@afokapu/atdd-bun 0.10.27 → 0.10.28

package/conventions/delivery/delivery.config-schema.convention.yaml

@@ -13,7 +13,7 @@ content:
   summary: >-
     The skill and the validators read the same policy. A misspelled key would be ignored by one and not the other, so a malformed block is a finding, and the rest of the run judges evidence against the package defaults.
   normative_text: |
-    The policy decides which models may write and review each stage, in preference order, how independent a reviewer must be, and when a model counts as unavailable. It is read by the delivery skill when it dispatches work and by this profile when it judges the record, so both must see the same, valid document.
+    The policy decides which models may write and review each stage (an allowed set, in no order; "*" allows any), how independent a reviewer must be, and when a model counts as unavailable. It is read by the delivery skill when it dispatches work and by this profile when it judges the record, so both must see the same, valid document.
   fix_hint: |
     Correct the key the finding names. The defaults are:
 

package/conventions/delivery/delivery.model-allowed.convention.yaml

@@ -4,31 +4,19 @@ kind: rule
 status: active
 name: Authors and reviewers are allowed models, and every fallback says why
 statement: >-
-  Each work entry's writer appears in that stage's writer list and each review's reviewer in its reviewer list; a model after the first is used only with a recorded fallback from each model before it, stating the kind of unavailability (outage, rate_limit, no_report, timeout), the failures observed (at least delivery.fallback.after_failures) within a window no longer than delivery.fallback.within_minutes, and the reason; or, where the operator ordered the substitution, kind operator with who decided, when, the ruling it is recorded in, and the reason (REQUIRED).
+  Each work entry's writer is a model its stage allows for writers, and each review's reviewer a model it allows for reviewers. A stage's list is the set of allowed models, in no order; "*" allows any model. Which allowed model does the work is the operator's choice at any moment, so no fallback has to explain it; the record names the model that actually ran (REQUIRED).
 terms:
 - term_id: fallback
   text: >-
     a recorded switch from a preferred model to a later one in the stage's list, with role (writer or reviewer) and a reason: an outage, a rate limit, or no auditable report, counted by failures; or the operator's order, referenced by its ruling. A REQUEST CHANGES verdict is never a reason.
 content:
   summary: >-
-    The lists are the operator's choice of who may do the work. A silent switch to another model hides an outage and can hide a reviewer chosen for being lenient.
+    The lists are the operator's choice of which models may do the work. Switching between allowed models needs no policy edit and no paperwork, so a rate limit or a preference change is handled on the spot; a model outside the list is still reported.
   normative_text: |
-    The first model in a list is used normally. When it is unavailable, the driver falls back to the next one and records it. A fallback is never triggered by a verdict: shopping for a reviewer until one approves is exactly what this rule exists to expose.
-    An operator can order a substitution faster than a policy change lands, as a standing ruling while a model is rate-limited. The record then states that ruling rather than failures nobody counted. The substitute must still be a model the stage lists in the policy the record is judged by, so the policy file keeps the final word; and only the operator rules: a driver never writes an operator fallback on its own authority.
+    The record must name the model that ran. Reviewer independence (a reviewer is never the model that wrote the work it reviews, under different-model) is judged by delivery.reviewer-independent, and a disputed finding upheld on a fresh review blocks the tranche, so switching reviewers cannot buy an approval.
+    A fallback entry (kind outage, rate_limit, no_report, timeout or operator) may still record why a model was used; it is optional and, when present, must be well formed.
   fix_hint: |
-    Record the fallback on the review:
-
-      fallback:
-        - { role: reviewer, from: codex, kind: rate_limit, failures: 3, window: { from: "2026-09-25T09:00:00Z", to: "2026-09-25T09:08:00Z" }, reason: "429 on 3 attempts in 10 minutes" }
-
-    or, for a substitution the operator ordered:
-
-        - { role: writer, from: claude-opus, kind: operator, decided_by: "the owner", decided_at: "2026-09-27T23:00:00Z", ruling: "atdd-frg-workstation #abc123", reason: "Standing substitution while Claude is rate-limited: kimi for Opus-tier work" }
-
-    or use a model the stage lists.
-  exceptions:
-  - >-
-    When a stage lists one model, no fallback is possible: an unavailable model blocks the tranche.
+    Use a model the stage lists, or have the operator add it (or "*") to the stage's list in atdd-bun.yaml; the integrity check reports an added model once for the operator to approve.
 metadata:
   aliases:
   - DELIVERY-MODEL-ALLOWED-001

