THE HEARING / THIRD-ROUND RESPONSE REGISTER
30 October 2026

Nine fresh independent AI role simulations. Three used the interface to varying extents; six read a frozen player-screen packet. No human reviewers participated. All original accepted reports are preserved without editing. Numbers below correspond to each report. Partial/deferred items identify what was not implemented.

T1 / LOW-PREPARATION FIRST-YEAR

- 1 Implemented: the neutral observation/action/option example and plain teaching purpose are available at entry and connection. Historical prose remains authored by the learner.
- 2 Partial: claim-specific limits now pair a positive recorded point with a plain boundary; baseline, denominator and other terms are available in the glossary. A pervasive inline translation of adjudicate/capacity is not added; unfamiliar-word testing remains in the pilot.
- 3 Retained: local card/table/source ID help and shared-source warnings. Source records and original links remain; opening two IDs is not independent corroboration.
- 4 Implemented: short target, immediate next action and complete requirements remain close to writing. Final self-check next actions now use imperatives rather than past-tense claims. No analytical gate removed.
- 5 Retained/strengthened: the fictional library example and existing request field model different expected patterns. The request prompt now asks explicitly what differs under the accounts; no historical answer is supplied.
- 6 Implemented: a compact dated Malawi sequence separates challenge, court/appeal, implementation and winner. Directly recorded sequence does not establish compliance motivation.
- 7 Partial: full source provenance is preserved, with the lengthy MEC numerical audit behind a labelled disclosure and a short relevant notice. Card-specific limits stay visible in source detail. The entire shared-record architecture is not replaced.
- 8 Partial: positive warrants, target and human-understanding prompts counter checklist thinking. Existing open counters and truthful labels remain; moving every counter to a new layout is deferred. Neither completion nor record opening claims mastery.

T2 / HIGH-SCHOOL STUDENT

- 1 Implemented and sampled in informed retest: visible office/holder, challengers/checks and opportunity map, including BN/PH and seat-control context, before expandable background.
- 2 Partial: plain source-limit/baseline wording, glossary definitions including civil forfeiture/misappropriation and local source guidance remain available. Not every technical term gains a new inline tooltip.
- 3 Implemented in part: card-specific supported point and limit lead the relevant source guidance; complete provenance/links remain. The detailed MEC audit is layered. Other records retain full metadata; no universal abbreviated metadata system is claimed.
- 4 Implemented: complete full-hearing contract includes target, complication, connection, both challenge/decision cycles, authored paragraphs, rival/request and review. Optional reflection stays optional.
- 5 Retained/strengthened: neutral library model and changed-option cue, with earlier situation explicitly an inference if unrecorded. No invented historical before-condition is introduced.
- 6 Implemented and sampled: choose reading, check/read feedback, decide about own explanation. One-at-a-time or Show both agenda; independent attempt and decision statuses. Both questions still required.
- 7 Implemented: final confirmation links state an action. Target, own connection and a blank comparative sentence frame carry into the second case. Plain comparison wording replaces failed analogy.

T3 / ADVANCED UNDERGRADUATE

- 1 Implemented: required short learner target with actor/period cue. It persists, travels into comparison/report, and is reviewed alongside the paragraph. It is a structural presence check, not semantic precision grading.
- 2 Implemented: Malawi sequence and card-specific positive warrant/limit separate remedy, appeal and implementation from the unestablished reason for compliance. Indirect judgment access remains explicit.
- 3 Implemented: decision reasons address a phrase or link in the learner paragraph/connection and cite a case ID. A current-connection inspector assists self-review. Recognition is not treated as an authored argument.
- 4 Implemented: both accounts name one uncertain link; the existing request asks for contrasting expected patterns, actor and period. Plausible alternative hypotheses need evidence, not automatic equal weight.
- 5 Partial: positive observation-specific warrants, original producer/locator data, shared-source warning and layered MEC audit. No new source graph is added; every original source record remains.
- 6 Implemented without more required prose: short target and existing separate connection/rival/request fields support planning; main 60-85 and comparison 30-45 ranges remain. No additional multi-field worksheet is imposed.
- 7 Implemented: named target/connection, neutral purpose labels, topic index and event/period/option/difference comparison frame. The full six-card file remains; comparable procedure is not assumed.
- 8 Retained/strengthened: honest minimum opening threshold plus substantive used-source confirmation and human criteria. Self-check now includes target/paragraph consistency; a board change still requires the learner to revise prose.

