emrg: the competition prompt asks a round for a probability and its round rules forbid stating it, and never says what a best score is #2431

What is wrong

emrg/server/competition_prompt.md gives a competition round two instructions that
answer the same question in opposite ways.

§5 requires a probability. Under the three readings the closing summary must carry:

How likely is the goal to be reached, and on what evidence? — a probability for the
goal §2 defines (the prize, or the standing), with the basis it is estimated from: the
metric's gap to the payout line or to the rank that pays, the submissions/budget left,
the deadline.

The round rules forbid stating it when it is discouraging. The task's own
installation carries, at the top of its precedence chain, a rule that no round may write
that the goal is unlikely, that the phrasing which says so is banned outright (probability
is low, hopes are slim, essentially impossible, expected value too small, chances are
remote, unreachable, ceiling, no leverage, converged), and that the only accepted form is:
the current reading, the gap measured in the competition's own metric, the list of levers
not yet tried, and the next action.

A round that reads both has to pick one. If the honest estimate is that the goal is far
away, §5 asks it to say so and the round rules forbid saying so — so it either writes a
number the round rules ban, or writes the gap-based form and leaves §5's question
unanswered. Every round reads two instructions and chooses silently, and which one it
chooses is not recorded anywhere.

Measured

emrg/server/competition_prompt.md on master:

• line 291 asks for "a probability for the goal";
• the strings out of reach, give up, not a conclusion and untried do not appear
  anywhere in the file.

So the prompt has no sentence that resolves the pair, and none that says what a
discouraging reading is to be written as instead.

The second half: a best score with no stated role

Phase D says "One improvement per round: local validation → submit → record the score
delta and the hypothesis behind the improvement", and tells the round to keep the best
score and the submission that produced it. Nothing says what a score is once it has been
recorded.

A round can therefore record a best score, re-state it next round, and call that
iteration. Nothing in the prompt distinguishes "the best score so far" — a reading, which
says where the work stands — from "the best score we can reach" — a conclusion, which
would end the work. The prompt also never asks a round for an attempt it had not tried
before, so a round that re-measures or re-states has met every stated requirement.

What a fix must satisfy

1. §5's probability question is answered in the competition's own metric — the current
   reading, the gap to the payout line or the rank that pays, the submissions and budget
   left, the deadline — and a discouraging reading is written as that gap plus the levers
   not yet tried and the next attempt, never as a verdict that the goal cannot be
   reached.
2. The resolved rule is stated once, in the section that already owns it, in the
   template's own voice — not appended at the end, and not duplicated.
3. Phase D states that a recorded best score is a reading, not a conclusion, and that a
   round brings at least one attempt it had not tried before, and says in its summary
   what that attempt is and why it is a better method than the last round's.
4. A content guard asserts both in the rendered prompt, and asserts that the
   contradictory phrasing is gone, so the pair cannot come back by being re-added on one
   side.