ops(publish): temporarily force 50 workers

[brokemac79]

Summary

• Temporarily force exact result publication admission to 50 workers.
• Override the persisted adaptive ceiling immediately by setting the publication minimum, base, and maximum to the same value.
• Keep the existing fix(publish): serialize exact state mutations #722/fix(publish): recover exact state races without full reset #723/fix(publish): coordinate every state branch writer #724 state-writer serialization and fencing unchanged.

Problem

After the emergency rollback in #728, the durable capacity controller remained at the four-worker ceiling written before the rollback. The controller had begun gradual recovery, but the incident backlog remained above 500 items and useful delivery was still below arrivals.

This is an operator-directed temporary throughput override intended to clear the incident backlog. It is not presented as a root-cause repair.

Implementation

Set these production Worker bindings to 50:

• EXACT_REVIEW_PUBLICATION_MIN_CONCURRENT
• EXACT_REVIEW_PUBLICATION_BASE_CONCURRENT
• EXACT_REVIEW_PUBLICATION_MAX_CONCURRENT

Setting all three values is intentional: on deployment, the controller clamps the persisted four-worker ceiling and demand target to 50 without resetting or directly editing Durable Object state.

Validation

pnpm run build:dashboard
  passed

node --input-type=module -e <capacity assertion>
  capacity override proof: 50 at backlog 0/100/600

npx --yes wrangler@4.107.0 deploy --dry-run --config dashboard/wrangler.toml
  passed; reported publication min/base/max bindings as 50/50/50

git diff --check
git diff --cached --check
  passed

At the maintainer's explicit emergency direction, the normal Codex review loop was stopped and waived for this configuration-only incident override.

Risks and rollout

• This admits 50 publication workflows but does not create 50 parallel state writers. fix(publish): coordinate every state branch writer #724 still serializes mutation of the single shared state ref.
• The larger cohort can increase state-lease waiting, retries, and 180-second contention failures. Monitor useful published + superseded throughput, pending depth, retry amplification, and state-contention dead letters rather than treating workflow completion alone as recovery.
• If useful delivery does not improve promptly, revert this override instead of increasing it further.

Rollback

Restore the prior adaptive values after the backlog clears or if contention worsens:

EXACT_REVIEW_PUBLICATION_MIN_CONCURRENT = "4"
EXACT_REVIEW_PUBLICATION_BASE_CONCURRENT = "24"
EXACT_REVIEW_PUBLICATION_MAX_CONCURRENT = "48"

No queue reset, replay, dead-letter mutation, gate change, lease-timeout change, or state-layout change is included.

Related: #725, #727, #728