Private improvement. Public learning, only by choice.
better loop.

Practice together. Keep your work private.

Choose a challenge.
Make it your own.

Sharpen your prompt. Check what happens. Practise with your own work, borrow an idea from another field, and keep the lesson that helps you.

One useful next step

Where does your task get stuck?

Start with a situation you recognise. Open a brief, choose a check that matters, and try it in your own workspace. No account needed.

All eight briefs can be adapted to general tasks. Choose the context of this task and make the check your own.

diagnosing error

Test the explanation before fixing the error

Practise: context gap detection, reasoning scrutiny, factual verification.

Open practice brief

Choose your acceptance check

Before the change, choose a check that reproduces the error and a check for an unaffected case. Record whether both pass afterward, and retain any regression or unresolved explanation.

Then try the next attempt

Use a new authorized local task. Select the difficulty and an appropriate acceptance check before trying the change. Review your selected private history, record your own decision and the checked outcome, and keep the lesson even if it did not help. Sharing is optional.

Relevant quality checks: correctness, verification, reproducibility. Adapt them to your task family and difficulty; a competent domain evaluator is still needed.

Prepare my local practice →Related evidence: Code · diagnose an error

reconciling data

Reconcile a discrepancy with traceable assumptions

Practise: goal definition, context gap detection, factual verification.

Open practice brief

Choose your acceptance check

Prespecify a total or invariant, a missing-entry case and a units check. Verify each against an authorized source. Keep original quantities private and record unresolved differences rather than forcing agreement.

Then try the next attempt

Use a new authorized local task. Select the difficulty and an appropriate acceptance check before trying the change. Review your selected private history, record your own decision and the checked outcome, and keep the lesson even if it did not help. Sharing is optional.

Relevant quality checks: correctness, completeness, reproducibility. Adapt them to your task family and difficulty; a competent domain evaluator is still needed.

Prepare my local practice →Related evidence: Analysis & finance · reconcile data

synthesizing evidence

Separate sources, inferences and uncertainty

Practise: audience definition, reasoning scrutiny, factual verification.

Open practice brief

Choose your acceptance check

Choose consequential claims before drafting. Check their original sources, whether the competing interpretation is represented fairly, and whether the conclusion discloses material uncertainty.

Then try the next attempt

Use a new authorized local task. Select the difficulty and an appropriate acceptance check before trying the change. Review your selected private history, record your own decision and the checked outcome, and keep the lesson even if it did not help. Sharing is optional.

Relevant quality checks: reasoning, verification, communication. Adapt them to your task family and difficulty; a competent domain evaluator is still needed.

Prepare my local practice →Related evidence: Research & strategy · synthesize evidence

quantitative reasoning

Challenge a calculation with an independent check

Practise: goal definition, reasoning scrutiny, factual verification.

Open practice brief

Choose your acceptance check

Prespecify a simple known case, a boundary case and a units or invariant check. Compare the result with an independent method; do not treat a confident explanation as proof.

Then try the next attempt

Use a new authorized local task. Select the difficulty and an appropriate acceptance check before trying the change. Review your selected private history, record your own decision and the checked outcome, and keep the lesson even if it did not help. Sharing is optional.

Relevant quality checks: correctness, reasoning, reproducibility. Adapt them to your task family and difficulty; a competent domain evaluator is still needed.

Prepare my local practice →Related evidence: Math & science · check a calculation

creating content

Make the brief testable for its reader

Practise: audience definition, quality examples, output structure, iterative refinement.

Open practice brief

Choose your acceptance check

Write a short review rubric for audience usefulness, required content and output constraints before drafting. Apply the same rubric to both attempts, retaining weaknesses and reviewer uncertainty.

Then try the next attempt

Use a new authorized local task. Select the difficulty and an appropriate acceptance check before trying the change. Review your selected private history, record your own decision and the checked outcome, and keep the lesson even if it did not help. Sharing is optional.

Relevant quality checks: communication, completeness, constraint compliance. Adapt them to your task family and difficulty; a competent domain evaluator is still needed.

Prepare my local practice →Related evidence: Writing & design · make a testable brief

coordinating plan

Choose the decisions that need a human checkpoint

Practise: goal definition, approach consultation, collaboration mode.

Open practice brief

Choose your acceptance check

Before execution, check that each required outcome has an actionable step, dependencies are consistent and consequential decisions have a human checkpoint. Later record which assumptions held and which caused rework.

Then try the next attempt

Use a new authorized local task. Select the difficulty and an appropriate acceptance check before trying the change. Review your selected private history, record your own decision and the checked outcome, and keep the lesson even if it did not help. Sharing is optional.

Relevant quality checks: completeness, reasoning, constraint compliance. Adapt them to your task family and difficulty; a competent domain evaluator is still needed.

