Pressure test.
Teaching edition.
Created by Yoav Duman · verified 29 October 2026 · offline prepublication checks.
What passed: eight current executable suites; independent content checks; live browser entry, country/repair/teach-back and exit/revision/export checks; 320/390-pixel layout checks; rendered review of all 21 printable pages. Full nine-country journeys were exercised by the browserless UI harness. Live browser checks used explicit synthetic QA fixtures for later stages; they are not a claim of an uninterrupted human playthrough or learning gain.
Narrative and replay refinements · 29 October 2026
All eight executable suites passed for this follow-up. The engine, PowerLab, learning instrument, country/tradeoff data, art/teaching modules, three PDFs and locked blind-review report are byte-identical to the preceding saved pack. The changes are presentation, optional cosmetic achievements and validated wrapper metadata. This follow-up adds no required prompt, lowers no class minimum and removes no country decision.
| Change | What was verified |
|---|---|
| Party voices and fictional news | Four one-line voices explain the existing party priorities. Static Port Meridian Wire strips describe modeled support changes, without flashing or extra clicks. The fixed core confidence motion is not invented as a named party walkout. The no-allies fallback is labeled a delivery dispute, not a coalition rupture. |
| Three optional strategy stamps | All conditions are reachable under unchanged arithmetic: presidential housing with an initially empty pact, Greens/Regions concessions and broader Civic deal yields 61 bill votes / 42 confidence seats; parliamentary Civic/Workers then broader Greens yields 61 / 59; semi-presidential Civic/Greens with jobs and reserve yields 54 / 59. The last is an opposing support agreement, not formal cohabitation, because this lab does not select a PM. |
| Evidence and rigor | A matching political outcome alone earns nothing until the constitutional prediction is correct or repaired. First mistakes remain recorded. Stamps add no practice points and do not affect required completion. Three canonical run/prediction proofs survive twelve later runs and restore from JSON after leaving the ten-run window. Forged results, mismatched predictions, future runs, wrong challenge conditions and conflicts with retained evidence are rejected. |
| Rapid entry dispatch | The same six first scenarios remain before teaching. Brief bullets and a short decisive phrase are invited. Five minutes is an aim, with no timer or speed score; timing is untested. Per-item presentation metadata identifies new first responses without relabeling older firsts when revised. Confidence, reported help, sealed baseline, deferred criteria and later exit/class requirements are retained. |
| Purposeful visuals and replay | A before/after table and proportion bars separate bill support, confidence support and reserve tokens. Explicit scales and text equivalents accompany the bars. A next-experiment prompt adds a replay aim without an extra essay. Existing flags, case diagrams and institutional atlas are retained; no decorative map or new country inference was added. |
| Live browser and narrow layouts | A concise bullet entry locked and advanced to item two. A live parliamentary run used the one-token broader deal: an incorrect legal prediction earned no stamp, its correction earned one, and reload preserved it. Real entry and lab JSON downloads passed the current validator, including the first mistake and repair. The chart, news and stamps were inspected at 320/390 CSS pixels; client/scroll widths matched at 305/305 and 375/375. A ledger heading contrast issue found in visual inspection was corrected. These are internal browser checks, not real-device or assistive-technology certification. |
This is an engineering follow-up, not a new blind panel or human learning study. Engagement gains and the five-minute expectation remain hypotheses for a real classroom pilot. All nine country pairs, eighteen decisions and original assessment criteria remain in the pack.
Focused teaching refinements · 29 October 2026
All seven executable suites passed after the four requested aids were added. The original engine, assessment module, country/tradeoff data, art/teaching module and all three PDFs are byte-identical to the preceding release. In PowerLab, only preview information changed; canonical state, vote arithmetic and saved evidence functions are unchanged.
| Aid | Guardrail and verification |
|---|---|
| Planning satisfaction and vote-bloc breakdown | Per-party scores, the deterministic next dispute and a tap/keyboard disclosure show the existing calculation, including the seat cap. The lowest-score/larger-party tie breaker, resource constraints and actual votes remain unchanged. Live browser verified Workers at zero, then Civic at two after a tied concession change. |
| Advisory writing reminders | Entry/exit/return answers under twelve words receive a neutral reminder. Reflection keyword absence can prompt a suggestion, with everyday synonyms accepted. No reminder assigns a score or adds a submission gate. Existing class minimums remain; brief entry uncertainty was accepted in the live browser without revealing criteria. |
| Earlier cohabitation preview | The PM selector previews opposing camps/cohabitation, same-party leadership, or mixed-party leadership with shared support. Including Workers alone does not imply a Workers PM or guarantee executive agreement. Live keyboard changes preserved selector focus; no government was automatically installed. |
| Three-act pacing and Save & Pause | Switchboard; Country Dossiers; optional Midnight Lab and final research/writing. Pause preserves screen, drafts, first responses and progress without advancing. Local-save failure instead requests a backup. Live download/reload resumed the exact route; automated quota and entry/exit timing checks passed. |
Planning cards were inspected at 320/390 CSS pixels. Client widths 305/375 matched scroll widths 305/375, with no page overflow. Ledger header contrast was corrected after visual inspection. No new student trial, learning gain, engagement effect or reduction in workload is claimed. The Midnight Session remains optional; all nine country pairs and assessment gates remain intact.
