Encevra SI Research Labs
Index

Notebook library

Research notebooks

Every research round has its own notebook. Pull one off the shelf to read its question, stated findings and effect on Noetis. Full experimental notes stay with the research team.
  1. 54E54-A structural diagnostic, E54-B live blind label audit and independent, repeated, resampled and deliberate reads, E54-C live rewriting review, flag only review and seeded proposals, and E54-D live stored conclusions and a stored procedure. Staged for review, not yet published. DOI reserved (10.5281/zenodo.24149599).Research paper · held out replication of seven domain general candidate propositions on new packets and new surface forms

    The Same Statements, New Cases: Held Out Replication of Domain General Propositions on New Packets and New Surface Forms

    Collaborative Intelligence · Artificial Cognitive Architecture · Reasoning & Verification · Evaluation Science10 findings Full notes private
  2. 53E53-A structural diagnostic, E53-B live blind label audit and independent, repeated, resampled and deliberate reads, E53-C live rewriting review, flag only review and seeded proposals, and E53-D live stored conclusion and stored procedure memory arms, on 188 items. All planned experimental calls attempted; 100 first outputs were unusable and were rerun once; 7 of those were rerun a second time; 3 remain empty and are disclosed. Exact run replay: 1367 checks pass. H53.1, H53.2, H53.3, H53.4, H53.5, H53.6, H53.10, H53.11 and H53.12 supported; H53.7, H53.8 and H53.9 not supported. DOI 10.5281/zenodo.24134801, Zenodo record published. No production system changed.Research paper · structural diagnostic and live tests of seven domain general candidate propositions on disagreement, review, memory and evidence, on replicated packets, held out non contract families and real SEC employment exhibits

    Stated Plainly, Tested to Break: Domain General Candidate Propositions on Disagreement, Review, Memory and Evidence, on Held Out Mixed Cases

    Collaborative Intelligence · Artificial Cognitive Architecture · Reasoning & Verification · Evaluation Science9 findings Full notes private
  3. 52E52-A structural diagnostic, E52-B live blind label audit, independent reads, resampled reads, one call deliberate reads and procedure card reads, and E52-C live cross and self validation with replayed routing, majority and recovery at matched call counts. All planned experimental calls attempted once; 21 of 797 outputs were unusable (one error, twelve empty, eight incomplete) and were not repeated; the remaining phases were reordered once before any of their outputs existed, and no call was stopped by the spend guard. Exact run replay: 395 checks pass. H52.1, H52.2, H52.4, H52.5 and H52.7 supported; H52.3, H52.6 and H52.8 not supported. DOI 10.5281/zenodo.24081926, Zenodo record published. No production system changed.Research paper · structural diagnostic and matched call baseline study of agreement routing, validation and recovery against same reader resampling, self validation and one call deliberation, on replicated Round 49 near miss packets, authored near threshold raw table claims and real SEC merger agreement termination fee claims

    The Second Reader or the Second Call? Matched Call Baselines for Agreement Routing, Validation and Recovery

    Collaborative Intelligence · Artificial Cognitive Architecture · Reasoning & Verification · Evaluation Science8 findings Full notes private
  4. 51E51-A structural diagnostic, E51-B live blind label audit and independent reads, and E51-C live validation of natural baselines with replayed preservation and recovery policies. All planned experimental calls attempted once; one reader output was empty and not repeated, and the first synthesis call returned no text before a second completed. One authored label (X01H) was found wrong after the live reads; both scorings are reported. Exact run replay: 372 checks pass. H51.1 and H51.6 supported; H51.2 and H51.3 not supported; H51.4 and H51.5 inconclusive. DOI 10.5281/zenodo.24053367, Zenodo record published. No production system changed.Research paper · structural diagnostic, transfer tests of evidence, agreement and rewrite checks on code review against a specification and numerical claims against raw tables, and real SEC covenant threshold claims

    Valid Evidence, Wrong Verdict: Transfer Tests on Code Review and Raw Table Claims, with Independent Recovery