T4 / CURIOUS NONSTUDENT ADULT

- 1 Adapted: readable drafts can be saved at any point in casual exploration, explicitly unfinished when requirements are missing. A smaller draft is not relabelled a completed hearing; full class and casual completion use equal gates.
- 2 Implemented in part: entry names the concrete evidence/opportunity/revision task and reusable research skill. A new sample saved historical answer is not supplied.
- 3 Implemented: office/holder, challengers/checks and opportunities appear before opening choices. Serbia distinguishes parliamentary seats from the campaigning president.
- 4 Partial: the new power map and case hook give a plain orientation. The historical main Serbia question is retained rather than replaced with a question that presupposes an electoral effect.
- 5 Implemented in part: recorded point plus source-supported observation/claim boundary. Full original card prose and provenance remain; some repetition is intentionally retained rather than deleting information.
- 6 Implemented: Check this source reading, explicit answer/feedback/own-decision sequence, and focused agenda. Existing reasoned keep/narrow/replace semantics remain.
- 7 Adapted: learner target, connection and topic index guide a full second file. All six cards remain equally accessible; the app does not hide four cards or choose a comparison for the learner.
- 8 Implemented in part: review/sharing and class-specific hand-in language now distinguish settings; casual draft banner clarifies stopping. Human judgment remains prominent rather than silently treating completion as quality.

T5 / EDUCATIONAL INSTAGRAM CREATOR LENS

- 1 Implemented: concrete first action, transferable purpose and short case hooks. The full 55-minute plan remains an estimate; exploration can pause/save. No short clip is advertised as completing the hearing.
- 2 Partial: three-space public-attention rationale is visible in evidence selection and the existing board. A second decorative hero-board mockup is deferred; all six cards remain readable.
- 3 Adapted: Explore at your pace and Save readable draft give voluntary exploration a stopping point. Existing Just for fun mode name and saved-hearing identity controls are retained, avoiding an unrelated navigation redesign.
- 4 Implemented in part: visible target, focused question agenda, next-action guidance, current-connection inspection and positive source/limit pairing. Full guidance and human/inclusion distinction remain; not every stage is visually rebuilt.
- 5 Implemented: dated Malaysia sequence distinguishes 2013 PR, May 2018 PH, and later reform; PH 113 and separate Warisan seats remain. Existing district/denominator explanations and non-fraud caveats remain.
- 6 Retained/strengthened: visible Serbia institutional map; card records and warrants distinguish stated rationale, activity, monitored content and investigated operation. Full campaign-slot/ordinary-news and voter-effect limitations remain; no new label implies nationwide measurement.
- 7 Implemented: Practice claim supplied by the activity label and separate own-explanation decision prompt. Both source questions, board-based selection and shared rival/request response remain.
- 8 Implemented: target and own action/option travel, empty comparison frame and neutral report labels. Report versus editable resume file and local finish/teacher hand-in boundary remain distinct. No posting or endorsement is claimed.

T6 / AVERAGE WORKING FIRST-YEAR

- 1 Implemented: entry states observation/inference, action/opportunity, rival and comparison in plain language, with concrete first actions.
- 2 Implemented: visible Serbia map identifies parliament, governing list, president campaigning without standing, and opposition/challenge institutions.
- 3 Implemented in part: baseline and source-limit language are plainer at use; definitions remain. Full research terms are preserved for discussion.
- 4 Adapted: an explicitly invented ballots/valid votes/seats denominator example is in the glossary. A full invented seat-allocation algorithm exercise is deferred; the file does not reconstruct Serbia 129 seats.
- 5 Implemented: recorded point, producer/locator and observation-specific warrant/claim boundary, with original source link and full limitations retained.
- 6 Adapted: the single existing request field now explicitly includes my prediction, the other prediction and what/whom/when to check. Neutral worked guidance remains; three extra required fields are not added.
- 7 Implemented: second-case topic index follows the learner target and keeps every card available. Compare asks for the event/period and a difference that bounds/tests the chosen claim.