Prepare my local practice →Related evidence: Operations & education · coordinate a plan

extracting information

Distinguish missing information from a negative answer

Practise: output structure, quality examples, factual verification.

Open practice brief

Choose your acceptance check

Prespecify known-present, known-absent and ambiguous cases, plus the required output structure. Check against the selected source without replacing missing values with invented facts or zeros.

Then try the next attempt

Use a new authorized local task. Select the difficulty and an appropriate acceptance check before trying the change. Review your selected private history, record your own decision and the checked outcome, and keep the lesson even if it did not help. Sharing is optional.

Relevant quality checks: correctness, completeness, constraint compliance. Adapt them to your task family and difficulty; a competent domain evaluator is still needed.

Prepare my local practice →Related evidence: General work · extract information

improving process

Change one habit and check the next attempt

Practise: approach consultation, iterative refinement, reasoning scrutiny.

Open practice brief

Choose your acceptance check

Freeze the outcome check and quality floor before trying the change. Compare a later compatible attempt, include known orchestration overhead, and keep neutral, adverse and missing findings. Lower token use alone is not better quality.

Then try the next attempt

Use a new authorized local task. Select the difficulty and an appropriate acceptance check before trying the change. Review your selected private history, record your own decision and the checked outcome, and keep the lesson even if it did not help. Sharing is optional.

Relevant quality checks: correctness, verification, constraint compliance. Adapt them to your task family and difficulty; a competent domain evaluator is still needed.

Prepare my local practice →Related evidence: General work · improve a process

Choosing a brief or copying a prompt is preparation, not a completed challenge.

About these practice briefs

These authored prompts are not validated benchmarks. Adapt the acceptance check to your task and choose a competent domain evaluator. Selection, copies and scoped Markdown changes do not establish participation, human improvement or learning transfer.

Look through one clear lens.

Find lessons about a similar problem, including ideas from other fields. Choose the task difficulty and the host you want to compare.

Choose the work you want to learn from.

Showing: Code · diagnose an error · moderate · codex.

Exact conditions for this comparison

software · diagnosing error · correctness. moderate difficulty, self-estimated; quality-threshold constraint only. Observational follow-up; same codex platform and standard model tier in both attempts.

Self-reported evidence, locally captured metrics. Required checks: correctness, verification; evaluator tool, locally recorded basis. All must be met, with no observed critical regression. Broad model tiers do not establish equivalent models or work.

Framework better-loop-fluency-0.1 · rubric bl-work-evidence-0.1 · task bl-task-0.1 · metrics bl-metrics-0.1. Neutral and adverse eligible outcomes are retained. Other conditions are outside these finite views.

Comparable progress

Quality and relative effort, side by side.

Look at reported quality and relative token use together. Use these descriptive comparisons to shape a question for your own next attempt.

Reported quality

Keep the quality floor visible

Current comparison unavailable

Try this view again when current evidence is available.

Higher is better on the reported quality rubric. Missing quality is never zero.

How to read this comparison

A fresh, complete group needs at least 20 distinct owners after all conditions and quality checks. No small-group count or estimate is shown. A known evaluator and recorded basis are required. Each owner has equal weight; metric groups can differ.

Uncalibrated descriptive reported progress. No equal-difficulty, causal improvement, percentile or ability-ranking claim.

Relative token use

Use fewer tokens, with checks

Current comparison unavailable

Try this view again when current evidence is available.

Baseline 100 is each report’s own baseline; lower means fewer tokens. Coverage is not aggregated.

How to read this comparison

A fresh, complete group needs at least 20 distinct owners after all conditions and quality checks. No small-group count or estimate is shown. A known evaluator and recorded basis are required. Each owner has equal weight; metric groups can differ.

Absolute usage and worker, judge, retry and cache coverage are unavailable here. Cost, time and human effort remain separate. Lower usage alone is not better quality.

Uncalibrated descriptive reported progress. No equal-difficulty, causal improvement, percentile or ability-ranking claim.

Who is included, and what can these numbers tell me?

Numbers require separate benchmark-aggregation permission and at least 20 distinct eligible owners after every condition and quality check. Each metric can have a different group; the panels are not a paired population result. No small-group counts or estimates are shown.

Uncalibrated descriptive reported progress does not establish equal difficulty, causal improvement, percentiles or ability rankings. Twenty-owner suppression and rounding are privacy heuristics, not an anonymity guarantee. These comparisons are separate from registered benchmark evidence. Role-evidence discovery has its own independent opt-in.

The loop stays yours

Keep the lesson.
Share only by choice.

Keep your prompt, working agreement and checked outcomes on your laptop. Come back with a clearer next move. Share a lesson only when you want to.

Start my private practice →Preview an optional story