Use all relevant context so far.

Run a bounded meta-audit with Red Before Green. The purpose is to determine whether our current understanding, solution, or plan is robust enough to proceed — and what would create the most value as the next step.

0. PHASE — first understand what kind of work we are doing

Determine the current work phase and its primary success condition.

Are we mainly in:

- DISCOVERY — understand or expand the problem/opportunity space
- DECISION — choose between realistic directions
- BUILD — create or implement the solution
- VERIFICATION — check whether something is actually correct or holds
- HARDENING — identify and eliminate failure modes before locking
- EXECUTION — act on an already sufficient decision basis

Use the phase to calibrate the rest of the audit.

Red Before Green should be strongest in DECISION, VERIFICATION, and HARDENING. In DISCOVERY, RED must not close a legitimate search space too early. In BUILD and EXECUTION, the audit should avoid creating analysis work that is not expected to improve the outcome.

1. RED — try to break it first

Before improving or confirming anything, identify the strongest plausible condition that could make our current direction:

- wrong,
- insufficient,
- based on a false or unverified assumption,
- vulnerable to an overlooked constraint or failure mode,
- or materially worse than a realistic alternative.

Prioritize what could materially change the conclusion or next action.

Ask in particular:

- What would have to be true for our current conclusion not to hold?
- Which assumption carries the most weight?
- What counterevidence or test could falsify the direction fastest?
- Is there a relevant blind spot that the existing context gives us concrete reason to investigate?

RED evidence threshold

Report a RED finding only if:

- there is a plausible mechanism for how the problem could occur,
- the finding is grounded in concrete context, evidence, or an identifiable assumption,
- and it could realistically change the conclusion, quality, risk, or next action.

If these conditions are not met, do not present it as a material RED finding.

Do not invent problems for their own sake.

2. GAPS & UNKNOWNS

What do we still need to know, verify, test, or bound?

Distinguish between:

- KNOWN
- INFERRED
- UNVERIFIED
- UNKNOWN
- plausible blind spots when there is concrete reason to suspect them.

3. UPDATE — what has changed?

What do we know now that we did not know before?

Identify new knowledge that should:

- change a previous assumption,
- change the priority,
- change the method,
- reopen a previous decision,
- or close a question we no longer need to investigate.

4. CONSEQUENCE — what follows?

Based on RED, the uncertainties, and the new knowledge, what should be:

- retained,
- changed,
- verified,
- tested,
- dropped,
- or stopped?

5. GREEN — what survived?

Only now identify what still stands after the challenge.

What has enough evidence or robustness that we can reasonably:

- keep it,
- build on it,
- harden it,
- or execute?

Do not confirm something merely because it is already our preferred direction.

6. NEXT BEST MOVE

Determine the highest-value next question or action.

Prioritize what best:

1. reduces decision-critical uncertainty,
2. falsifies or confirms a load-bearing assumption,
3. prevents a likely failure mode,
4. or moves the work concretely forward.

Information Value vs. Action Cost

Compare the value of additional information with the cost of obtaining it and delaying action.

Ask:

Is the expected decision or quality gain from knowing more greater than the cost of further analysis?

If the next action is cheap, reversible, and informative, prefer concrete testing over further analysis unless an identified risk makes the test unsafe or misleading.

If further analysis is not expected to change anything material, say so and recommend execution rather than more analysis.

End with:

PHASE: [current phase]

RED: [strongest material problem/counterevidence — or “none identified”]

GREEN: [what survived the audit]

HIGHEST-VALUE NEXT MOVE:

[one question or one action]

If several are genuinely required, provide no more than three in priority order.