Introduction | For the thesis project, only 1-shot and 5-shot have finished running; 10-shot has not been trained yet. This article explains how to,without fabricating any numbers, finish the sections that can be written first and reserve placeholders for the ones that cannot.
1. What's blocking you isn't writing ability—it's not knowing what can be written
For a few-shot image classification thesis project, a common progress point is: the 1-shot and 5-shot results are out, while 10-shot is still in the training queue.
At this point, the typical state of sitting in front of the computer is: you want to write, but you don't dare to—afraid that if you write down the 10-shot Top-1 accuracy, that would be fabricating it.
The problem can actually be broken down: which sections of the paper do not need experimental data at all? Which sections are only temporarily missing data? Once you distinguish them, you can start immediately.
2. Step One: Upload Existing Experimental Data
Upload the existing 1-shot and 5-shot results (CSV / JSON) to MuchSci Paper Sailor. It will first profile the data: field types, missingness, and distribution.
First get a clear view of “what I actually have on hand”; this step determines how accurate the later gap marking will be.
3. Step Two: Section–Data Gap Marking, ✅ and ❌ at a Glance
After uploading the data, generate the paper outline, then run the gap mapping; each section will be marked:
| Marker | Meaning | Current action |
| ✅ | Enough data | Can write now |
| ⚠️ | Barely sufficient | Narrow the conclusions before writing |
| ❌ | Data missing | Use placeholders first, wait for experiments |
For the 1/5/10-shot example, the 10-shot column in the Results section is clearly ❌—it will not pretend that it can write it.
4. Step Three: Put Placeholders Where Data Is Missing, Not Fabricated Numbers
At positions where data is missing, MuchSci Paper Sailor will keep placeholders:
{{TBD_AFTER_EXPERIMENT}}
That is: for the cell for 10-shot Top-1 accuracy, leave it blank and leave a marker; never fill in a number that “looks reasonable.”
This rule sounds conservative, but it is actually the smartest approach—you can go ahead and finish the whole section, leaving only a few holes, and fill them in as soon as the experiment finishes.
5. Step Four: Get a To-Do Experiment Checklist
Besides the markers, it also outputs a to-do experiment checklist: it clearly tells you “still missing 10-shot Top-1 / 5-way results / confidence intervals,” etc.
This checklist is your training plan for the next few days, rather than a vague “the experiments are not enough.”
6. So Which Parts Can Be Written First?
Divided by the nature of the sections, it is actually quite clear:
• Can be written immediately: Abstract (leave placeholders for values), Introduction, Related Work, Method, Experimental Setup.
• Can mostly be written: the structure of the Results section, the skeleton of figures and tables, descriptive sentence patterns—leave placeholders where values go.
• Must wait for data: the parts of the Conclusion and Discussion that involve 10-shot.
Write the first two categories first, and your draft will be 70–80% complete.
7. A Complete Walkthrough
1. Upload the 1-shot / 5-shot experimental results.
2. Profile the data and confirm the data structure.
3. Generate an outline and get the section skeleton.
4. Run gap mapping: mark the 10-shot-related sections with ❌.
5. Placeholders: write {{TBD_AFTER_EXPERIMENT}} where data is missing.
6. Write the sections that can be written and produce a to-do experiment checklist.
7. Experiments finish → fill in numbers → recheck → finalize.
8. Common Pitfalls
• “Make up the numbers first and revise later.” This is the most dangerous self-soothing—once you forget to revise, it becomes fake data in the submitted version.
• Leaving an entire section blank and not writing it. The right approach is to write it fully and leave only holes, not to stop writing.
• Inconsistent placeholder formats. Use {{TBD_AFTER_EXPERIMENT}} consistently, do a global search before finalizing, and confirm that none remain.
• Treating ⚠️ as ✅. “Barely usable” means the conclusions need to be narrowed in tone; you cannot write as if they were “fully validated.”
9. Conclusion
Not having finished running 10-shot should not be a reason to stop writing. MuchSci's Paper Sailor marks “what can be written and what cannot” very clearly, then uses placeholders to hold the spots for you—you haven't fabricated any numbers, yet you've already written more than half of the paper.
Self-check before submission:
• Have all sections marked ❌ been supplemented with experiments?
• Search the whole document for {{TBD_AFTER_EXPERIMENT}}—are there any remaining?
• Is there any number that I “estimated”?