package/conventions/delivery/delivery.operating-model.convention.yaml

@@ -11,13 +11,13 @@ terms:
     Owns throughput, never the work. It splits the program into tranches with explicit dependencies and writes why, the scope and the split in docs/delivery/index.adoc; activates a tranche once its own dependencies have merged (one still waiting may run plan and its review, nothing after); starts one driver per tranche in its own worktree; keeps provider health for the program, telling every driver to skip a model that is down; intervenes when a tranche is idle without being blocked, or has deliverables and no commit after two hours; and hands BLOCKED disputed-finding to a human with both sides. It never writes, reviews or merges.
 - term_id: driver
   text: >-
-    Owns one tranche end to end. For each stage of delivery.stages, in order, it starts the stage's writer (the first model in its writer list, or the next after a recorded fallback) in the tranche worktree, commits red on its own before green starts, passes the stage's gate, and appends a work entry; for each stage with a reviewer it starts a review. It never reviews.
+    Owns one tranche end to end. For each stage of delivery.stages, in order, it starts the stage's writer (any model in its writer list, as the operator chooses) in the tranche worktree, commits red on its own before green starts, passes the stage's gate, and appends a work entry; for each stage with a reviewer it starts a review. It never reviews.
 - term_id: review_run
   text: >-
     Every review is a fresh process in its own detached worktree at the exact SHA (git worktree add --detach), launched with a prompt that gives it the delivery.review convention, the stage, the base SHA, the SHA and the RED commit. Afterwards git status --porcelain must be empty and HEAD still the SHA, or the review is a REVIEWER_FAILURE and does not count. The raw output is kept in the tranche folder and named in the review's report.
 - term_id: fallback
   text: >-
-    After fallback.after_failures failures within fallback.within_minutes (outage, rate limit, no auditable report, timeout), the driver uses the next model in the list and records the fallback with its role, kind, failures and window. A request-changes verdict is never a failure. With the list exhausted, when_exhausted block emits BLOCKED provider-unavailable; wait keeps retrying the last model.
+    Any model a stage lists may be used, so a switch needs no fallback: when a model is down or rate-limited, the driver moves to another listed model and records the one that ran. A fallback entry may still say why (outage, rate limit, no auditable report, timeout, or the operator's order); it is optional. A request-changes verdict is never a reason to switch reviewers: a disputed finding upheld on a fresh review blocks the tranche. When no listed model is available, when_exhausted block emits BLOCKED provider-unavailable.
 - term_id: repair
   text: >-
     For each finding of a request-changes review the writer either fixes it or writes one rebuttal with evidence; a fresh review of the same stage follows. If that reviewer upholds a disputed finding, the driver emits BLOCKED disputed-finding and never disputes it again. Any commit, regenerated file, conflict fix or rebase after an approval cancels it and the change goes back to that stage's review.

