---
vault_clearance: KETER
halo:
  classification: RESTRICTED
  confidence: HIGH
  front: "38_Project_FIREWALL — replication bounties for IP-clean re-derivation"
  custodian: "Jixiang Leng"
  created: 2026-05-03
  wing: READY
  containment: "Operator IP-firewall execution checklist. Update statuses as bounties complete. Each replication SOLVED gets a WORLDLINE entry with verification receipts."
---

# FIREWALL — Bounty Board

Replication bounties for every NIH-tainted derivation in the vault that needs to ship as commercial / patent / external-IP claim. Each bounty has the full verification protocol from [HALO_FIREWALL_DOCTRINE.md §4](HALO_FIREWALL_DOCTRINE.md). For inventory + provenance see [HALO_REPLICATION_TARGETS.md](HALO_REPLICATION_TARGETS.md).

**Status legend:** `OPEN` (not started) · `IN PROGRESS` (replication running) · `BLOCKED` (waiting on dependency) · `SOLVED` (verified, OpenTimestamps anchored) · `DEFERRED` (postponed; see notes) · `NOT_NEEDED` (already clean by doctrine §2)

**Bounty ID prefix:** `FW-` for FIREWALL.

---

## Tier 0 — Infrastructure setup (this week, ~$200-400 + ~6 hours operator-time)

Without these, no replication can run. Personal-compute isolation is the foundation.

| ID | Title | Status | Cost | Time | Notes |
|---|---|---|---|---|---|
| **FW-I1** | **Set up operator-personal GCP project** with separate billing-account-ID. Verify: project NOT in `desynchronization-theory` org; billing → operator-personal credit card. | OPEN | $0 setup | 30 min | New project name suggestion: `firewall-clean-room` or operator's preferred. Enable Compute Engine + Cloud Storage + Vertex AI APIs. Set IAM such that operator is sole owner. |
| **FW-I2** | **Spin up FIREWALL VM** — n2-highmem-32 (256 GB RAM, 32 vCPU) on personal-billed GCP. Confirm SSH from operator-personal hardware ONLY. Persistent SSD 2 TB. | BLOCKED on FW-I1 | ~$1.50/hr running; ~$340/mo storage | 30 min spin-up | Boot Ubuntu 22.04+; install conda + Docker + sra-toolkit + cellranger + samtools + pysam + pandas + scvelo + scanpy. **STOP the VM when not actively running** to avoid $1k+/month idle. |
| **FW-I3** | **Personal AWS account verification** (alt route if GCP isolation feels mixed) | OPEN — optional | $0 setup | 15 min | If operator has separate personal AWS, this is the lowest-NIH-cross-contamination compute option. Otherwise stick with FW-I1 GCP. |
| **FW-I4** | **Download GRCh38 + GENCODE + cellranger refs** to FIREWALL VM | BLOCKED on FW-I2 | ~$2 storage cost | 1 hour | `cellranger refdata-gex-GRCh38-2024-A` (~11 GB); GRCh38 primary FASTA (~3 GB); GENCODE v44+ GTF (~50 MB); RepeatMasker GRCh38 BED (~500 MB). Hash and store paths. |
| **FW-I5** | **Build verification harness** — automated check that confirms (a) no NIH-VM in compute graph, (b) no NIH-cohort accession in inputs, (c) outputs are deterministic from public-data inputs | OPEN | $0 | 4-8 hours | Tool: `firewall_verify.py` — takes a replication output bundle, asserts compute-isolation + data-source provenance + tool-version pinning. Fail-fast on any contamination. Commit to operator-personal GitHub. |
| **FW-I6** | **OpenTimestamps stamping pipeline** for replication outputs | OPEN | $0 | 1 hour | Reuse `scripts/ots_stamp` from `05_Project_LENG`. Set up auto-stamping cron on FIREWALL VM to anchor every output's parquet/JSON within 10 min of creation. |
| **FW-I7** | **Personal-compute receipt format** — agreed schema (what receipts look like) | OPEN | $0 | 1 hour | Schema: GCP invoice + IAM dump + VM ops log + pip freeze + accession list + SHA256 + .ots proofs. Document in `38_Project_FIREWALL/RECEIPT_SCHEMA.md`. |

**Critical path:** FW-I1 → FW-I2 → FW-I4 → first replication. Total setup: ~3-4 hours operator + ~2 hours compute.

---

## Tier 1 — Splicing thread (35_TheHats / BT104+105+106) — highest commercial-IP priority

Unlocks VitaDAO submission + senolytic-timing-biomarker patent provisional.