Previously verified scope retained
| Area | Verification and result |
|---|---|
| Institutional rules | Both blind faculty panels audited all nine countries and eighteen keyed moves; a fresh faculty acceptance reviewer found no content blocker. Portugal’s article attribution was corrected to 195(1)(f); all prior correct-choice texts remain unchanged for save compatibility. All eighteen new synthetic assessment cards and 72 human criteria were checked for internal consistency. |
| Core model | Three origin/survival rules, ordinary bill defeat, formal confidence/censure, legal continuation, cohabitation, original formation, six-slot constraints and immutable first/revised evidence passed. |
| Choice-driven lab | 1,473 distinct deterministic sessions sampled across 5,187 legal configurations/responses. All four bill/confidence outcome combinations, token/slot limits, same-support cross-system comparisons, replay retention and canonical import/prediction checks passed. |
| Learning instrument | 11 dedicated tests passed. Six entry, six exit and six optional return prompts are retained. First explanations/confidence/help are locked; revisions wait for all initials and retain a why-statement. Entry seals when instruction starts; exit follows final comparison, tradeoffs and submitted research. Legacy timing is labeled unverified. No semantic grade or gain calculation is produced. |
| Fresh full expedition | All18 country decisions, nine saved office/rule/consequence traces, three cross-case comparisons, final research, optional/partial/complete entry, six exit and six return firsts, revisions, restore/reload and 22 malformed teaching/lab records passed. |
| Backup boundary | One 33,554,432-byte UTF-8 budget is used at import/export/wrapper boundaries. The maximum supported-writing fixture is 22,838,979 bytes and passes. Full explanation bodies are no longer duplicated in the journal. A localStorage quota failure prompts a downloadable backup. Old records missing original formation evidence are rejected honestly and can be downloaded for recovery. |
| Live browser | Entry answers survived reload; all six exit first explanations were collected before criteria; a short class answer was blocked; reported reference use and a why-revision survived real JSON/TXT downloads. A wrong Portugal answer, its repair, the exact denominator graphic and the saved trace were exercised. Exported complete JSON passed the current validator. One automated pointer activation missed and a later download wait timed out; keyboard activation completed both exports. This is recorded rather than treating every attempted activation as a successful check. |
| Phone layouts | Same-origin frames at 320 and 390 CSS pixels showed the Portugal threshold graphic and three trace controls. Document client widths 305/375px matched scroll widths 305/375px, with no horizontal page overflow. Text equivalents remain readable. This is layout QA, not a real-device or screen-reader certification. |
| Print materials | Student Guide 3 pages, Instructor Guide 9 pages, Worksheet 9 pages. All 21 final pages rendered and visually inspected; six later changed pages re-rendered and rechecked. Embedded fonts and all 9+9 answer keys verified. |
| Website | No deployment, audience change or catalog update was performed. The publication flag remains false. The separate website integration test could not complete because the website runtime dependency was absent; it is not counted as a passed offline-game check. |
Findings fixed during the final acceptance pass
A fresh blind student-perspective reviewer found entry-resume after teaching and premature exit access. Shared teaching/exit gates now close both paths, including help, reload and restore. Imports preserve old responses while labeling unknown timing, and obsolete sealed-entry screens redirect to the portfolio. Focused regressions passed and the independent reviewer verified these corrections separately from its unchanged blind report.
The same pass aligned setup and guides around the complete nine-country default, three resumable sessions, the failure-condition/counterfactual tradeoff prompt and a visible three-revision allowance. The institutional atlas is a voluntary disclosure until the dossier traces are complete, so the initial route does not duplicate nine long evidence cards.
Release package check
The teaching pack passed ZIP integrity checks. Its editable source was extracted into a clean directory, rebuilt the identical standalone HTML byte for byte, and passed the embedded offline UI test there. The download includes the game, all three PDF guides, sources, unchanged blind reviews and their correction verification.
