Implement Slice 5 applied Skill learning #2

Merged
alexander merged 58 commits from mcp/migrate-gh-pr-44-slice-5-applied-skill-learning into main 2026-09-20 14:19:11 +02:00
Owner

Historical migration note: This PR was recreated in Forgejo from the former GitHub Slice-5 PR. GitHub references and CI statuses are historical only and are not current acceptance evidence.

Original source branch: feature/slice-5-applied-skill-learning

Current Forgejo acceptance state is recorded below.

Implements Slice 5 — Applied Skill learning + self-taught plateau from the merged Slice-5 integration contract.

Player-facing result

  • Swordsmanship is learned only from explicitly authored relevant practice: the deliberate sword drill and the west sentry's sword Method.
  • Learning is credited from lived execution time, diminishes toward the current ceiling, snaps to the literal self-taught limit, and banks no hidden overflow.
  • The fixed sentry problem becomes progressively less useful for broad Swordsmanship as retained capability rises and eventually grants zero broad Skill learning.
  • Swordsmanship transfers to explicitly relevant Methods and can improve execution even when exact-task familiarity is zero; exact Mastery, broad Mastery and Skill remain independent ledgers.
  • A Skill remains hidden until a resolved positive qualifying practice episode establishes recognition. Once recognized, the existing Mastery destination becomes Learning and presents Skill separately from Mastery / Familiarity.
  • If first Swordsmanship recognition occurs through fatal sentry practice, the existing loop summary surfaces that recognition before the ordinary loop log is cleared.
  • Action buttons keep the actual resulting time/cost primary; an accessible Why this time? disclosure explains exact familiarity, broad familiarity and known relevant Skill contribution.
  • Genuine positive Skill gains below two-decimal display precision render as <0.01, never as 0.00.
  • Plateau messaging states only that new instruction or insight is required; Slice 5 does not reveal an undiscovered teacher, location, curriculum or next ceiling.

Engineering / persistence

  • Adds explicit ActionMethod.skillRelationship metadata; Skill relevance is never inferred from category, equipment, labels, IDs or animation.
  • Authors baseline ceiling/rate/snap as per-Skill balance under skillLearning.skills.<skillId>, while shared performance caps and challenge profiles remain common configuration.
  • Persists sparse Skill competence, current ceiling and recognition state with safe old-save/malformed-state handling and no historical Mastery/Aptitude conversion.
  • Repository/static and runtime/editable validation enforce equivalent Skill-learning bounds and own-property reference ownership for Skill IDs, per-Skill tuning, challenge profiles, and Mastery groups.
  • Prototype-colliding unknown IDs such as toString, constructor, and __proto__ are rejected rather than resolving through Object.prototype.
  • Empty or whitespace-only challengeProfileId values are rejected in both static and editable/runtime validation even if a matching blank challenge-profile key is present.
  • Engine runtime also fails closed for owned malformed/null per-Skill tuning, challenge profiles and fixed-profile Mastery-group dependencies: no Skill learning/performance fallback and no uncaught dereference.
  • Safe offline practice credits only execution time actually consumed offline; the first live completion tick cannot add the synthetic 0.001s Skill-practice step.
  • Preserves Slice-4 exact/broad Mastery behavior and persistence semantics.

Acceptance coverage

Automated coverage includes lived partial practice, cancellation, fatal sentry resolution and loop-summary recognition, exact safe-offline completion, offline ineligibility for the sentry, diminishing learning, exact plateau/no overflow, raised-ceiling re-learning, malformed persistence, no historical inference, fixed challenge relevance, malformed owned runtime metadata, anti-farming, Skill/Mastery ledger independence, transfer at zero exact Mastery, the 1.75× Skill cap, the 8× combined progression cap, unrelated valid Methods receiving 1× Skill contribution, self-taught-limit precedence over challenge irrelevance, per-Skill balance ownership, static/runtime validation equivalence, prototype-colliding reference IDs, empty/whitespace-only challenge-profile references, and Learning UI/browser accessibility at 400px with 1.3× text.

The final acceptance-test repair uses the actual current contracts:

  • action timing mutation uses baseTimeCostSec;
  • actionProgressionBreakdown() assertions use { mastery, skillSpeed, totalSpeed } and mastery.* fields;
  • exactFamiliarityComplete() uses the exact-task rep ID plus the authored task Mastery curve;
  • unrelated-Method coverage clones a valid authored Method and removes skillRelationship rather than inventing stale fields;
  • previously accepted Skill/Mastery independence, zero-exact transfer/cap, and plateau-reason precedence tests are restored.

