / server timings you can re-run
-- No spam GGUF re-upload when the official file is the stand-behind artifact (S5)
+- No off-Spark (rented cloud) numbers relabeled as Spark
 
 ## Reproducible method
 
@@ -158,4 +157,4 @@ See [`docs/REPRODUCE.md`](./docs/REPRODUCE.md) for exact clone SHAs, cmake flags
 
 ## Disclaimer
 
-Independent personal measurements on one DGX Spark. Not affiliated with, endorsed by, or representing Poolside, Nous Research, Hermes Agent product claims, or Ainfera Inc. product lines. “Hermes-class” means OpenAI-compatible tool-calling agent runtime shape only — **not** a Nous endorsement.
+Independent personal measurements on one DGX Spark. Not affiliated with, endorsed by, or representing Poolside or Nous Research. “Hermes-class” means OpenAI-compatible tool-calling agent runtime shape only — **not** a Nous endorsement.

SPARK.md

@@ -1,9 +1,9 @@
 # SPARK.md — Laguna-S-2.1 on DGX Spark (GB10)
 
-personal · not Ainfera · not Neptune · not Fin · not Gate0
+Personal DGX Spark notes only.
 
 Hardware target: **NVIDIA DGX Spark only** (GB10, Grace+Blackwell, ~128 GB unified).  
-Do not run this pack’s claimed numbers on Verda or rented cloud and relabel them.
+Do not run this pack’s claimed numbers on rented cloud hosts and relabel them as Spark.
 
 ## Engine pin (measured on this Spark)
 
@@ -214,6 +214,6 @@ If you measure vLLM on Spark, add a dated subsection here with image digest + to
 
 ## Forbidden host confusion
 
-- Do **not** attribute Verda/cloud jobs to this file.  
-- Do **not** mix Ainfera gateway base URLs into sample commands.  
-- Personal HF only for publish.
+- Do **not** attribute rented-cloud jobs to this file.  
+- Do **not** mix non-local / third-party gateway base URLs into sample commands.  
+- Personal HF only for publish (`hizrianraz/*`).

docs/BUILD_SPARK.md

@@ -1,6 +1,6 @@
 # Build notes — poolside llama.cpp laguna on DGX Spark
 
-personal · not Ainfera · not Neptune · not Fin · not Gate0
+Personal Spark build notes.
 