**FW-S3, FW-S4, FW-S5 status update (2026-05-04)**: First-pass replication landed on **D03 Calandrelli HUVEC** (CellxGene Census, 59,605 cells, glucose+TNF metabolic-stress time-course) — operator-personal compute, fully Zone-1-clean. **All 4 success criteria PASSED.** See [replications/calandrelli_d03_huvec/MEMO_2026-05-04_FULL_REPLICATION.md](replications/calandrelli_d03_huvec/MEMO_2026-05-04_FULL_REPLICATION.md) for the full numbers + caveats.

The D03 replication validates the framework on a complementary dataset: different cell type (HUVEC vs WI-38), different induction (metabolic stress + inflammation vs etoposide DNA damage). The original GSE226225 path (planned in this bounty board) is still the canonical WI-38 + etoposide replication — running it gives **two independent replications on two distinct induction biologies**, which is the strong-evidence-grade claim.

**Status of FW-S1..S5 below reflects the D03 first-pass replication; GSE226225 path remains as confirmation.**

| ID | Title | Status | Public dataset | Time | Notes |
|---|---|---|---|---|---|
| **FW-S1** | Replicate BT102 structure-first reading on public WI-38 timecourse | BLOCKED on FW-I4 | GSE226225 (etoposide day 0→10) + Tabula Sapiens EC reference | 4-6 hours compute | Verify: EC-specific senescence progression observable on public data; Δ pseudotime in EC vs other lineages reproduces. |
| **FW-S2** | Replicate BT103 7-axis multi-dim PCA on GSE226225 | BLOCKED on FW-S1 | GSE226225 + GSE250041 | 2-4 hours | Verify: ≥7 PCs >5% variance; Module 1 senescence-network + Module 2 proliferation modules emerge. SRSF3/PTBP1 singletons surface. |
| **FW-S3** | **Replicate BT104 peak-and-release biology** — Module 1 IR PEAKS at cluster 5, crashes at cluster 6 | **PARTIAL — D03 PASSED 2026-05-04**; GSE226225 confirmation pending FW-S2 | D03 Calandrelli (DONE) + GSE226225 (planned) | 2-4 hours each | D03 verdict: peak at archetype 9, crash at archetype 6, EC-identity TFs ~2× higher at pre-loading. ✅ Direction correct on independent cell type + induction. GSE226225 etoposide path still planned for canonical WI-38 confirmation. |
| **FW-S4** | **Replicate BT105 NEAT1+MALAT1 paraspeckle dissolution** — 3× / 2× induction at cluster 5; 64× / 137× drop at cluster 6 | **PARTIAL — D03 PASSED 2026-05-04**; GSE226225 confirmation pending | D03 Calandrelli (DONE) + GSE226225 (planned) | 2 hours each | D03 verdict: NEAT1 ~4.5× pre vs term, MALAT1 ~4.8× pre vs term — direction reproduces, magnitude on D03 stronger than NIH cohort prediction. Crash magnitude on D03 (~5×) smaller than NIH cohort (64-137×) — likely induction-method difference. GSE226225 confirmation will resolve magnitude question. |
| **FW-S5** | **Replicate BT106 dark-transcript ecosystem** — PVT1, ENSG00000259345, ENSG00000251314, MIR31HG, etc. peak-and-release pattern | **PARTIAL — D03 PASSED 2026-05-04**; GSE226225 confirmation pending | D03 Calandrelli (DONE) + GSE226225 (planned) | 2 hours each | D03 verdict: PVT1 + ENSG00000259345 found, both show peak-and-release direction. ENSG00000251314 + MIR31HG not in D03's gene set — confirm on GSE226225. PVT1 magnitude weak in absolute (HUVEC vs cohort's HAEC+PBMC); direction correct. |
| **FW-S6** | **GNRA hairpin scan** (already clean — sequence-only on GRCh38) | OPEN | Public GRCh38 + RepeatMasker | 30 min | **Verification gate:** PVT1 carries 49 ± 5 strict 6bp-perfect GNRA hairpins on personal-compute re-run. ENSG00000251314 carries 79 ± 5. Re-run for receipt. |
| **FW-S7** | **`monotone_glm.py` re-run on public data** — produce GLM coefficients on GSE226225 | BLOCKED on FW-I4 | GSE226225 | 1-2 hours | **Verification gate:** SERPINE1, RALA, RB1, GLB1, CXCL1/8 show negative slopes (release-leg dominant) on public data, reproducing the BT104 finding. |
| **FW-S8** | **`staff_tomography.py` family re-run on public BAMs** | BLOCKED on FW-I4 | GSE226225 BAMs | 4-6 hours | Tools are operator-built (clean tier); re-run produces clean output receipts. |
| **FW-S9** | **HALO_STRUCTURE_FIRST_VELOCITY.md split-restate** — separate clean framework section from tainted results section | BLOCKED on FW-S3 through FW-S5 | n/a | 2-3 hours operator | Output: clean-framework HALO + lab-data appendix HALO. Clean version becomes Sovereign-shippable; appendix stays vault-internal. |
| **FW-S10** | **Cross-validation on GSE131907** Korean NSCLC EC subset | BLOCKED on FW-S5 | GSE131907 (208,506 cells, 44 patients) | 4-8 hours | **Optional but high-value:** EC-specific replication in disease context. Reproduces peak-and-release in cancer-EC subset. |
| **FW-S11** | **Replication receipt package + Sovereign WORLDLINE entry** | BLOCKED on FW-S3, S4, S5, S6, S7 | n/a | 2-4 hours operator | Final deliverable: clean-replication tarball + WORLDLINE entry confirming IP-firewall-clean for splicing thread. **Unblocks Sovereign SOV-1b (patent provisional)** and **VitaDAO proposal submission**. |