Documentation remains synchronized in CONTENT_AUTHORING.md, docs/gameplay/skill.md, docs/gameplay/gameplay-tables.md, docs/development-status.md, and plan.md.

Acceptance-fix reconciliation

Independent Agent-50 passes identified and the implementation now includes only the requested Slice-5 corrections:

  1. per-Skill learning tuning ownership;
  2. static/editable validation semantic equivalence;
  3. own-property reference validation for Skill/tuning/profile/Mastery-group IDs;
  4. exact safe-offline Skill-practice accounting;
  5. matching static/editable rejection of blank challengeProfileId;
  6. fail-closed runtime handling for owned malformed Skill/profile metadata;
  7. player-visible fatal first-recognition in the existing loop summary;
  8. truthful <0.01 display for genuine subprecision positive gains;
  9. repaired/restored acceptance tests against the current ActionMethod/Mastery APIs.

No Slice-6 mechanic, teacher route, lesson interaction, raised-ceiling source, curriculum, Bellhollow hint, second Skill, or future-ceiling disclosure was added.

Validation

Final accepted state

  • Exact accepted HEAD: 0acfde0eef200f1e2b2901c8a3c5e2c3fc8ee4db
  • Base main: 151d2e2a61e1aac52f0343e037a1a267c2864e66
  • Authoritative Forgejo Actions run: #60 — PASS
  • Independent Agent 50 — Code Reviewer / Game Acceptance verdict: ACCEPTED

Forgejo Actions run #60 passed on the exact accepted HEAD:

  • Core / static and content — PASS
    • content validation
    • strict TypeScript
    • ESLint
    • formatting
    • production web build
  • Core / tests — PASS
    • resolved-content tests: 25/25
    • full Vitest suite: 222/222
  • Browser / Playwright smoke — PASS
    • 6/6
  • Security / production npm audit — PASS
    • 0 vulnerabilities

The conditional Tauri/platform validation lane is not applicable to this PR.

Agent 50 rechecked the previously rejected Slice-5 areas and found no blocking or medium-severity defects remaining. Acceptance applies specifically to exact HEAD 0acfde0eef200f1e2b2901c8a3c5e2c3fc8ee4db. Slice-6 functionality remains out of scope and was not introduced.

Historical GitHub CI/migration blocker text has been retired from the current acceptance record. Forgejo Actions is authoritative.

Explicitly out of scope

Slice 6 teacher-raised ceilings, teacher/curriculum fiction, lesson interactions and next-ceiling disclosure are not implemented here.

Slice 5 is accepted by independent Agent 50 on exact HEAD 0acfde0eef200f1e2b2901c8a3c5e2c3fc8ee4db with Forgejo Actions run #60 green. Merge remains a separate manual owner action.