 ## Engine
 
@@ -24,7 +24,7 @@ Measured fix that builds clean:
 if (!std::isfinite(v)) {
 
 // after (Spark / GNU 13.3)
-if (!::isfinite(static_cast<double>(v))) /* manwe-isfinite-fix */ {
+if (!::isfinite(static_cast<double>(v))) /* spark-isfinite-fix */ {
 ```
 
 Do **not** treat a failed UI-asset HF fetch during build as fatal — `llama-server` still links; only the optional embedded UI may stay stale.

docs/REPRODUCE.md

@@ -1,6 +1,6 @@
 # Reproducible method — Laguna-S-2.1 Spark pack
 
-personal · not Ainfera · not Neptune · not Fin · not Gate0
+Personal reproducible method.
 
 ## Pins (card freeze)
 
@@ -39,9 +39,9 @@ sha256sum -c SHA256SUMS
 
 ## Personal HF publish surface
 
-- User: `hizrianraz` (personal). Not `ainfera-ai` without separate founder Go.
-- Suggested repo id: `hizrianraz/laguna-s-2.1-spark`
-- Line 1 of card must remain: `personal · not Ainfera · not Neptune · not Fin · not Gate0`
+- User: `hizrianraz` (personal)
+- Repo id: `hizrianraz/laguna-s-2.1-spark`
+- Card framing: personal measurements only; no org product claims
 
 ## DIY quant policy
 

eval/agent_smoke/cases.json

@@ -1575,7 +1575,7 @@
         },
         {
           "role": "user",
-          "content": "Remember project codename. What file should hold README for ORCA? Just path sandbox/ORCA/README.md \u2014 if you must use a tool, list_dir sandbox first."
+          "content": "Remember project codename. What file should hold README for ORCA? Just path sandbox/ORCA/README.md — if you must use a tool, list_dir sandbox first."
         }
       ],
       "tools": [
@@ -1627,8 +1627,8 @@
           "write_file"
         ],
         "content_forbids": [
-          "Neptune",
-          "Ainfera"
+          "PoseidonLabs",
+          "ACMECorp"
         ]
       },
       "notes": ""
@@ -1799,7 +1799,7 @@
       "messages": [
         {
           "role": "system",
-          "content": "You are a careful tool-using coding agent. Only call tools from the provided schema. Never invent tool names. When a tool is needed, emit proper tool calls. Prefer concise answers. You must not claim Ainfera or Neptune branding."
+          "content": "You are a careful tool-using coding agent. Only call tools from the provided schema. Never invent tool names. When a tool is needed, emit proper tool calls. Prefer concise answers. You must not invent org or product branding. Keep the pack label personal spark pack only."
         },
         {
           "role": "user",

eval/agent_smoke/run_smoke.py

@@ -2,7 +2,7 @@
 """Fixed agent_smoke runner for Laguna OpenAI-compatible servers.
 
 Pass/fail only against the pinned cases.json — no free-form grading.
-personal · not Ainfera · not Neptune · not Fin · not Gate0
+personal fixed smoke runner
 """
 
 from __future__ import annotations

hermes/README.md

@@ -1,6 +1,6 @@
 # Hermes-class sample client
 
-personal · not Ainfera · not Neptune · not Fin · not Gate0
+Personal sample client for local Spark serve.
 
 OpenAI-compatible client notes for **Hermes-class / tool-agent** runtimes against
 local `llama-server` serving Laguna-S-2.1 on DGX Spark.
@@ -11,7 +11,7 @@ local `llama-server` serving Laguna-S-2.1 on DGX Spark.
   + `tools` / `tool_calls` shape common to agent stacks in that family.
 - **Not** a Nous Research endorsement.
 - **Not** Hermes Agent source, branding, or product affiliation.
-- **Not** Ainfera gateway.
+- **Not** a third-party gateway product.
 
 ## Connect
 

hermes/sample_client.py

@@ -2,7 +2,7 @@
 """Hermes-class / tool-agent compatible sample client for local Laguna serve.
 
 Wording: OpenAI-compatible tool-calling shape used by Hermes-class agent runtimes.
-This is NOT a Nous Research endorsement, not Hermes Agent product code, and not Ainfera.
+This is NOT a Nous Research endorsement and not Hermes Agent product code.
 
 Usage:
   export OPENAI_BASE_URL=http://127.0.0.1:8000/v1

results/MEASURED.md

@@ -1,6 +1,6 @@
 # Laguna-S-2.1 on DGX Spark — measured 2026-07-28
 
-personal · not Ainfera · not Neptune
+personal Spark measurement note
 
 ## Artifacts
 - Model: poolside official `laguna-s-2.1-Q4_K_M.gguf` (90G)

results/agent_smoke.log

@@ -0,0 +1,42 @@
+PASS tool_json_01 (tool_json) ok
+PASS tool_json_02 (tool_json) ok
+PASS tool_json_03 (tool_json) ok
+PASS tool_json_04 (tool_json) ok
+PASS tool_json_05 (tool_json) ok
+PASS tool_json_06 (tool_json) ok
+PASS tool_json_07 (tool_json) ok
+PASS tool_json_08 (tool_json) ok
+PASS multi_01 (multi_step) ok
+PASS multi_02 (multi_step) ok
+PASS multi_03 (multi_step) ok
+PASS multi_04 (multi_step) ok
+PASS multi_05 (multi_step) ok
+PASS multi_06 (multi_step) ok
+PASS multi_07 (multi_step) ok
+PASS multi_08 (multi_step) ok
+PASS repair_01 (error_repair) ok
+PASS repair_02 (error_repair) ok
+PASS repair_03 (error_repair) ok
+FAIL repair_04 (error_repair) HTTP 500: b'{"error":{"code":500,"message":"Failed to parse tool call arguments 
+PASS repair_05 (error_repair) ok
+PASS repair_06 (error_repair) ok
+PASS no_invent_01 (no_invented_tools) refused without inventing
+PASS no_invent_02 (no_invented_tools) refused without inventing
+PASS no_invent_03 (no_invented_tools) refused without inventing
+PASS no_invent_04 (no_invented_tools) refused without inventing
+PASS no_invent_05 (no_invented_tools) refused without inventing
+PASS no_invent_06 (no_invented_tools) refused without inventing
+PASS code_01 (short_code) ok
+PASS code_02 (short_code) ok
+PASS code_03 (short_code) ok
+PASS code_04 (short_code) ok
+PASS code_05 (short_code) ok
+PASS code_06 (short_code) ok
+PASS long_01 (long_horizon) ok
+PASS long_02 (long_horizon) ok
+PASS long_03 (long_horizon) ok
+PASS long_04 (long_horizon) ok
+PASS long_05 (long_horizon) ok
+FAIL long_06 (long_horizon) KeyError: 'tool'
+wrote results/agent_smoke.json
+SUMMARY pass=38/40 rate=0.95 elapsed=96.66s

results/hf_publish.json

@@ -24,5 +24,9 @@
     "scripts/pull_official_gguf.sh",
     "scripts/serve_spark.sh"
   ],
-  "private": false
+  "private": false,
+  "launch_locked": true,
+  "launch_at_wib": "2026-08-03T12:00:00+07:00",
+  "latest_commit": "bebc09aa213eda2ff1105800f322aefbfb7f57cb",
+  "measured_json_on_hf": true
 }

scripts/pull_official_gguf.sh

@@ -12,7 +12,7 @@ mkdir -p "${DEST}"
 cd "${DEST}"
 echo "pull ${URL}"
 curl -L --fail --retry 8 --retry-delay 15 -C - \
-  -A 'manwe-laguna-pack' \
+  -A 'hizrianraz-laguna-spark-pack' \
   -o "${FILE}" \
   "${URL}"
 sha256sum "${FILE}" | tee "sha256_${FILE}.txt"

scripts/serve_spark.sh

@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
 # Serve official Laguna-S-2.1 GGUF on DGX Spark via poolside llama.cpp (laguna).
-# personal · not Ainfera · not Neptune · not Fin · not Gate0
+# personal DGX Spark serve helper
 set -euo pipefail
 
 ROOT="${LAGUNA_ENGINE:-$HOME/src/llama.cpp-laguna}"