**Critical path (splicing thread):** FW-I4 → FW-S1 → FW-S2 → FW-S3 → FW-S4 → FW-S5 → FW-S11. **Total: ~3-5 days from clean start.**

> **Note 2026-05-04 — out-of-scope work moved out:** Initially this board carried FW-S12 (Cai Lei load-vs-release on Pineda + Yadav). That was a misclassification: Pineda + Yadav are NOVEL framework deployments on datasets the operator has never analyzed before, not replications of NIH-cohort findings. They've been moved to **[10_Project_DiscordIntoSymphony/experiments/als_load_vs_release_2026-05-04/](../10_Project_DiscordIntoSymphony/experiments/als_load_vs_release_2026-05-04/)** where the K_RG framework + novel cellular-state archetype work lives. 38_FIREWALL stays focused on its actual job: cleaning NIH-tainted derivations on public data so they can ship as commercial IP. The IP-hygiene guards (`assert_personal_storage`, `assert_personal_compute`) carry over to the moved work because they're general practice, not 38-specific. The rotenone DA neuron memo at [replications/rotenone_DA_neurons/MEMO.md](replications/rotenone_DA_neurons/MEMO.md) **stays in 38** because it tested transfer of the NIH-derived peak-and-release pattern (BT104+105+106) — that IS replication-adjacent work.

---

## Tier 2 — Recycler / Fungal cell (28_RedFromTheGrave) — second commercial-IP priority

Unlocks chitin-high persister-cancer biomarker panel + cancer-subtype-stratification consulting + parallel VitaDAO/Astera proposals.