> **Historical migration note:** This PR was recreated in Forgejo from the former GitHub Slice-5 PR. GitHub references and CI statuses are historical only and are not current acceptance evidence. > > Original source branch: `feature/slice-5-applied-skill-learning` > > Current Forgejo acceptance state is recorded below. Implements **Slice 5 — Applied Skill learning + self-taught plateau** from the merged Slice-5 integration contract. ## Player-facing result - Swordsmanship is learned only from explicitly authored relevant practice: the deliberate sword drill and the west sentry's sword Method. - Learning is credited from lived execution time, diminishes toward the current ceiling, snaps to the literal self-taught limit, and banks no hidden overflow. - The fixed sentry problem becomes progressively less useful for broad Swordsmanship as retained capability rises and eventually grants zero broad Skill learning. - Swordsmanship transfers to explicitly relevant Methods and can improve execution even when exact-task familiarity is zero; exact Mastery, broad Mastery and Skill remain independent ledgers. - A Skill remains hidden until a resolved positive qualifying practice episode establishes recognition. Once recognized, the existing Mastery destination becomes **Learning** and presents Skill separately from Mastery / Familiarity. - If first Swordsmanship recognition occurs through fatal sentry practice, the existing loop summary surfaces that recognition before the ordinary loop log is cleared. - Action buttons keep the actual resulting time/cost primary; an accessible **Why this time?** disclosure explains exact familiarity, broad familiarity and known relevant Skill contribution. - Genuine positive Skill gains below two-decimal display precision render as `<0.01`, never as `0.00`. - Plateau messaging states only that new instruction or insight is required; Slice 5 does not reveal an undiscovered teacher, location, curriculum or next ceiling. ## Engineering / persistence - Adds explicit `ActionMethod.skillRelationship` metadata; Skill relevance is never inferred from category, equipment, labels, IDs or animation. - Authors baseline ceiling/rate/snap as **per-Skill balance** under `skillLearning.skills.<skillId>`, while shared performance caps and challenge profiles remain common configuration. - Persists sparse Skill competence, current ceiling and recognition state with safe old-save/malformed-state handling and no historical Mastery/Aptitude conversion. - Repository/static and runtime/editable validation enforce equivalent Skill-learning bounds and own-property reference ownership for Skill IDs, per-Skill tuning, challenge profiles, and Mastery groups. - Prototype-colliding unknown IDs such as `toString`, `constructor`, and `__proto__` are rejected rather than resolving through `Object.prototype`. - Empty or whitespace-only `challengeProfileId` values are rejected in both static and editable/runtime validation even if a matching blank challenge-profile key is present. - Engine runtime also fails closed for owned malformed/null per-Skill tuning, challenge profiles and fixed-profile Mastery-group dependencies: no Skill learning/performance fallback and no uncaught dereference. - Safe offline practice credits only execution time actually consumed offline; the first live completion tick cannot add the synthetic `0.001s` Skill-practice step. - Preserves Slice-4 exact/broad Mastery behavior and persistence semantics. ## Acceptance coverage Automated coverage includes lived partial practice, cancellation, fatal sentry resolution and loop-summary recognition, exact safe-offline completion, offline ineligibility for the sentry, diminishing learning, exact plateau/no overflow, raised-ceiling re-learning, malformed persistence, no historical inference, fixed challenge relevance, malformed owned runtime metadata, anti-farming, Skill/Mastery ledger independence, transfer at zero exact Mastery, the `1.75×` Skill cap, the `8×` combined progression cap, unrelated valid Methods receiving `1×` Skill contribution, self-taught-limit precedence over challenge irrelevance, per-Skill balance ownership, static/runtime validation equivalence, prototype-colliding reference IDs, empty/whitespace-only challenge-profile references, and Learning UI/browser accessibility at 400px with 1.3× text. The final acceptance-test repair uses the actual current contracts: - action timing mutation uses `baseTimeCostSec`; - `actionProgressionBreakdown()` assertions use `{ mastery, skillSpeed, totalSpeed }` and `mastery.*` fields; - `exactFamiliarityComplete()` uses the exact-task rep ID plus the authored task Mastery curve; - unrelated-Method coverage clones a valid authored Method and removes `skillRelationship` rather than inventing stale fields; - previously accepted Skill/Mastery independence, zero-exact transfer/cap, and plateau-reason precedence tests are restored. Documentation remains synchronized in `CONTENT_AUTHORING.md`, `docs/gameplay/skill.md`, `docs/gameplay/gameplay-tables.md`, `docs/development-status.md`, and `plan.md`. ## Acceptance-fix reconciliation Independent Agent-50 passes identified and the implementation now includes only the requested Slice-5 corrections: 1. per-Skill learning tuning ownership; 2. static/editable validation semantic equivalence; 3. own-property reference validation for Skill/tuning/profile/Mastery-group IDs; 4. exact safe-offline Skill-practice accounting; 5. matching static/editable rejection of blank `challengeProfileId`; 6. fail-closed runtime handling for owned malformed Skill/profile metadata; 7. player-visible fatal first-recognition in the existing loop summary; 8. truthful `<0.01` display for genuine subprecision positive gains; 9. repaired/restored acceptance tests against the current ActionMethod/Mastery APIs. No Slice-6 mechanic, teacher route, lesson interaction, raised-ceiling source, curriculum, Bellhollow hint, second Skill, or future-ceiling disclosure was added. ## Validation ### Final accepted state - **Exact accepted HEAD:** `0acfde0eef200f1e2b2901c8a3c5e2c3fc8ee4db` - **Base `main`:** `151d2e2a61e1aac52f0343e037a1a267c2864e66` - **Authoritative Forgejo Actions run:** **#60 — PASS** - **Independent Agent 50 — Code Reviewer / Game Acceptance verdict:** **ACCEPTED** Forgejo Actions run #60 passed on the exact accepted HEAD: - **Core / static and content — PASS** - content validation - strict TypeScript - ESLint - formatting - production web build - **Core / tests — PASS** - resolved-content tests: **25/25** - full Vitest suite: **222/222** - **Browser / Playwright smoke — PASS** - **6/6** - **Security / production npm audit — PASS** - **0 vulnerabilities** The conditional Tauri/platform validation lane is **not applicable** to this PR. Agent 50 rechecked the previously rejected Slice-5 areas and found no blocking or medium-severity defects remaining. Acceptance applies specifically to exact HEAD `0acfde0eef200f1e2b2901c8a3c5e2c3fc8ee4db`. Slice-6 functionality remains out of scope and was not introduced. Historical GitHub CI/migration blocker text has been retired from the current acceptance record. Forgejo Actions is authoritative. ## Explicitly out of scope Slice 6 teacher-raised ceilings, teacher/curriculum fiction, lesson interactions and next-ceiling disclosure are not implemented here. **Slice 5 is accepted by independent Agent 50 on exact HEAD `0acfde0eef200f1e2b2901c8a3c5e2c3fc8ee4db` with Forgejo Actions run #60 green. Merge remains a separate manual owner action.**
alexander force-pushed mcp/migrate-gh-pr-44-slice-5-applied-skill-learning from 5badc50198 to 5ec42fa80a
Some checks failed
Loopbound PR verification / Core / static and content (pull_request) Failing after 15s
Loopbound PR verification / Core / tests (pull_request) Failing after 11s
Loopbound PR verification / Browser / Playwright smoke (pull_request) Failing after 33s
Loopbound PR verification / Security / production npm audit (pull_request) Successful in 7s
2026-09-20 13:08:02 +02:00
Compare
test(slice-5): fix relevance acceptance assertions
Some checks failed
Loopbound PR verification / Core / static and content (pull_request) Has been cancelled
Loopbound PR verification / Core / tests (pull_request) Has been cancelled
Loopbound PR verification / Browser / Playwright smoke (pull_request) Has been cancelled
Loopbound PR verification / Security / production npm audit (pull_request) Has been cancelled
dea9fc8c67
test(slice-5): fix loop transition expectation
Some checks failed
Loopbound PR verification / Core / static and content (pull_request) Has been cancelled
Loopbound PR verification / Core / tests (pull_request) Has been cancelled
Loopbound PR verification / Browser / Playwright smoke (pull_request) Has been cancelled
Loopbound PR verification / Security / production npm audit (pull_request) Has been cancelled
f181ba2b62
test(slice-5): stabilize content studio acceptance setup
All checks were successful
Loopbound PR verification / Core / static and content (pull_request) Successful in 15s
Loopbound PR verification / Core / tests (pull_request) Successful in 10s
Loopbound PR verification / Browser / Playwright smoke (pull_request) Successful in 21s
Loopbound PR verification / Security / production npm audit (pull_request) Successful in 7s
0acfde0eef
Author
Owner

