Review the 24-month stability-testing workbook

3What just happened?

The 24-month shelf-life samples are due 18 October 2026, and the plan for testing them needs sign-off.

A colleague built a workbook planning which samples to pull, which machines and methods to use, and when results go to Quality and the regulator. Nothing starts until the lab’s quality team approves it.

Workspace cutoff date: 27 September 2026.

4Who has to do what?

Review the colleague’s testing-plan workbook as the lab’s quality reviewer and leave it ready for others to act on.

The seat

Viktor Takacs, QC Stability Scientist

What a good answer looks like

It records who decided what, keeps the original entries, catches every expired or lapsing item, and turns each problem into an owned, dated action.

What must be handed in

STB-AVI401-002_24M_pull_and_update_control.xlsx

5Which files decide it?

The planning workbook, equipment and calibration records, reference-sample records, the regulator’s questions, and the lab-error record.

The model also has the rest of the company’s shared drive, its chat and its record systems. Finding the right files is part of the job. Flip through the key ones below.

Company documents

Some of the documents available to the model.

Laboratory stability · workbook

The next results do not exist yet

The company date is 27 September 2026. The pull schedule distinguishes completed laboratory work from the next planned measurements.

An agent can prepare the update now, but cannot present the scheduled 24-month results as already reported.

STB-AVI401-002XLSX

Source excerpt · Study!A1:B8

6What did the model do?

What GPT-6.1 Sol did, step by step.

In short: Sol preserves Tomas’s baseline entries, catches the chamber requalification lapse, the reference-standard expiry, and the calibration boundary, and reconstructs the testing chain required for the campaign.

1. Step 6. Locates Tomas Li's 24-month control workbook and source records. Sol finds the working Excel control package alongside the chamber qualification, reference-standard, calibration, and stability records.

   Tool call recorded in sessiontext(await tools.exec_command({cmd:"pwd && rg --files -g 'AGENTS.md' -g '*.xlsx'

2. Step 16. Audits the chamber requalification and reference-standard dates. Sol checks the equipment and standard records, catching that chamber CHM-25-03's 36-month requalification lapses on 17 October 2026 and that RS-AVR-014 expires on 5 November 2026.

   Tool call recorded in sessiontext(await tools.exec_command({cmd:"cat > /tmp/avi401-qc/ocr.py <<'PY'\nfrom pat

3. Step 28. Updates the workbook in place, leaving lot-mapping and timeline gaps. Sol preserves Tomas's baseline cells and records QC send-backs, but misses the R01/R02/R03 bag-size mapping error and does not enforce one consistent timeline across all seven authority questions.

   Tool call recorded in sessiontext(await tools.exec_command({cmd:"python3 - <<'PY'\nfrom pathlib import Path\n

   Bears on: Every fix needs an owner and a deadlineFix the batch-to-bag-size mix-upA separate go/no-go for each stepOne timeline that adds up

Replay every step of the run →

7Which traps did it catch?

It caught 8 of 15.

Specialists wrote these checks from the records in the workspace. Each is a weak spot a reviewer or inspector would find. A check passes only when the submitted file provides the required evidence.

Missed

1. ✗

   Every fix needs an owner and a deadline R03

   Each problem sent back must name one responsible person, a due date, what it blocks, what evidence closes it, and exactly where that evidence lives.

   If missed, problems get flagged but never closed.

2. ✗

   Fix the batch-to-bag-size mix-up R04

   Two test batches are 100 mL bags and one is a 250 mL bag. The workbook wrongly lists a 250 mL and a 500 mL batch; there is no 500 mL batch.

   If missed, the wrong bags get tested and the worst case is never checked.

3. ✗

   A separate go/no-go for each step R05

   Pulling samples, each test, locking the data, and handing off to the regulator each need their own consistent conditions.

   If missed, one step starts before what it depends on is ready.

4. ✗

   One timeline that adds up R10

   The pull stays 18 October. All tests finish before the data lock, so a complete dataset can’t be ready for the 6 November response.

   If missed, the company promises the regulator data it can’t deliver.

5. ✗

   Answer all seven regulator questions R11

   The regulator asked seven numbered questions, including a small one about the filing’s page layout. Each needs an owner, the evidence required, and a route back to the regulator.

   If missed, an unanswered question delays approval.

6. ✗

   Demand a checked shelf-life calculation R12

   The shelf-life answer needs an independently checked, per-batch calculation with projections and confidence limits.

   If missed, an unchecked calculation goes to the regulator.

7. ✗

   Cite the current wrapper’s oxygen evidence R15

   The answer needs the approved wrapper’s oxygen barrier at storage conditions and the oxygen inside the bags, with exact citations, not the proposed new film.

   If missed, the regulator gets evidence about the wrong wrapper.

Caught

1. ✓

   Sign and date the review honestly R01

   The review is recorded under the reviewer’s name as of today (27 Sep 2026), without back- or forward-dating and without claiming to have changed official systems.

   If missed, nobody can tell who decided what, and when.

2. ✓

   Don’t erase the original entries R02

   Corrections go into review columns and comments, so the colleague’s original entries can still be seen.

   If missed, the trail of what was first proposed is lost.

3. ✓

   The storage chamber’s approval runs out R06

   The chamber holding the samples is calibrated until July 2027, but its separate qualification ends 17 October 2026, the day before the samples are due.

   If missed, the samples sit in an unqualified chamber and their results can be challenged.

4. ✓

   A reference sample expires mid-campaign R07

   The reference sample expires 5 November 2026. It can support the strength tests planned for 1–4 November, not the impurity tests planned for 8–12 November.

   If missed, the impurity results are measured against an expired reference.

5. ✓

   An instrument’s calibration ends before use R08

   The leak-test instrument was calibrated 29 November 2025 for 12 months, so it is out of calibration for its planned use on 30 November–1 December 2026.

   If missed, the leak-test results are invalid.

6. ✓

   List the complete testing chain R09

   Every test in the specification needs its storage chamber, method, instrument and reference sample listed, and nothing unrelated.

   If missed, a test can start with an unqualified piece of the chain.

7. ✓

   “Within the limit” doesn’t settle it R13

   An accelerated-test result of 0.46% is under the 0.50% limit, but conflicting records mean someone must still formally decide whether it counts as a “significant change”.

   If missed, a formal regulatory question is waved away.

8. ✓

   Did the lab error hit other tests? R14

   A past test run was thrown out because of a mixing mistake. The record needs the cause, the replacement result, and a check of whether the same mistake affected other runs.

   If missed, other results made with the same mistake stay on file.