Theory of Collaborative Intelligence

Status rule

A proposition is marked Supported only where a published round's own text reports a supporting finding, and that round is cited. Everything else is Open. Supported means supported in the bounded setting the round describes, not proven in general.

Working propositions (v0)

1. v0.1

   Reliability is a property of the process

   Reliability is a property of the system's process, not only of the underlying model.

   Working position from Theory v0, not yet tested as stated.

   Open
2. v0.2

   Finding evidence and judging it are separate steps

   Locating a quotation and judging whether it supports a claim are separate operations.

   Working position from Theory v0, not yet tested as stated.

   Open
3. v0.3

   Confidence belongs to the final answer

   The object that should carry confidence is the final revised answer, not the first draft.

   Working position from Theory v0, not yet tested as stated.

   Open
4. v0.4

   Improvement counts only against a matched baseline

   An architecture change counts as an improvement only when measured against a matched baseline with the same number of calls on an adjudicated endpoint.

   Working position from Theory v0, not yet tested as stated.

   Open

257 open hypotheses; none yet tested. These propositions are positions to be tested, not results.

Round 40 (NCT-40) propositions

1. P1

   Structural abstention

   A composition can hold only if some component can produce HOLD or a disagreement can produce review.

   Round 40 reports H40.1 as Supported (E40-A: a forced majority releases 108/243 states unsafely, the same as one witness, and holds 0/16 HOLD-truth states without abstaining witnesses).

   Round 40Round 39

   Supported
2. P2

   Read and quote, then decide

   Models should locate and quote; a deterministic, span checked resolver should decide.

   Round 40 reports H40.2 as Supported (E40-A: 4/12 unsafe with the span check versus 8/15 without).

   Round 40Round 39

   Supported
3. P3

   Disagreement buys a hold, not a vote

   The second independent read should only be able to stop a release.

   Round 40 reports H40.3 as Supported (E40-A: disagreement routing raises the fewest faults for an unsafe release to 2). The false hold cost was not observed.

   Round 40

   Supported
4. P4

   Decorrelation is the safety budget

   A routed topology is as safe as its readers are decorrelated.

   Round 40 notes all 4 unsafe routed states are identical misreadings; joint misreading rates are not yet measured.

   Round 40

   Open
5. P5

   Evidence-rewriting stages add risk

   Stages that can rewrite evidence add risk.

   Round 40 reports H40.4 as Supported (E40-C: on X24 a validation stage that rewrote a correct ledger produced the only composed error).

   Round 40

   Supported

Round 40 names each proposition as a candidate for the Round 41 to 50 block, with a removal test that could reject it.

Roadmap to Round 55 · currently Round 54

1. Round 41

   Hard pool: items on which witnesses disagree, so topologies separate on accuracy

   published
2. Round 42

   Measured decorrelation: joint misreading rates across witness pairs and prompts (P4)

   published
3. Round 43

   Record conventions and flag only validators versus rewriting validators (P5)

   published
4. Round 44

   Record-level status: abstention at scale with externally reviewed HOLD labels (P1, Round 39 handoff item 1)

   published
5. Round 45

   Resolver versus reader: where fixed rules are wrong and witnesses are right (P2)

   published
6. Round 46

   Review routing load: false hold rates and what a reviewer must see (P3)

   published
7. Round 47

   Supersession and independent review: when a replacement record is not a cure

   published
8. Round 48

   Independent review and successor evidence on contract packets and a second corpus (scientific papers with citations or policy documents). The reviewer extracts and decides from the packet before seeing any brief, and separately scores whether one item replaces another versus whether the underlying requirement is actually met, testing whether the Round 47 finding transfers beyond contracts.

   IRE-48: Extract First, or Brief First? · DOI 10.5281/zenodo.24009036

   published
9. Round 49

   Working memory and procedure reuse (the item Round 47 skipped), in two domains: when prior ledger and procedure reuse helps, and when it poisons later decisions.

   WMR-49: Remember the Answer, or Remember the Procedure? · DOI 10.5281/zenodo.24036028

   published
10. Round 50

    Open removal tests, not another contract refinement: joint error when both roles can fail in the same direction, and rewrite gate behavior on genuinely wrong rewrites.

    ORT-50: When Both Readers Are Wrong: Open Removal Tests for Review and Rewriting · DOI 10.5281/zenodo.24036963

    published
11. Round 51

    A third clearly different task, not another document-and-clause packet: code review against a spec, numerical or experimental claims with raw tables, or news claims against primary sources. Propositions that fail transfer cannot enter the broader theory unchanged.

    XFR-51: Valid Evidence, Wrong Verdict: Transfer Tests on Code Review and Raw Table Claims, with Independent Recovery · DOI 10.5281/zenodo.24053367

    published
12. Round 52

    Matched call baselines for surviving propositions, so improvements count only against a fair resource comparison.

    MCB-52: The Second Reader or the Second Call? Matched Call Baselines for Agreement Routing, Validation and Recovery · DOI 10.5281/zenodo.24081926

    published
13. Round 53

    Candidate theory statements in domain-general language about roles, evidence, disagreement, review and memory, tested by trying to break them on leftover mixed cases. No contract-only language in the statements.

    DGP-53: Stated Plainly, Tested to Break: Domain General Candidate Propositions on Disagreement, Review, Memory and Evidence, on Held Out Mixed Cases · DOI 10.5281/zenodo.24134801

    published
14. Round 54

    Held-out replication across the standing domains, with new packets and surface forms, on the same propositions.

    planned
15. Round 55

    Theory v1 draft, a multi-day round, retaining only what survives Rounds 40 to 54. Contract and other-domain results support statements; they do not define the theory's language. If the evidence fails, narrow or defer claims rather than force a general theory.

    planned

Round 40 originally scheduled Round 48 transfer, Round 49 matched baselines and Round 50 Theory v1. That schedule was not met because later rounds followed the real next checks. This roadmap supersedes the forward plan, not the published history. Contracts are only one test bed: the aim is a broader domain-general theory, not a claim of universal proof or a guaranteed result. From Round 48 onward every round uses at least two kinds of material, one of them neither contracts nor SEC exhibits.

A proposed research path, not a published commitment to particular results.