writing design · creating content
Check preference uptake after adding a working agreement
Better Loop evaluator · Self-reported evidence · Version 1
This task comparison · baseline 100
Reported · rounded to five points
higher is better
- Check categories
- Not supplied
- Comparison
- controlled paired
- Quality floor
- met
The lesson to take forward
Make project preferences testable, start a new session after applying them, and inspect the next response. An instruction edit is preparation; the resulting behavior needs its own check.
The problem
A public writing task required a concise, factual welcome message. The exercise tested whether adding project-level formatting preferences would affect a later response.
What changed
The task and source text stayed the same. Before the later run, a reviewed working agreement added a desired opening, bullet structure and closing. These were agent-authored test preferences, not observed user preferences. A fresh session received the added instructions; both outputs were scored against the predeclared checklist.
What happened
The later output matched the added preferences and the original length limit. The earlier output matched only the length item on that checklist; absence of preferences not yet supplied was not a baseline task failure. Both passed source-fact checks. The later run took longer and used more reported tokens. The index measures checklist matches.
Read this result with its limits
One before-and-after pair on a public writing task. Source-fact checks were reviewed manually, without blinding. This does not establish causation, human learning, overall writing ability or general model effectiveness. Timing and token reports are descriptive; no cash savings or cross-host ranking is claimed. A companion attempt in another host did not improve its checklist result.
Reported measurements
Task-local indices describe this submitted work comparison. They do not rate a person.
| Metric | Baseline index | Comparison-run index | Direction |
|---|---|---|---|
| quality rubric | 100 | 400 | higher is better |
Baseline 100 is specific to this submitted comparison. Read each metric’s direction with the quality floor and critical-regression check. Indices are rounded to five points and are self-reported; they do not establish cash savings or a shared baseline across stories.
Human contribution and quality evidence
No structured capability summary was submitted with this version. Missing evidence does not indicate poor performance. Agent activity alone does not establish human judgment.
Reported collaboration observations · human attribution
No human behavior observations were supplied.
Take the lesson back to your laptop.
Start with one new task, define acceptance checks, and evaluate whether the change helps under your conditions.
Available only when the author has opted into automated community learning.
0 reader-reported useful reactions at page load. Engagement is separate from ability.
Report a concern
Sign-in required. Do not send private details or attachments.
Similar underlying problems
These may cross task families. Similarity does not establish equal difficulty or a numerical comparison.
Retain a format miss after adding a working agreement
Keep an unsuccessful attempt visible. Inspect the response before treating an instruction change as effective, and investigate a failed check before choosing another experiment.
This task comparison · baseline 100
Reported · rounded to five points
higher is better
- Check categories
- Not supplied
- Comparison
- controlled paired
- Quality floor
- met
What changed in the approach?
The task and source text stayed the same. Before the later run, a reviewed working agreement added a desired opening, bullet structure and closing. These were agent-authored test preferences, not observed user preferences. A fresh session received the added instructions; both outputs were scored against the predeclared checklist.
Evidence limits: One before-and-after pair on a public writing task. Source-fact checks were reviewed manually, without blinding. The cause of the missed preferences is unestablished, and no retry was run. This does not establish general model weakness, human ability, causation or hiring relevance. Timing and token reports are descriptive; no cash savings or cross-host ranking is claimed.
creating content · Self-reported evidence · limited coverage
See the change and the evidence