    Collaborative Intelligence · Artificial Cognitive Architecture · Reasoning & Verification · Evaluation Science8 findings Full notes private
  5. 50E50-A structural diagnostic, E50-B live blind review, procedure generation and independent reads, and E50-C live ledger rewrites. All phases attempted and the round is complete, with five individual outputs (two empty, three truncated) disclosed as unavailable and not repeated; not every call completed. Exact run replay: 194 checks pass. H50.3, H50.4, and H50.6 supported; H50.1 and H50.2 not supported; H50.5 consistent with its safety benefit unmeasured. DOI 10.5281/zenodo.24036963, Zenodo record published. No production system changed.Research paper · structural diagnostic, open removal tests of the second read, procedure memory and a status preservation gate on contract and non-contract policy packets, and real SEC exhibit windows

    When Both Readers Are Wrong: Open Removal Tests for Review and Rewriting

    Collaborative Intelligence · Artificial Cognitive Architecture · Reasoning & Verification · Evaluation Science8 findings Full notes private
  6. 49E49-A structural diagnostic, E49-B contract memory study with stored commitment lock, E49-D quality release record memory study, and E49-C on 24 new SEC exhibits with three note styles executed. H49.2, H49.4, and H49.7 supported; H49.1, H49.3, H49.5, H49.6, and H49.8 not supported as stated. No production system changed.Research paper · structural diagnostic, working memory and procedure reuse study in contract and quality release record packets, stored commitment lock, and real SEC exhibit note styles

    Remember the Answer, or Remember the Procedure?

    Artificial Cognitive Architecture · Collaborative Intelligence · Evaluation Science7 findings Full notes private
  7. 48E48-A structural diagnostic, E48-B Domain A live discharge and extract first study, E48-D Domain B clinical packets, and E48-C on 24 new SEC exhibits executed; 15 Witness B ledgers and the synthesis were completed the same day as a disclosed post hoc completion. H48.1 supported on gold and on live Witness B ledgers, H48.2 supported, H48.5 supported but uninformative, H48.3, H48.4, and H48.6 not supported as stated. No production system changed.Research paper · structural diagnostic, cure discharge study, extract first independent review, dual domain clinical packets, and real SEC exhibit validators

    Extract First, or Brief First?

    Reasoning & Verification · Collaborative Intelligence · Evaluation Science6 findings Full notes private
  8. 47E47-A structural diagnostic and E47-B live supersession, routing, and independent reviewer study executed, with the budget tail completed the same day as a disclosed post hoc completion; E47-C executed on 24 new SEC exhibits. H47.2 supported, H47.4 supported but uninformative, H47.1, H47.3, and H47.5 rejected. No production system changed.Research paper · structural diagnostic, supersession study, disagreement routing, independent reviewer views, and real SEC exhibit validators

    When a Replacement Is Not a Cure: Supersession, Schedule Thresholds, and Independent Review

    Reasoning & Verification · Collaborative Intelligence · Evaluation Science6 findings Full notes private
  9. 46E46-A (structural certificates, gold ledgers under every rule set, scripted reader mistakes, the routing certificate, and exhibit and gate certificates, 0 model calls) and E46-B (blind label review, record set ledgers, direct readings, four routing compositions, and three reviewer views on 5 authored families in 20 variants, 141 completed calls; 12 Witness B ledgers stopped by the planned budget tail were completed post hoc the same day) and E46-C (48 validator calls on 24 real SEC exhibits) executed; H46.1 (post hoc completion), H46.2, H46.4, and H46.5 supported, H46.3, H46.6, and H46.7 rejected; no production system changedResearch paper · structural diagnostic, blind label review, live consistency test study with four fixed rule sets, disagreement routing, reviewer views, and real SEC exhibit validators

    Who Must Look Again? Disagreement Routing, False Hold Load, and What a Reviewer Must See

    Reasoning & Verification · Collaborative Intelligence · Evaluation Science8 findings Full notes private
  10. 45E45-A (structural certificates, gold ledgers under every rule set, scripted reader mistakes, and exhibit and gate certificates, 0 model calls) and E45-B (blind label review, record set ledgers by Witnesses B, C, and D, and direct readings on 6 authored families in 30 variants, 152 completed calls) and E45-C (72 validator calls on 24 real SEC exhibits) executed, with the phases after the first budget stop completed the same day (post hoc); H45.1 to H45.4 and H45.6 supported, H45.5 rejected, H45.7 not supported live; no production system changedResearch paper · structural diagnostic, blind label review, live record set ledger study with four fixed rule sets, direct reader comparison, and real SEC exhibit validators

