NOVION · Novel Biotech · personal decision memo
Keep Codex running.
CONCEPT STUDY · RESEARCH ONLY · NO HUMAN TESTING · NOT AVAILABLE · NO DEVELOPMENT PROGRAM EXISTS
Historical recommendation: CONTINUE. The 12 August 2026 memo asked whether the Codex subscription should continue and argued for renewal through the research phase to mid-2027. This reading edition preserves that reasoning; it is not current product, model, pricing, or subscription advice.
Preserved source identity
Source PDF: output/pdf/codex-continuation-justification-2026-08-12.pdfPrivate repository source
SHA-256: e7326c6a230d2a572ee30d407ff78f8445659d4857fdb230ebf2cef4907a995a
A hash establishes byte identity, not truth or current authority. The single dense source page was rendered and inspected. Model names, usage comparisons, task counts, program counts, and the 2027 horizon are preserved as historical statements only.
Source page 1 · opening
One question — should the Codex subscription continue?
The memo described Codex “(GPT-5.6 Sol, xhigh)” as the production engine behind Novel Biotech: multi-task orchestration doing volume work that no single interactive assistant could sustain. It asked to judge that claim against what had already been produced, what the system was for, and what only it could carry.
| Value | Source description |
|---|---|
| 9 | Orchestrated Codex tasks — 1 Integrator + 8 specialist lanes |
| 3 | Concept threads taken to evidence-gated records: MF-01 · BPNI · LIMEN |
| 139 | Files in the LIMEN R0/R1 research scaffold, across 4 lanes |
| 34 | Independently QA-passed canonical BPNI science claims |
| → 2027 | Authorized research-only horizon, through mid-June |
Source section 01
The engine room — what Codex actually does
The memo described a standing nine-task topology — Integrator, Science, Frontend, Backend, Business, QA / Red-Team, Legal & Ethics, Brand, and Product Design — at xhigh reasoning. Its job was said to be high-volume, long-horizon work: deep prior-art research, science claim registers, engineering contracts, traceability matrices, and adversarial self-review.
It characterized the output as hundreds of long-context documents, produced and QA-checked continuously. It claimed premium interactive models could not carry that load and recorded that a single review pass over the repository consumed 13% of a Claude allotment. Codex was called the workhorse; scarce premium models were reserved for spot review and integration.
Source section 02 · FACT
Already delivered
- A truth-disciplined program frame — every claim tagged FACT / PLAN / HYPOTHESIS / AMBITION — with product, design, and authority documents.
- A rebuilt public site in the Graphite instrument identity; a fail-closed application intake; the Observatory preserved.
- MF-01 concept study: research record, nonfunctional model, evidence boundaries, kill criteria.
- BPNI program: independently-QA’d science packet + accepted product research program; visual reports rendered and delivered.
- PROJECT LIMEN: a four-lane R0-synthetic / R1-non-biological research scaffold, fully boundary-labeled and gate-tracked.
Source section 03 · PLAN
What it is for next
- Close the Science simulator assumptions contract (G8) and Backend review; then decide which research branch or branches to retain.
- Build the R0 synthetic simulator and R1 non-biological bench — offline, non-actuating, each separately authorized.
- Keep the site as an honest recruiting surface for exceptional early-career builders.
- Hold evidence and gate discipline: no premature product, no external claims, no unauthorized spend.
- AMBITION: make the human body measurably less fragile within a lifetime; earned by evidence, never asserted.
A small visible annotation beside the ambition in the source reads “ONE PER PAGE”; it is preserved here as a layout-era source artifact, not interpreted as a program instruction.
Source section 04
Why Codex specifically
The memo argued that the work was parallel, structured, and self-reviewing by design: many bounded lanes producing evidence, one Integrator reconciling it. It said that shape mapped to a subscription capable of sustained multi-agent generation at xhigh reasoning, not to per-interaction premium billing.
It framed the realistic alternative not as “do it cheaper elsewhere,” but as doing far less. Without Codex, it claimed the science registers, engineering contracts, and QA evidence that made the program credible would not be produced at all.
CONTINUE
Renew Codex for the research phase through mid-2027. The memo called it the only sustainable way to run the stated volume of structured research and adversarial review at once, and the direct source of every artifact listed above.
Candid caveat preserved from the source: Codex produces drafts and evidence, not authority. Every packet still needs human acceptance, and lanes can over-produce; LIMEN grew large and fast, ahead of its gates. The value was said to hold only while Integrator + QA discipline held.
Grounding printed in the memo: local NOVION working tree @ 2026-08-12; HANDOFF.md §0–23 · PRODUCT.md · DESIGN.md · docs/{science,design-comps,roadmap,marketing} · sites/evidence-lab. The public reading edition removes the private absolute repository path preserved in the source edition.
The source closed with: “Boundary — concept stage. No company, lab, product, trial, or approval exists.”