Repository navigation
Description
The competition task template (emrg/server/competition_prompt.md) leaves two decisions to the installation that it should state itself, and three competition tasks on this machine independently had to derive both.
What was observed
1. The template asks for a gap between submission batches without saying how the gap is computed.
§1 requires the round to open with a per-competition submission-quota ledger, and its quota bullet ends: "The fraction, the batch bound and the gap are calibrated from the platform's own reset time and allowance — what the ledger may not do is omit them." The requirement is stated; the arithmetic is not. Measured in three competition tasks' own memory, each having worked it out for its own allowance:
| Allowance the platform states | Interval derived | Note |
|---|---|---|
| 1 per day | 43200 |
previously 14400, from an allowance guessed as 5 |
| 10 per day | 7854 |
86400 / 11 |
| 2 per day | 28800 |
86400 / 3; the configured allowance said 10 |
All three are 86400 ÷ (n + 1). Three independent derivations of the same formula is a general method with no home in the template.
2. The allowance was twice read from a counter that cannot carry it.
In the first case the round inferred five submissions per day from the platform reporting no submissions remaining after a spend. That reading is consistent with any allowance: after its last spend, a platform reports zero whatever n is. The platform's own wording stated an allowance of one, and the competition was paced six times too fast for as long as the guess stood.
3. The template archives a competition that has "ended" without saying what establishes that.
The cross-round record rules say "a competition that ends (deadline passed, abandoned) moves to an archive entry with its round range, not deleted silently", and the error table has a Deadline passed row. Neither says what evidence ends a competition — and the same §1 that defines the traversal set demands an increment from every active competition each round, with no exemption for one that has been archived. So the judgement that removes a competition from the round's work is the one judgement the template leaves entirely to the round, while the reason closest to hand ("the gap is too wide", "the field is too strong", "there is nothing to do but wait") is exactly the reasoning the template's §0.0 forbids as grounds for slowing the task down.
Why it matters
A pace set from a guessed allowance spends a competition's attempts at the wrong rate, and the error is silent: the submissions are accepted. An end declared on the wrong evidence retires a live competition, or keeps traversing a dead one forever — and a round with an empty set has no rule at all, so it either invents work to keep itself running or edits the host's task configuration.
What the fix must satisfy
- §1 states the gap as arithmetic:
86400 ÷ (n + 1)seconds for an allowance ofnsubmissions per day, with the reason the divisor isn + 1(the reading window after the last spend), the direction an uneven division rounds, and86400for an allowance that is not daily or is spent. - §1 states that the allowance is a reading: the platform's own stated number, or where it states none, the largest number of accepted submissions across the competition's own days of history — and never "no submissions remain", which is true for every
n. - §1 states that the platform's number governs when a configured allowance disagrees with it, and that the difference is recorded as its own reading.
- §4 states the evidence that ends a competition, as named forms each written with the field, quote or timestamp that carries it — the platform's end field passed, the award settled, a stage gate that closes what could still be submitted, or a hard constraint of §3 in new evidence.
- §4 states that an archived competition leaves the traversal set §1 defines, so §1's increment rule no longer applies to it.
- §4 states that difficulty is not that evidence — a wide gap, a strong field, a slow lever and a long wait are assessment material.
- §4 states that when nothing in the set is live the round says so in its closing summary, does not manufacture an increment, and does not switch the task off, because the task's configuration is the host's.
Acceptance
Reading the template, a round can answer "how far apart do I place this competition's submissions, and how do I know the allowance" and "what makes a competition finished, and what does not" without consulting any installation's configuration or memory. Guards hold each rule at both the file and the rendered level, and the file's existing guards still pass.
Activity
how2how2how2-arch commentedon Nov 17, 2026
Handled by #2944