T7 / INSTRUCTIONAL DESIGNER

- 1 Implemented: optional unprompted closing reflection asks a supported statement, further inference and reusable check. Guide includes later unfamiliar material; neither response nor multiple choice proves transfer.
- 2 Implemented: each hearing reason refers to own writing/link and evidence; current-connection inspector available while drafting. Human pilot checks whether recognition reaches prose, without new auto-grading.
- 3 Implemented: rivals concern an uncertain motivation, implementation or broader-effect link. Keep directly recorded events/rules rather than manufacture doubt.
- 4 Implemented: target and own connection carry; blank frame and topic index distinguish similar opportunity, different stage and limiting difference. No automatic causal verdict.
- 5 Implemented: full in-game cards/source details used are core; external original links are optional verification. Opening remains a weak mechanical indicator, with substantive human confirmation.
- 6 Implemented in part: visible basic power relationships, positive card warrants, specific limits, Malaysia/Malawi timelines and layered detailed audit. Full provenance remains; no information deletion.
- 7 Retained with support: word ranges are unchanged; short target, separate planning fields and optional support reduce compression burden. No broad exemption that lowers analytical requirements.
- 8 Implemented: six-page guide includes phase-time/help notes, reasoning observation sheet, carryover point and two-session option. 55 minutes remains conditional and unmeasured.

T8 / POLITICAL-SCIENCE FACULTY

- 1 Implemented: one named event/opportunity, actor and period before connection; carries into comparison and final self-review. Different office/outcome measures remain explicitly non-interchangeable.
- 2 Implemented in part: optional Russia reported-backing actor/action starters added; positive warrants keep supported survey observations legitimate. Missing effect links must be named. No seat decomposition or unestablished electoral effect is invented.
- 3 Implemented: dated institutional power maps and topic routes preserve different entry, organization, information and remedy configurations. Competitive-authoritarianism concept note is separate; it does not classify every case alike.
- 4 Partial/retained: distinct-observation criterion, source maker/origin metadata and shared-document warnings. The app does not require different publishers. A new mandatory source-independence response is deferred to avoid another field; teacher sheet checks distinct supporting observations.
- 5 Implemented: common uncertain-link rival framing and differing expected observations; documented facts need not be denied. Alternative possibility does not receive equal evidentiary weight automatically.
- 6 Implemented: Malawi opening now says the reported 2020 winner held both majority and largest share. The threshold alone would not choose a different winner; possible strategic effects require further evidence.
- 7 Implemented: all 24 cards have a positive source-supported point paired with a claim-specific boundary. The full original text and source qualifications remain.
- 8 Implemented: optional own-words reflection and human pilot/deferred transfer plan. Completion, substantive participation and mastery stay separate; no causal learning claim.

T9 / GAME DESIGN, UX AND ART DIRECTION

- 1 Implemented: short research hook on every case choice, keeping dates, offices and known outcomes. It frames an inquiry, not a country ranking.
- 2 Partial: evidence selection makes the public-attention limit explicit. Existing replacement/history remains; a new pre-commit outgoing/incoming preview is deferred.
- 3 Partial/retained: selected board cards already link back to their cards/sources; positive warrants and target stress purposeful review. Aggregate open counters remain truthful and no points are awarded. A complete counter-layout redesign is deferred.
- 4 Implemented: own connection inspector beside later drafting; main target/actor/action/option carry into comparison, with optional inference and evidence attachments visible. No prose generation.
- 5 Partial/retained: own-link decision reasons, target/board review notices, first idea, current writing and revision history. A new unified before/current dashboard is deferred; unchanged prose is never silently repaired.
- 6 Deferred: active and legacy practice claims/options/feedback are preserved to avoid an unreviewed item-bank rewrite. Own-writing decision prompts and optional independent reflection strengthen the ceiling; new plausible distractor sets need a separate item-validation round.
- 7 Implemented in part: target and connection travel, topic index and blank comparison frame establish intention. A separate forced-step selector and additional planner fields are deferred; final authored comparison remains.
- 8 Implemented in part: a focused hearing agenda extends the existing paper/collage atmosphere with calm hierarchy. Existing resume controls and pair cues remain; new timers, electoral effects, political scores or decorative institutional imagery are not added.

CROSS-CUTTING VERIFICATION

- 594 assertions: 268 new plus 326 prior. Complete-record fixtures include the new short target; every prior analytical gate remains.
- All 24 historical card bodies and 12 active/12 legacy answer keys/feedback preserved, with original source identities and metadata.
- Informed high-school-role retest sampled power map, target and comparison carry-through, own-connection inspector, focused agenda and both decisions required. Not a new blind round or complete replay.
- Retest wording findings corrected: pending self-checks now direct an action; failed analogy replaced with a plain comparison-does-not-fit phrase and an optional glossary definition.
- Engineering caught inconsistent comparison labels in the exported report; they now match neutral screen wording.
- Exact source/build hashes, full review reports, sampled retest scope and setup exclusions are retained. Physical-phone use, full assistive-technology audit, actual classroom times and learning gains remain unverified.