    The Rule or the Reader? Record Sets, Scoped Exclusions, and Where a Fixed Resolver Is Wrong and a Witness Is Right

    Reasoning & Verification · Collaborative Intelligence · Evaluation Science13 findings Full notes private
  11. 44E44-A (eight structural certificates, 0 model calls), E44-B (blind label review, record level ledgers by three witnesses, condition ledger baseline, and direct readers on 8 authored families in 24 variants, 128 completed calls), and E44-C (144 validator calls on 24 real SEC exhibits) executed; budget stopped phases completed later with unchanged settings; H44.1, H44.3, H44.5, and H44.6 supported, H44.2 preliminary, H44.4 inconclusive; no production system changedResearch paper · structural diagnostic, blind label review, live record level ledger study, and fixed real SEC exhibit track

    This Record Fails, or No Record Exists? Record Level Status Ledgers, Completeness Records, and Independently Reviewed HOLD Labels

    Reasoning & Verification · Collaborative Intelligence · Evaluation Science7 findings Full notes private
  12. 43E43-A (seven structural certificates, 0 model calls), E43-B (6 authored families in 18 variants, 201 calls, 3 stopped by the budget rule), and E43-C (24 fixed real SEC exhibits, 96 calls) executed; H43.2 to H43.5 supported; H43.1 rejected as stated (holds on U, not on T); no production system changedResearch paper · structural diagnostic, live declared convention study, and real SEC exhibit validator track

    Open Record or Closed Record? Declared Record Conventions, Near Miss Records Read as Negation, and Flag Only Against Rewriting Validators

    Reasoning & Verification · Collaborative Intelligence · Evaluation Science6 findings Full notes private
  13. 42E42-A (structural certificates and Frechet bounds, 0 model calls), E42-B (6 authored contrast families, 72 calls), and E42-C (20 fixed real SEC exhibits, 120 calls) executed, 0 missing; H42.1, H42.2, H42.3, and H42.5 supported; H42.4 not testable; no production system changedResearch paper · structural diagnostic, live contrast families read in two roles, and real SEC exhibit track

    Role or Witness? Measured Joint Misreading of Undetermined Facts by One Witness in Two Roles and by Two Witnesses in One Role

    Artificial Cognitive Architecture · Evaluation Science · Reasoning & Verification6 findings Full notes private
  14. 41E41-A (exact selection enumeration, 0 model calls), E41-B (10 authored twin families), and E41-C (20 fixed real SEC exhibits) executed with 200 live calls, 0 missing; H41.1 and H41.2 supported structurally only; H41.3 rejected; H41.4 and H41.5 supported; no production system changedResearch paper · exact selection diagnostic, live isomorphic twin screen and holdout, and real SEC exhibit track

    Hard by Construction or Hard by Selection? Disagreement Mined Item Pools, Isomorphic Twin Holdouts, and the Consensus Errors That Disagreement Cannot Find

    Evaluation Science · Artificial Cognitive Architecture · Reasoning & Verification6 findings Full notes private
  15. 40E40-A (exact topology enumeration, 0 model calls), E40-B (17 authored packets), and E40-C (16 real SEC exhibits) executed with 330 live calls, 0 missing; H40.1 to H40.4 supported; H40.5 partly supported; no production system changedResearch paper · exact topology diagnostic, live six-topology benchmark, and real SEC exhibit track

    Composition Before Capability: Six Noetis Composition Topologies Compared on Release Safety, Abstention, and Calls per Decision

    Artificial Cognitive Architecture · Evaluation Science · Reasoning & Verification6 findings Full notes private
  16. 39E39-A (112 exact states), E39-B (96 live witness evaluations, 0 mocks), and E39-C (33 completed evaluations on 24 real SEC exhibits) executed; H39.1-H39.3 and H39.5-H39.7 supported; H39.4 rejected; no production system changedResearch paper · exact finite diagnostic, live multi-model benchmark, and real SEC exhibit track

    When the Records Disagree: Conflicting Collateral Evidence, Evidentiary Lens Balancing, and Abstention Failure in Multi-Witness Contract Audits

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture7 findings Full notes private
  17. 38E38-A (512 raw / 464 feasible post-TDR-37 states) and E38-B (96 live witness evaluations across 4 role-decorrelated architecture families, 0 mocks) executed; H38.1-H38.6 supported; no production system changedResearch paper · exact finite diagnostic and live multi-model benchmark

