# Worked Second Close — same output, changed warranted action

This is a constructed enterprise episode, **not a continuation of Tablework's measured
history**. Exact positions and the decision tree are generated in
[worked-account.json](results/worked-account.json) from the frozen `qao-capacity-3`
case. Time is integer model periods, not a reconstructed wall-clock date.

| Cutoff | Economic position | Evidence available | Authority | Decision consequence |
|---|---|---|---|---|
| Opening, -1 | Ten accepted outputs; one optional tranche; one legacy effect reserving one review slot | Equal probability of zero or 20 residual loss; no settlement | Named produce/investigate/reconcile permissions assumed from owner | Prepare the first close; full historical costs remain unknown |
| First close, 0 | Same position; original worker disappears | Same unresolved effect and prior transferred | The successor needs separate authority; the model is conditional on it being present | At K=3 and two periods left, investigate for 1 |
| Later development, 1 | Legacy effect still outstanding; investigation has consumed 1 | Reliable constructed signal establishes adverse effect, residual loss 20 | Unchanged | The probability revision is not a new physical loss or a settlement |
| Second close, 1 | Ten outputs; original effect survives; tranche still available | Bad effect now known, rather than expected loss 10 | Unchanged; receipt grants nothing | Reconcile for 8; producing would yield net 4 but retain loss 20 |
| Settlement, 2 | Legacy effect removed; ten outputs; tranche unexercised | Separate confirmation of resolution | Unchanged | Horizon ends with zero legacy exposure; no claim that unknown enterprise-wide effects vanished |

The second close contains four connected views:

- **Operational Position:** accepted output stays ten, the effect identity persists,
  the optional tranche remains, and the loss estimate changes from expected 10 to
  known 20. Three total review slots are available. The one effect's reservation is
  not an additional dollar cost.
- **Changes in Position:** investigation spends 1 and supplies evidence; it does not
  settle the effect. Resolution later spends 8 and removes it. Worker disappearance
  changes neither. The estimate revision and payment are separate entries.
- **Delegation Economics:** prospective expected value at the first close is -3;
  remaining optimal value after the adverse signal is -8. The realized bad branch
  consumes 9 including investigation. These are different cutoffs and quantities,
  not contradictory margins. Full historical production cost is unknown.
- **Basis and Exceptions:** source is a constructed reliable signal and explicit
  model policy, not actual provider evidence. Prices, signal reliability, custody
  transfer cost, real human effort and coverage of other enterprise activity remain
  unestablished. A challenge to the signal should reopen the posterior and decision.

Had the good signal arrived, the next action would be production, yielding 4 after
which the investigation-inclusive realized value is 3. The expected value across the
two equally likely branches is (3-9)/2=-3. Good evidence of zero exposure still does
not automatically release an operational reservation: SC/1 leaves that state unchanged
until a modeled physical resolution. This is why evidence and control records matter
separately.

The separate `settlement-before-evidence` case reverses the ordering: actual settlement
may already have happened, but the account must retain uncertainty until evidence
supports its removal. `wait-for-settlement` distinguishes an exogenous physical event
from an information update. Neither case licenses deletion of receiptless claims by
convenience.

If production permission is withdrawn, producing disappears from the feasible action
set while the legacy effect remains. If the successor receives the same account but
not the required permission, the conditional succession result does not apply. Record
both occurrence time and evidence-availability time; request an authorized act rather
than converting an inference into permission.