| ID | Title | Status | Public dataset | Time | Notes |
|---|---|---|---|---|---|
| **FW-R1** | Replicate Phase 1 cross-kingdom voltage / trafficking conservation | BLOCKED on FW-I4 | Multi-organism public scRNA-seq + ortholog DBs | 1-2 days | Yeast (GSE-wide), C. neoformans, mammalian senescent cohorts. Verify 8/8 family directional concordance. |
| **FW-R2** | Replicate Phase 2 opisthokont-ancestral persister-trafficking gene-level conservation | BLOCKED on FW-R1 | Same as FW-R1 | 4-8 hours | Verify gene-level mapping (PMA1, ergosterol pathway, TRK1, V-ATPase) between yeast and mammalian senescence. |
| **FW-R3** | Replicate Phase 3A cancer sterol-biosynthesis up-regulation | BLOCKED on FW-I4 | CellxGene Census + GTEx + GSE131907 NSCLC | 4-8 hours | **Verification gate:** ≥1.0 log₂ up-regulation in cancer vs normal sterol-biosynthesis genes. |
| **FW-R4** | Replicate Phase 3B amphotericin B selective damage | DEFERRED | NO direct public replication; needs independent wet-lab OR public cell-line-effect databases | n/a | **Mentor-lab tri-culture data is mentor-lab IP.** For commercial claim, either (a) external wet-lab collaborator runs independent assay, or (b) frame as mechanism prediction with public cell-line-effect DB validation only. |
| **FW-R5** | **Replicate Phase 3E chitin-high persister-cancer 9% panel** | BLOCKED on FW-I4 | GSE131907 NSCLC + GBM cohorts (Darmanis, ~338k cells) + Suva GBM | 1-2 days | **Verification gate:** identifiable 5-15% sub-population with CHI3L1 + senescence + sterol panel co-expression. Critical commercial claim — high-priority. |
| **FW-R6** | Three-arm therapy mechanism re-derivation | BLOCKED on FW-R3, FW-R5 | Public mechanism literature | 1 day | Mechanism is clean (literature-anchored); specific clinical recommendations need independent replication if used commercially. |
| **FW-R7** | Antifungal-as-anti-cancer mechanism (separate from FW-R4 wet-lab) | BLOCKED on FW-R3 | Public itraconazole NSCLC trial data + amphotericin B literature | 1 day | Mechanism preserved; specific tri-culture numbers retire to mentor-lab IP. |
| **FW-R8** | **`HALO_THE_FUNGAL_CELL_FOUR_LINES.md` split-restate** — clean lines 1-3 → Sovereign-shippable preprint; line 4 (wet-lab) → vault-internal appendix | BLOCKED on FW-R1, R2, R3, R5 | n/a | 1 day operator | **Major deliverable:** the Sovereign Fungal Cell preprint (per `37_Project_Sovereign/BOUNTY_BOARD.md` SOV-4) becomes shippable when this lands. |

**Critical path (Recycler):** FW-I4 → FW-R1 → FW-R3 → FW-R5 → FW-R8. **Total: ~5-8 days from clean start.**

---

## Tier 3 — UHRF1 / TE silencing (12_BloodyEchoes) — anti-aging intervention IP

Unlocks BloodyEchoes-anchored anti-aging consulting + Bryan Johnson Don't Die outreach.

| ID | Title | Status | Public dataset | Time | Notes |
|---|---|---|---|---|---|
| **FW-B1** | **Replicate UHRF1 -71% senescence drop on public WI-38** | BLOCKED on FW-I4 | GSE226225 + GSE150247 + Tabula Sapiens senescence | 1 day | **Verification gate:** UHRF1 expression drops ≥40% senescent vs proliferative on public data. Tolerance: ≥40% reproduces the directional finding even if absolute % differs. |
| **FW-B2** | **Replicate TE-silencing collapse cascade** (DNMT1, TRIM28, SETDB1, MORC2 sequential drops) | BLOCKED on FW-B1 | Same as FW-B1 | 1-2 days | **Verification gate:** ≥3 of 4 cascade genes show ≥30% drop with sequential timing pattern. |
| **FW-B3** | **Replicate ADAR / SAMHD1 / APOBEC3 alarm response** | BLOCKED on FW-B2 | Same | 1 day | **Verification gate:** ≥2 of 3 alarm genes show ≥50% up-regulation on senescent. |
| **FW-B4** | **Replicate mito-checkpoint lncRNA ENSG00000289474** correlation with mitochondrial encoded genes | BLOCKED on FW-B1 | Same + GTEx mitochondrial reference | 1 day | **Verification gate:** ρ ≥ 0.30 between ENSG00000289474 and mtRNA fraction on public data. |
| **FW-B5** | **HALO_BLOODYECHOES_PROTOCOL.md split-restate** | BLOCKED on FW-B1, B2, B3, B4 | n/a | 1 day operator | Already mostly clean (theory + literature); just verify references are public-only. |

**Critical path (BloodyEchoes):** FW-I4 → FW-B1 → FW-B2 → FW-B3 → FW-B5. **Total: ~4-5 days from clean start.**

---

## Tier 4 — Annotation-free atlas (20_MarathonLament) — methodology paper IP

Unlocks the annotation-free methodology Genome Biology / Nature Methods candidate paper.