    Signed Is Not Operative: Conditional Effectiveness, Counterpart Authority, and Sub-Clause Scope Partition in Multi-Witness Contract Audits

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture5 findings Full notes private
  18. 37Completed exact diagnostic, live multi-model benchmark, and stale-register probeResearch paper · exact finite diagnostic and live multi-model benchmark

    When the Contract Changed After Signing: Event-Date Resolution, Register-Anchored Completeness, and Stale Registers in Multi-Witness Contract Audits

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture3 findings Full notes private
  19. 36Completed exact diagnostic and live matched-compute multi-model benchmarkResearch paper · exact finite diagnostic and live multi-model benchmark

    When One Complete Document Is Still Half the Truth: Cross-Document Reference Closure, Precedence DAGs, and Compositional Clause Chains Across Multi-Instrument Quorums

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture2 findings Full notes private
  20. 35Completed exact diagnostic and live matched-compute multi-model benchmarkResearch paper · exact finite diagnostic and live multi-model benchmark

    When Grounded Witnesses Still Agree on a Falsehood: Correlated Semantic Failures, Verbatim Span Grounding, and Matched-Compute Multi-Model Quorums

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture2 findings Full notes private
  21. 34Completed exact finite diagnostic · model-facing pilot BLOCKEDResearch paper · exact finite trust-graph diagnostic

    Can Trust Bootstrap Itself? Circular Attestation, Sybil Issuers, and the Minimum Grounded Trust Base

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture2 findings Full notes private
  22. 33Completed exact finite diagnostic · model-facing pilot BLOCKEDResearch paper · exact finite certification diagnostic

    Who Certifies Independence? Provenance Completeness and the Limits of Self-Attested Witness Quorums

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture2 findings Full notes private
  23. 32Completed sandbox diagnostic · model-facing pilot BLOCKEDResearch paper · authored synthetic sandbox diagnostic

    Who Watches the Witness? Boundary Hardening and a Blinded Mutation Holdout for the RISE Gate

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture2 findings Full notes private
  24. 31Sandbox conformance executed · live E30-B BLOCKEDResearch paper · authored sandbox conformance diagnostic

    Did the Root Actually Move? A Signed Sandbox Dry Run of the RISE Gate

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture2 findings Full notes private
  25. 30Readiness diagnostic executed · BLOCKED · shadow experiment proposedResearch paper · deterministic repository-readiness diagnostic

    Can We Intervene for Real? A Fail-Closed Readiness Gate for Active Evidence-Lineage Trials

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture2 findings Full notes private
  26. 29Exact finite diagnostic executed · live experiment proposedResearch paper · exact finite diagnostic

    Round 29 - Make the Evidence Move: Active Source Interventions and the Limits of Dependency Discovery

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture2 findings Full notes private
  27. 28Exact constructed trace replay executed · active intervention and live-agent transfer untestedResearch paper · exact constructed trace replay

    Round 28 - Confidence About What? Separating Target Probability, Structural Ambiguity, and Model Adequacy

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture2 findings Full notes private
  28. 27One exact finite constructed diagnostic executed · one shadow experiment proposedResearch paper · finite-model diagnostic

    Round 27 - Which Check Should Run Next? Model-Set-Safe Verification Under a Fixed Budget

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture2 findings Full notes private
  29. 26One finite-model identifiability analysis executed · one prospective shadow experiment proposedResearch paper · finite-model diagnostic

    Round 26 - What Can the Watchers Ever Know? Minimum Observables for Detecting Omitted Change

    Reasoning & Verification · Artificial Cognitive Architecture · Evaluation Science2 findings Full notes private
  30. 25One bounded adversarial-omission diagnostic executed · one prospective experiment proposedResearch paper · bounded diagnostic

    When Silence Looks Like Stability: Detecting Missing Change Evidence in Collaborative AI Systems

    Reasoning & Verification · Artificial Cognitive Architecture · Evaluation Science1 finding Full notes private
  31. 24One bounded source-authority audit and authored conformance diagnostic executed · one prospective experiment proposedResearch paper · bounded diagnostic

    Do Two Watchers Count as Two? Authority Lineage and Common-Origin Failure in Claim Revalidation