What has not been established
These are AI reviews, software checks and document/layout inspections. No recruited students or professors participated. Actual comprehension gains, retention, enjoyment, workload, completion time, explanation-score reliability, physical print quality, 200% browser zoom, assistive technology and real-device behavior remain unmeasured. The game supplies a portfolio and a pilot instrument; it does not award a validated mastery result.
Use the instructor guide’s pilot protocol for first-time students, preserve first answers and help history, and score explanations with phase-blind human raters where feasible. Treat entry/exit differences as changes over an instructional interval, not automatically a causal game effect.
Narrative/replay executable evidence
engine · PASS
Power Circuit engine: formation, policy/confidence separation, three survival rules, replacement, cohabitation, immutable records, classroom writing and import validation passed. Power Circuit repeated repair/reflection histories, original formation evidence and replacement consistency passed.
lab · PASS
Power Circuit Midnight Session: 1473 distinct deterministic sessions sampled across 5187 legal configurations/responses; resource constraints, all four outcome combinations, fixed-support system comparisons, replay evidence, bounded history and canonical import validation passed.
learning · PASS
✔ all eighteen items have stable IDs, open questions and four human criteria (2.77469ms) ✔ fresh records and summaries make a missing baseline explicit without scoring meaning (3.617928ms) ✔ record returns a new frozen state and locks the first response without word-quality checks (2.156662ms) ✔ all six initial responses are required before any revision in that form (13.247621ms) ✔ three revisions are retained and a fourth is rejected without losing the first (1.167955ms) ✔ skipped or partial entry records cannot become a retroactive baseline (2.8474ms) ✔ starting a later form automatically seals an offered entry record (1.878473ms) ✔ form/index checks and response text, confidence, help and revision limits are enforced (3.574883ms) ✔ timestamps are canonical and cannot put a revision before the full initial form (4.194956ms) ✔ malformed imports, mismatched identifiers and inconsistent histories are rejected (2.72151ms) ✔ browser global export works without Node or DOM and uses the same stable schema (4.868878ms) ℹ tests 11 ℹ suites 0 ℹ pass 11 ℹ fail 0 ℹ cancelled 0 ℹ skipped 0 ℹ todo 0 ℹ duration_ms 183.195112
round2-boundaries · PASS
Power Circuit round-2 boundaries passed: shared six-slot rules, honest original formation, compact journal, genuine engine round-trip, and 22,838,979-byte maximum writing fixture within 33,554,432 UTF-8 bytes.
ui · PASS
Power Circuit browserless UI: class gating, coalition negotiation, all three circuits, corrected predictions, 18 country moves, historical replay, lab, semantic save validation, 52 research revisions, embedded offline runtime, tradeoff/research gates, report/export, autosave/reload, restore and malformed backup rejection passed.
round2-ui · PASS
Power Circuit round-2 UI passed: fresh full nine-country journey, eighteen country decisions, instruction-sealed entry and post-submission exit timing, optional/partial/complete entry, six exit and return explanations before criteria, reported help, class writing gates, draft/reload/restore, immutable first reasoning, reachable revisions, retained lab replay evidence, 22 malformed teaching/lab records, and UTF-8 file limits.
assists · PASS
Focused assists passed: deterministic risk/tie previews and capped vote breakdown; three executive arrangements before formation; advisory-only writing with original class gates; exact pause/resume, quota fallback, backup compatibility and unchanged first evidence.
engagement · PASS
Engagement checks passed: all three reachable canonical challenges; prediction/repair gate; unchanged scores, 9 countries/18 decisions and baseline gates; archived stamp persistence; semantic import rejection; legacy hydration; brief dispatches with instruction metadata and immutable first responses.
Focused-refinement executable evidence
engine · PASS
Power Circuit engine: formation, policy/confidence separation, three survival rules, replacement, cohabitation, immutable records, classroom writing and import validation passed. Power Circuit repeated repair/reflection histories, original formation evidence and replacement consistency passed.
lab · PASS
Power Circuit Midnight Session: 1473 distinct deterministic sessions sampled across 5187 legal configurations/responses; resource constraints, all four outcome combinations, fixed-support system comparisons, replay evidence, bounded history and canonical import validation passed.