Agent 50 — Final Slice-5 Acceptance

This comment records the completed independent Agent 50 — Code Reviewer / Game Acceptance review. It is not Agent 40 self-approving its own implementation.

  • Verdict: ACCEPTED
  • Exact reviewed HEAD: 0acfde0eef200f1e2b2901c8a3c5e2c3fc8ee4db
  • Base main: 151d2e2a61e1aac52f0343e037a1a267c2864e66
  • Forgejo Actions run #60: PASS
  • No blocking or medium findings remain.
  • Acceptance applies only to the exact reviewed HEAD.
  • Slice-6 functionality remains out of scope and was not introduced.
### Agent 50 — Final Slice-5 Acceptance This comment records the completed **independent Agent 50 — Code Reviewer / Game Acceptance** review. It is not Agent 40 self-approving its own implementation. - **Verdict:** ACCEPTED - **Exact reviewed HEAD:** `0acfde0eef200f1e2b2901c8a3c5e2c3fc8ee4db` - **Base `main`:** `151d2e2a61e1aac52f0343e037a1a267c2864e66` - **Forgejo Actions run #60:** PASS - No blocking or medium findings remain. - Acceptance applies only to the exact reviewed HEAD. - Slice-6 functionality remains out of scope and was not introduced.
alexander deleted branch mcp/migrate-gh-pr-44-slice-5-applied-skill-learning 2026-09-20 14:19:11 +02:00
Sign in to join this conversation.
No reviewers
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
alexander/Loopbound!2
No description provided.