# Source reconciliation and feasibility decisions

These are execution interpretations and identified constraints, not a replacement strategy. Original files remain unchanged. The week briefs preserve the original source instruction and add precise implementation checks; where the original wording conflicts with measured evidence, report the evidence and retain the missed target.

| Issue in the source material | Execution treatment |
|---|---|
| v3 describes 16 weeks/Vara; v4 describes 20 weeks; v5 supersedes both | Use Week 00 setup + Weeks 01–40 from v5. v4 only supplies retained details. |
| v5 Part 0 says three instruments by week 20; actual schedule builds trialkit in 25–26 and labdocs in 33–35 | Half 1 delivers batchlens and FermLit; the three instrument target belongs to week 40. |
| General shape says two study blocks throughout; Part 7 says one in Half 2 | Preserve six 90-minute blocks plus Sunday hour; Tuesday is formal study and Thursday becomes applied review/market work. |
| '100 real batches' conflicts with IndPenSim's simulated origin | Call it an open simulated industrial-fermentation benchmark. Verify actual source, batch count and fault metadata. |
| Example headlines assume hybrid superiority or fixed corpus counts | Treat them as draft concepts. Replace with actual findings/N, including losses and incomplete datasets. |
| Gold-set tuning followed by scoring the same 50 papers | Freeze paper-level development/held-out partitions; report the distinction. Independent annotation remains human. |
| Kappa requested for all schema fields | Categorical kappa where meaningful; numeric tolerance/error agreement for continuous values. Retain pre-adjudication labels. |
| kLa from DO alone; exact Monod recovery with SINDy; lag from basic Monod | Require sufficient measured variables, appropriate model/library and identifiable parameters. An honest non-identifiability finding completes the bounded exercise. |
| CC BY 4.0 proposed for all extracted data | Assess source terms and what is redistributed. Do not copy full papers into a public release or assume source quotations can be relicensed. |
| Week 12 endorsement messages reference Paper 1 before its scheduled submission | Use the real report/dataset/draft link; update when public. |
| Week 19 asks for two preprints before Paper 2's week 20 gate | Application assets label Paper 2 as manuscript/draft or submitted, as actually true. |
| Week 16/20 goals alternate between submitted and public | Record both. Local submission readiness is controllable; public gates require public IDs and may remain pending. |
| Endorsement, co-authorship, submission and acceptance blend in reputation levels | Keep separate. No L4-co-authored without a real co-authored public paper; no L5 from submission alone. |
| Founder outreach targets mix messages, booked calls and completed conversations | Track each separately. Week 8 gate: five messages; relationship ambition: five conversations. |
| Paper 3 depends on a partner lab | Start relationship work earlier; require written contribution/protocol agreement at 24. Computational fallback preserves useful work but does not satisfy co-authorship/wet-lab targets. |
| Gate 28 says any invoice; main week asks sold and half paid | Track agreed scope, issued invoice and received payment separately; retain the stronger sales target. |
| Week 29 shifts to selling if 28 misses, while 30 assumes delivery | Use the explicit recovery branch and log the delivery/research reforecast. |
| Six-to-eight-week pilot begins week 37 | Continuation runs through weeks 42–44; week 40 can rely on an earlier delivered paid audit. |
| JOSS/venue timing is external | Recheck eligibility and current requirements; maintain sustained public development from early releases. Submissions are not guaranteed acceptance. |

## Capacity decisions to make during execution

1. Week 6: time the first three hand annotations. Fifty complete papers with provenance in 4.5 build hours may be infeasible. Report the forecast, use available buffer, and identify whether scope, outside assistance or dates must change; do not trade off label integrity.
2. Weeks 5–9: corpus acquisition and bulk extraction need an observed cost/throughput pilot. The inherited API budget is USD 150–200 for the first 20 weeks, not authorization to spend it. Independent annotation and manuscript review costs are additional source estimates, not already funded.
3. Weeks 8, 13, 18 and 35: long articles and releases compete for build/buffer time. Reuse figures and draft in the designated work blocks; the Sunday hour cannot contain a full article from scratch.
4. Weeks 25–26 compress broad instrument scopes. Ship one statistically valid end-to-end path first, then record unmet source breadth explicitly. Image scoring is not in v5's scheduled deliverables and stays parked.
5. Weeks 29–30: a 20-hour audit consumes all two-week capacity, leaving no separate study/Paper 3 time. Before promising the audit, make a concrete choice of smaller fixed scope, longer delivery window, or explicitly increased availability. Until Peter chooses, prepare drafts and flag the commitment conflict.
6. Weeks 37–40: client delivery, talks and job processes share the same budget. Set one primary route based on actual commitments; log what is deferred.

## Assumptions left visible

English is the planning and international recruiting default, with Spanish versions where useful for LATAM contacts. No start date, desired salary, work authorization, travel preference, client tax identity, API spend authorization or exact role seniority is invented. Week 00 preparation gathers these when they materially affect execution. The source's equipment and employment history are to be confirmed against Peter's actual current situation.

The programme can strengthen an evidence-backed scientific profile; a particular publication, title, job, co-author, discovery or revenue level is an aspiration, not a promised result. A suitable accepted role can be a valid early alternative outcome under v5 Part 10.

## 2026-09-29 · v7 re-sequence: hybrid, lead-first

Week 1 showed a 30-minute biology block taking five hours, and the v5/v6 scope (three preprints, two journal submissions, JOSS, three products, protein, extremophile and field-trial work, a fund offer) could not fit 400 hours in any order. Peter's goal is a well-paid role helping biotech/pharma implement AI with a credible brand, not a research-scientist track. Decision: [ROADMAP-v7](./ROADMAP-v7.md) governs from week 5 on. labdocs (regulated-document AI) and GridX case studies move forward; FermLit ends in one technical report plus dataset with DOIs; the job search starts at week 19 with a market and salary check at week 5; Papers 1–3, journal submissions, protein/extremophile, trialkit and the fund offer are parked with a revisit condition. Study is one core source per slot, 60 minutes, stopped by the understanding check.

## 2026-09-29 · Weeks 1–4 in mission format, trimmed study core

Core study cards for weeks 1–4 cut from 30 to 21 (v6 Learn-it lists and materials.md; demoted items kept under Deeper). Each day in the week 1–4 specs is now a mission (Hook, Do, Done when), designed for Peter's twice-exceptional profile: interest-led hooks, one stop signal per day, boring parts delegated to AI, Saturday catch-up plus rabbit-hole slot, Sunday "show it". Later weeks are prepared on a rolling two-weeks-ahead basis; see [ROADMAP-v7](./ROADMAP-v7.md#how-the-rest-of-the-weeks-get-built).