learning · PASS
✔ all eighteen items have stable IDs, open questions and four human criteria (2.614141ms) ✔ fresh records and summaries make a missing baseline explicit without scoring meaning (1.374658ms) ✔ record returns a new frozen state and locks the first response without word-quality checks (2.49371ms) ✔ all six initial responses are required before any revision in that form (14.238678ms) ✔ three revisions are retained and a fourth is rejected without losing the first (1.47652ms) ✔ skipped or partial entry records cannot become a retroactive baseline (3.352622ms) ✔ starting a later form automatically seals an offered entry record (3.161083ms) ✔ form/index checks and response text, confidence, help and revision limits are enforced (2.517285ms) ✔ timestamps are canonical and cannot put a revision before the full initial form (2.076732ms) ✔ malformed imports, mismatched identifiers and inconsistent histories are rejected (2.684196ms) ✔ browser global export works without Node or DOM and uses the same stable schema (5.006858ms) ℹ tests 11 ℹ suites 0 ℹ pass 11 ℹ fail 0 ℹ cancelled 0 ℹ skipped 0 ℹ todo 0 ℹ duration_ms 214.177925
round2-boundaries · PASS
Power Circuit round-2 boundaries passed: shared six-slot rules, honest original formation, compact journal, genuine engine round-trip, and 22,838,979-byte maximum writing fixture within 33,554,432 UTF-8 bytes.
ui · PASS
Power Circuit browserless UI: class gating, coalition negotiation, all three circuits, corrected predictions, 18 country moves, historical replay, lab, semantic save validation, 52 research revisions, embedded offline runtime, tradeoff/research gates, report/export, autosave/reload, restore and malformed backup rejection passed.
round2-ui · PASS
Power Circuit round-2 UI passed: fresh full nine-country journey, eighteen country decisions, instruction-sealed entry and post-submission exit timing, optional/partial/complete entry, six exit and return explanations before criteria, reported help, class writing gates, draft/reload/restore, immutable first reasoning, reachable revisions, retained lab replay evidence, 22 malformed teaching/lab records, and UTF-8 file limits.
assists · PASS
Focused assists passed: deterministic risk/tie previews and capped vote breakdown; three executive arrangements before formation; advisory-only writing with original class gates; exact pause/resume, quota fallback, backup compatibility and unchanged first evidence.
Original round-two executable evidence
engine · PASS
Power Circuit engine: formation, policy/confidence separation, three survival rules, replacement, cohabitation, immutable records, classroom writing and import validation passed. Power Circuit repeated repair/reflection histories, original formation evidence and replacement consistency passed.
lab · PASS
Power Circuit Midnight Session: 1473 distinct deterministic sessions sampled across 5187 legal configurations/responses; resource constraints, all four outcome combinations, fixed-support system comparisons, replay evidence, bounded history and canonical import validation passed.
learning · PASS
✔ all eighteen items have stable IDs, open questions and four human criteria (5.918013ms) ✔ fresh records and summaries make a missing baseline explicit without scoring meaning (3.264381ms) ✔ record returns a new frozen state and locks the first response without word-quality checks (12.635827ms) ✔ all six initial responses are required before any revision in that form (33.248183ms) ✔ three revisions are retained and a fourth is rejected without losing the first (7.155211ms) ✔ skipped or partial entry records cannot become a retroactive baseline (9.221782ms) ✔ starting a later form automatically seals an offered entry record (4.566755ms) ✔ form/index checks and response text, confidence, help and revision limits are enforced (5.113656ms) ✔ timestamps are canonical and cannot put a revision before the full initial form (3.256368ms) ✔ malformed imports, mismatched identifiers and inconsistent histories are rejected (7.819778ms) ✔ browser global export works without Node or DOM and uses the same stable schema (6.959757ms) ℹ tests 11 ℹ suites 0 ℹ pass 11 ℹ fail 0 ℹ cancelled 0 ℹ skipped 0 ℹ todo 0 ℹ duration_ms 327.879749
boundaries · PASS
Power Circuit round-2 boundaries passed: shared six-slot rules, honest original formation, compact journal, genuine engine round-trip, and 22,838,979-byte maximum writing fixture within 33,554,432 UTF-8 bytes.
legacy_ui · PASS
Power Circuit browserless UI: class gating, coalition negotiation, all three circuits, corrected predictions, 18 country moves, historical replay, lab, semantic save validation, 52 research revisions, embedded offline runtime, tradeoff/research gates, report/export, autosave/reload, restore and malformed backup rejection passed.
round2_ui · PASS
Power Circuit round-2 UI passed: fresh full nine-country journey, eighteen country decisions, instruction-sealed entry and post-submission exit timing, optional/partial/complete entry, six exit and return explanations before criteria, reported help, class writing gates, draft/reload/restore, immutable first reasoning, reachable revisions, retained lab replay evidence, 22 malformed teaching/lab records, and UTF-8 file limits.
Read the six locked blind reports, fresh acceptance reviews and redesign rationale.