    Reasoning & Verification · Artificial Cognitive Architecture · Evaluation Science1 finding Full notes private
  32. 23One bounded diagnostic executed · one larger prospective experiment proposedResearch paper · bounded shadow diagnostic

    Can the Gate Catch Tomorrow's Change? A Prospective Shadow Trial of Claim Revalidation

    Reasoning & Verification · Artificial Cognitive Architecture · Evaluation Science1 finding Full notes private
  33. 22Paper complete · diagnostic executed · prospective experiment proposedResearch paper · deterministic repository diagnostic · Version 1.0

    When Does an Audit Expire? Dependency-Triggered Revalidation for Rapidly Changing Agent Systems

    Reasoning & Verification · Artificial Cognitive Architecture · Evaluation Science3 findings Full notes private
  34. 21Paper complete · diagnostics executed · shadow experiment proposedResearch paper · deterministic development diagnostics · Version 1.0

    Where Did the Trace Break? An Attempt-Bound, Fail-Closed Measurement Gate for Noetis

    Reasoning & Verification · Artificial Cognitive Architecture · Evaluation Science2 findings Full notes private
  35. 20Paper complete · diagnostics executed · held-out study proposedResearch paper · deterministic development diagnostics · Version 1.0

    What Must the Log Reveal? Minimal-Disclosure Receipts for Verifiable Evidence Delivery

    Reasoning & Verification · Artificial Cognitive Architecture · Evaluation Science2 findings Full notes private
  36. 19Paper complete · diagnostics executed · reconstruction study proposedResearch paper · deterministic diagnostics and implementation audit · Version 1.0

    Did the Evidence Reach the Agent? Reconstructable Evidence Lifecycles for Reliable Experiments

    Evaluation Science · Artificial Cognitive Architecture · Reasoning & Verification2 findings Full notes private
  37. 18Paper complete · diagnostic executed · model study proposedResearch paper · deterministic constructed diagnostic · Version 1.0

    When Evidence Changes: Versioned Support and Bounded Belief Repair

    Artificial Cognitive Architecture · Reasoning & Verification · Evaluation Science2 findings Full notes private
  38. 17Paper complete · diagnostics executed · model study proposedResearch paper · finite constructed diagnostics · Version 1.0

    When the Question Changes: Recoverable Evidence Under a Context Budget

    Collaborative Intelligence · Artificial Cognitive Architecture · Evaluation Science2 findings Full notes private
  39. 16Paper complete · diagnostics executed · model study proposedResearch paper · finite-world diagnostics · Version 1.0

    Before the Answer Closes: Detecting Missing Decision Conditions

    Reasoning & Verification · Evaluation Science · Artificial Cognitive Architecture3 findings Full notes private
  40. 15Internal research complete · experiments proposedSummary only · internal notebook

    What makes a message worth sharing?

    Collaborative Intelligence · Artificial Cognitive Architecture · Evaluation Science3 findings Full notes private
  41. 14Published · experiment proposedPaper and exact diagnostics

    When the judges are uncertain: measuring revision gains without an infallible evaluator

    Evaluation Science · Reasoning & Verification3 findings Full notes private
  42. 13Published · pilot proposedAudit memorandum

    From final scores to measurable corrections

    Evaluation Science · Reasoning & Verification5 findings Full notes private
  43. 12Published · experiment proposedPaper and proposed experiment

    After the last revision: calibrating release risk

    Reasoning & Verification · Evaluation Science1 finding Full notes private
  44. 11Published · experiment proposedExpanded revision 2

    When should collaborators spend their next unit of compute?

    Artificial Cognitive Architecture · Reasoning & VerificationNo stated findings Full notes private
  45. 10Published · experiment proposedWorking paper

    Evidence exchange, selective revision, and a testable path

    Collaborative Intelligence · Artificial Cognitive ArchitectureNo stated findings Full notes private
  46. 09Published · agendaResearch agenda

    A research agenda for the lab

    Artificial Cognitive ArchitectureNo stated findings Full notes private
  47. 08Published · noteWorking paper

    Delayed feedback can act on a temporary trace

    Artificial Cognitive ArchitectureNo stated findings Full notes private
  48. 03Published · reviewReview

    Hippocampal ripples and compositional inference

    Artificial Cognitive Architecture · Collaborative IntelligenceNo stated findings Full notes private

These are the public records of each round. Full experimental notes, prompts and implementation procedures are kept by Encevra SI Research Labs.