| ID | Title | Status | Public dataset | Time | Notes |
|---|---|---|---|---|---|
| **FW-M1** | **Build `atlas_public.db`** — annotation-free atlas equivalent to atlas_full6.db on public BAMs | BLOCKED on FW-I4 | Public BAMs from multiple GEO sources (GSE226225, GSE131907, GSE131928 GBM, etc.) | 3-5 days | **Verification gate:** ≥80,000 unique structural transcripts; pattern of ribosomal-protein-gene introns + snoRNA hosts reproduces. Largest time cost in FIREWALL — the atlas build is compute-intensive. |
| **FW-M2** | Re-derive 2,230 ghost-of-transcript pairs on `atlas_public.db` | BLOCKED on FW-M1 | atlas_public.db | 1 day | **Verification gate:** ≥2,000 ghost pairs identified with similar distribution (snoRNA-host dominant). |
| **FW-M3** | **Replicate BT70 fossil-layer senescence enrichment** (~57% lncRNA + pseudogene + repeat-containing) | BLOCKED on FW-M1 | GSE226225 + GSE250041 atlas slices | 1-2 days | **Verification gate:** ≥40% enrichment of lncRNA + pseudogene transcripts in senescent vs proliferative. |
| **FW-M4** | Replicate chr9:19.2-19.4M DENND4C 15-inner-transcript case | BLOCKED on FW-M1 | atlas_public.db | 1 day | Verify the standout pattern reproduces. |
| **FW-M5** | Replicate snoRNA-host-gene rediscovery pattern | BLOCKED on FW-M1 | atlas_public.db + RefSeq snoRNA annotations | 1 day | Verify SNORA31/TPT1, SNORA7A/RPL32, SNORD101/RPS12 type patterns. |
| **FW-M6** | **HALO_GHOSTS_OF_TRANSCRIPT.md split-restate** | BLOCKED on FW-M1, M2, M3 | n/a | 1 day operator | Framework section stays clean; specific count results updated to public-atlas numbers. |

**Critical path (MarathonLament):** FW-I4 → FW-M1 (the long pole) → FW-M2 + FW-M3 + FW-M5 in parallel → FW-M6. **Total: ~6-9 days from clean start.**

---

## Tier 5 — Audit-only items (verify clean status; no full replication needed)

| ID | Title | Status | Time | Notes |
|---|---|---|---|---|
| **FW-W1** | **Audit Quantum-BioAntenna Zenodo deposit** for source provenance | OPEN | 4-6 hours | Verify all 16 files in deposit derive from public data. If any traces to lab data, identify and flag. |
| **FW-W2** | **Verify GNRA Mg²⁺/Li²⁺ MD simulation** Biowulf compute was operator-personal billing | OPEN | 1 hour | Pull Biowulf billing records; confirm operator-personal account. If yes: NA. If no: re-run on personal compute. |
| **FW-W3** | **Audit spliceosome catalogue (208 introns)** — verify all sources are public | OPEN | 1 day | Per-intron provenance check. Flag any that traced to lab data. |
| **FW-W5** | **Audit K-tensor framework** — per-claim provenance for specific coupling values | OPEN | 1-2 days | Mixed status; per-finding audit. |
| **FW-D1** | **Audit DeadShape lysosome shape result** for source provenance | OPEN | 4-6 hours | Verify proteomics + cell biology data sources are public. |
| **FW-D2** | **Audit DeadShape vacuole-fungal homology methodology** | OPEN | 4-6 hours | Per-claim audit. |
| **FW-D3** | **Audit `recycler_homology_test.py` outputs** | OPEN | 1-2 hours | Verify output inputs were public. |
| **FW-MM2** | Verify Idenbraid GNRA MD simulations Biowulf billing | OPEN | 30 min | Same as FW-W2 (overlapping). |
| **FW-MM3** | **Audit XIST gain-control framework** sources | OPEN | 4-6 hours | Per-claim provenance. |
| **FW-MM4** | **Audit HALO_QIP.md framework** sources | OPEN | 4-6 hours | Per-claim provenance. |

These don't require full re-derivation if provenance audit shows clean — they require receipts.

