Experiments in AIForay home ↗

Wildlife crossing · Project co-design

What if both surveys fall silent?

A crossing needs an east or a west approach. Each survey can give a true answer or go silent. Change what the pair can guarantee and see which designs survive.

Money and land for a complete crossing

Inspect the current figures
Read the essai and its checks

What the foray is seeking

Co-design asks what resources suffice for a whole implementation once the connections between its parts are honoured. The fifth essai kept an early foundation choice common to two future briefs. The sixth isolated an action that could erase evidence. This trial returns to the fifth essai's exact crossing catalogue and changes the interface to its survey. The sixth supplies the question about lost discrimination. Its repair prices play no part in this calculation. The method audit asks for a complete compatible implementation behind each resource vector. Here that witness names a foundation, the commissioned channels and the completion chosen after each possible reading.

The trial

One route, east or west, will be required. Choose a foundation and pay for zero, one or two channels before the requirement is known. Each commissioned channel costs the stated price even if it is silent. A nonblank report is always correct. After reading the reports, choose a completion compatible with the foundation and with every world still possible. Resources are worst-case money and land in the catalogue's invented units.

In all four treatments either channel can fail individually. Under the separate-failure guarantee, at least one of two commissioned channels answers. Under unrestricted shared silence, a double blank is possible under either route requirement. In the two directional controls the pair may be blank only when east is needed, or only when west is needed. The team knows the contract before commissioning the channels. No probabilities, independence assumptions or observed ecological performance enter either treatment. The rule may use the decoded report East, West or Blank. A blank has exactly the meaning allowed by its contract. When two transcripts give the same possible worlds, their physical choices and up-front costs are identical, so a separate response to the channel's identity cannot improve the frontier.

Price per channel Separate-failure guarantee, money / land Shared silence, money / land
0.50 6 / 2, 8 / 1 7 / 4, 8 / 1
1.00 7 / 2, 8 / 1 7 / 4, 8 / 1
1.25 7 / 4, 7.5 / 2, 8 / 1 7 / 4, 8 / 1
1.50 7 / 4, 8 / 1 7 / 4, 8 / 1

Read one implementation

At a price of 1, reserve the junction for 4 money and 1 land, buy both channels for 2, and fit the direct route on the reported side for 1 money and 1 land. Either branch costs 7 / 2. At the same money the cheapest fixed crossing needs 4 land. Compact integration at 8 / 1 remains worthwhile where land is tight.

If a double blank is possible in both worlds, the rule for that observation must work in both worlds from the same early foundation. Take that fallback and use it in both worlds without commissioning a survey. Its resources fit within the original rule's worst-case envelope and it avoids the survey cost. This explains why the whole robust frontier reverts to the fixed frontier. More information could still improve typical outcomes under a different objective.

Close the long corridors and remove the compact site. At unit price, 7 / 2 survives when one answer is guaranteed. Unrestricted shared silence now makes the joint brief infeasible, although either route separately is buildable. The joint failure claim is itself a substantive assumption to test before relying on a second supplier.

The counterexample to the original reading

Suppose the pair can fall silent together only when the west route is needed. Reserve the junction and buy both surveys. Fit east after an east report, west after a west report, and west after a double blank. The same 7 money and 2 land suffice, even on the restricted site. Reversing the contract makes the correct blank response east. A single commissioned survey can still be blank in either world and cannot supply this distinction.

The table above gives the directional contracts in its first result column as well. At a channel price of 1.25, all three budgets in that column are genuine choices. At 1.5 the surveyed junction uses 8 / 2 and is dominated by the fixed compact crossing at 8 / 1. At zero price, several distinct implementations can attain one budget. The reader retains each of those choices.

For each possible transcript, the channel contract supplies a set of possible route requirements. A valid completion must serve every member of that set and fit the foundation already chosen. Asking suppliers which transcripts remain possible on each route gives the design team the information it needs to apply that test. The restricted site requires an unambiguous side before construction. A survey guarantee may supply that distinction even when neither channel returns a reading. This is a stipulated support contract, whose reliability a real provider would need to substantiate.

Method and limits

The trial enumerates 124 admissible decoded rules under the separate-failure contract and 174 when shared silence is allowed, with the default sites. The extra rules in the latter case include policies with an explicit fallback. Their count says nothing about usefulness. All 48 combinations of four contracts, two sites and six prices are compared with the original solver's timely-survey and fixed controls. Every retained policy is checked against every world that admits each of its readings. Censi's implementation and resource orders supply the formal comparison. The trial models bounded erasure and guaranteed success. An expected-cost analysis would need frequencies and a different objective.

Shared project end · A common beginning · Keep the question answerable · Censi on co-design · Huang and Zardini on adaptation

Reproduce

From the repository root run node project-co-design/research/silent-surveys/run.cjs. The recorded frontiers preserve implementation names, observation-dependent choices and the restricted-site control.