---

## Tier 6 — Operator-action only (cannot be done by AI seats)

| ID | Title | Status | Time | Notes |
|---|---|---|---|---|
| **FW-OP1** | **Operator IP-attorney consultation** on federal-employee-invention scope (37 CFR 501) | OPEN | 1-2 hour consult | Bring HALO_REPLICATION_TARGETS.md as inventory; per-finding determination of duty-related vs personal. |
| **FW-OP2** | **Verify NIH-employment IP-disclosure form** that operator filed at hiring | OPEN | 1 hour | Pull from HR / NIH records. Establishes pre-employment IP baseline. |
| **FW-OP3** | **Document timeline of operator-personal compute use** for vault content | OPEN | 4-8 hours | Spreadsheet: for each major HALO/finding, when did operator first develop the framework, on what hardware. Receipts where possible. |
| **FW-OP4** | **NIH OTT consultation decision** — yes/no/timing on formal IP scope determination | BLOCKED on FW-OP1 | n/a | Per IP attorney guidance; do NOT consult OTT before attorney walks through. |

---

## Tier 7 — Continuous

| ID | Title | Cadence | Notes |
|---|---|---|---|
| **FW-C1** | **OpenTimestamps anchor** every replication output within 10 min of creation | per-output | Automated via FW-I6. |
| **FW-C2** | **Verification harness run** before any replication output leaves FIREWALL VM | per-output | Automated via FW-I5. |
| **FW-C3** | **Receipt package backup** to operator-personal cloud storage | per-replication | Multi-region GCS or AWS S3 personal. |
| **FW-C4** | **Update Sovereign WORLDLINE** with each replication SOLVED | per-event | Cross-link to source domain project + Sovereign tier-changes. |
| **FW-C5** | **Quarterly NIH-billing audit** — verify no FIREWALL compute crossed into NIH-billed project | quarterly | Pull GCP billing for `desynchronization-theory` org and confirm no FIREWALL-related charges. |

---

## Aggregate timeline + budget

**Best-case clean replication of highest-priority commercial IP** (FW-I tier + FW-S tier + FW-R5 + FW-B1):

- Operator time: ~40-60 hours over 2-3 weeks (largely in evenings/weekends)
- Compute cost: ~$200-400 on personal GCP
- Storage cost: ~$100/month carrying cost during active replication
- **Total cash outlay: ~$400-600** for the unlocking of all of: VitaDAO submission + senolytic-timing patent + chitin-high cancer panel + UHRF1 anti-aging intervention IP

**Full FIREWALL (every PENDING bounty completed):**
- Operator time: ~150-250 hours
- Compute cost: ~$1,500-3,000
- Storage cost: ~$200-400/month carrying during replication
- **Total: 6-10 weeks operator-time + ~$2,000-4,000 cash**

---

## Cross-references

- **Source projects** that own each replicated finding: `35_Project_TheHats`, `28_Project_RedFromTheGrave`, `12_Project_BloodyEchoes`, `20_Project_MarathonLament`, `27_Project_WingsAboveMorning`, `34_Project_DeadShape`, `13_Project_MemoryOfMind`
- **Sovereign tier-1 bounties** that unblock when FIREWALL clears specific findings:
  - SOV-1b (peak-and-release patent provisional) → unblocked by FW-S11
  - SOV-2 (file Fungal Cell patent provisional) → unblocked by FW-R8
  - SOV-4 (Fungal Cell preprint) → unblocked by FW-R8
  - VITADAO_PROPOSAL_DRAFT → unblocked by FW-S11
  - ASTERA_ESSAY_DRAFT → already submittable (operator-personal voice; no firewall dependency)
- **Out-of-scope work (now in proper home):**
  - Cai Lei load-vs-release Pineda + Yadav novel deployment: **moved to [10_Project_DiscordIntoSymphony/experiments/als_load_vs_release_2026-05-04/](../10_Project_DiscordIntoSymphony/experiments/als_load_vs_release_2026-05-04/)** (bounties ALS-1, ALS-2, ALS-3 there)
  - Project 39_MindWithBody load-vs-release HALO + tool inventory live there, not here
  - 38 stays focused on cleaning NIH-tainted derivations; novel framework work belongs in the project that owns the framework
- **Doctrine reference:** [HALO_FIREWALL_DOCTRINE.md](HALO_FIREWALL_DOCTRINE.md)
- **Inventory reference:** [HALO_REPLICATION_TARGETS.md](HALO_REPLICATION_TARGETS.md)
- **Personal compute reference:** [HALO_PERSONAL_COMPUTE.md](HALO_PERSONAL_COMPUTE.md)