package/integrity.json

@@ -1,5 +1,5 @@
 {
-  "version": "0.10.27",
+  "version": "0.10.28",
   "files": {
     "HOOK_AUDIT.md": "5329d840db37671b1918f688ead26865473b87db73dbc75f7c8b2a8bbe8d6d43",
     "PLANNER_PORT.md": "fb5935bac8b7ac18994de21e43ace3a5ef8cd55f85b0e3349fca261280054f11",
@@ -94,12 +94,12 @@
     "conventions/coder.htmx/coder.htmx.verb-mutation-signals-progress.convention.yaml": "ae85c5ea01a09a3c41b939988088a8e5da91848d2710a1f0a1e0c86308638ac9",
     "conventions/delivery/delivery.approved-sha-resolves.convention.yaml": "68f2ecc10870e780abb05bb31d211b1923d47b56672df4b3950f437aa9cf8ca8",
     "conventions/delivery/delivery.board.convention.yaml": "4c544f51de80b09961a8df41028867023ab802aa4b31dcb0a1ce29e890552e6d",
-    "conventions/delivery/delivery.config-schema.convention.yaml": "4c7b449074a8bd9f89b70c702ab3e0df229105ddb4458772f894f0bf37813e9f",
+    "conventions/delivery/delivery.config-schema.convention.yaml": "b59a69b7789ad0c3d0b1ac9ea2b955fbbb5dfa07a9ee169a037b406a198813f0",
     "conventions/delivery/delivery.evidence-schema.convention.yaml": "9c6a22495dfb3e2fcbca700392866c3b5660084b1f5594230eb320b499d5e5cb",
     "conventions/delivery/delivery.findings-resolved.convention.yaml": "f6574e93a9a2d4bdd6d6f5645720b1a647fc0048d95f7675fe783d2ddb62c05b",
     "conventions/delivery/delivery.merge-gate.convention.yaml": "48745ec23625b68be9b8fe2c6dcdcba88b2b38924f6593abf6987391ff0ee243",
-    "conventions/delivery/delivery.model-allowed.convention.yaml": "7bf69b1803ec388a77b2fbc8aa727ff4ac6fd0cf4e5d5736519117bf86ac8688",
-    "conventions/delivery/delivery.operating-model.convention.yaml": "7982157d957938ade878a114d3f5c2bdaa8f02e52ac21614634d9af4be9692f1",
+    "conventions/delivery/delivery.model-allowed.convention.yaml": "e9690e95f3604483f182b3032ae60eac20109eb458509f14651945ccf07692f2",
+    "conventions/delivery/delivery.operating-model.convention.yaml": "1a86a1fe84695968a2fb719a899629af9a8baadcd90bdb8776f1f0208cebbbb4",
     "conventions/delivery/delivery.review.convention.yaml": "b033579c2d90fcb098d325dcc07c7b084e048cca43fbe8e28159ec08a412018a",
     "conventions/delivery/delivery.reviewer-independent.convention.yaml": "8c436d66ef5b1f6d6bc5dd55a235caca2490020e88665eecadb2ca41cd4149c5",
     "conventions/delivery/delivery.stages-complete.convention.yaml": "0e5dd8057ad51eff7906e3424a9686a81be1814447afb9a30acab43d39e49b4c",
@@ -1179,7 +1179,7 @@
     "planner-schemas/binding-lock.schema.json": "f2455606702f45fafa194549126d4a66e6c286ba1c4ba9e6bdcf932286a6d34d",
     "planner-schemas/component.schema.json": "d20d3502c155627a120cb10d5e9bdec4ba2c73af9c3e9a9d10e3c351f167c72a",
     "planner-schemas/contract-registry.schema.json": "b9dbaf6c87d4f4d676e3b4251344b9eada523b1c5e739c0423cabe267012c12e",
-    "planner-schemas/delivery-config.schema.json": "16a9b66a7d36edf8a49c68d9786f2516cd2ff5392be6dd60b766a09ffc95a9c6",
+    "planner-schemas/delivery-config.schema.json": "7cd9c15ea4345ad83195b9a1c7d09a0b66275e9a3fdf8b5d6905d96bbec579d3",
     "planner-schemas/delivery-evidence.schema.json": "72ba1450abe5482c43f2069e9ee7d121398008a0ddef36c6b269d38a16872e0e",
     "planner-schemas/feature.schema.json": "eb4aa40d89067ae5c86a079bb3d8e4f45c8727f9e15db2b86c08dacf8e4ae56a",
     "planner-schemas/journey.schema.json": "1ff705bcf89a2823cf60c9e1597e254f97844e7a74dc58299d53192ad2c9f8c3",
@@ -1199,7 +1199,7 @@
     "src/board.ts": "bc70e479874f2f05a44293226e53eb9e763b661bdcebade4f8948b0dd2ee2ead",
     "src/ci.ts": "67e54d2cbfc44a9e42837d5e750af526d6255c539d3c452a4d5fca3226c0d9ac",
     "src/cli.ts": "e5ab4d2435fdcbebf462b4c2b67b495bd975d5deb668b088d8cba04a9d8b78a5",
-    "src/delivery.ts": "f6bb6a7c42693d795b964c77b6186e1ca3a1556fd43932911a0ad42e72e7e441",
+    "src/delivery.ts": "2eaaf4fd87aea26dce70b4c88a4e0a2f25b9de165bc2e38330fd7eb449587901",
     "src/docs-capability.ts": "8a7a4360976773fe462b1860eaf26182b016bcb1fbcd6a6d5423bd81379dbe67",
     "src/enforce.ts": "41a6f8c058a2a034996c817f224191403cc66ccc4ed4e04bbefc430014c783d8",
     "src/hooks.ts": "2feeec2674c1462c94256de02ce59ccc6e3ff671ec4dca1bf9903c32888b600b",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@afokapu/atdd-bun",
-  "version": "0.10.27",
+  "version": "0.10.28",
   "repository": {
     "type": "git",
     "url": "git+https://github.com/afokapu/atdd-bun.git"

package/planner-schemas/delivery-config.schema.json

@@ -76,8 +76,8 @@
       "type": "array",
       "minItems": 1,
       "uniqueItems": true,
-      "items": { "type": "string", "pattern": "^[a-z][a-z0-9-]*$" },
-      "description": "Models in preference order: the first is used normally, each later one only as a recorded fallback."
+      "items": { "type": "string", "pattern": "^(\\*|[a-z][a-z0-9-]*)$" },
+      "description": "The models allowed for the role, in no order: any listed model may do the work, and \"*\" allows any model."
     },
     "stage": {
       "type": "object",

package/src/delivery.ts

@@ -39,7 +39,7 @@ export type DeliveryPolicy = {
 };
 
 /** The default operating model: two reviews, the plan and the whole change; the stages between them are written and
- * held by their deterministic gates. Lists are preference orders; a later model is used only as a recorded fallback. */
+ * held by their deterministic gates. Each list is the set of models allowed for the role, in no order; "*" allows any. */
 const DEFAULT_STAGES: Partial<Record<Stage, { writer?: string[]; reviewer?: string[] }>> = {
   plan: { writer: ["codex", "claude-opus", "kimi"], reviewer: ["glm", "claude-opus", "codex", "kimi"] },
   red: { writer: ["glm", "claude-sonnet", "deepseek-flash", "claude-opus", "codex", "kimi"] },
@@ -122,10 +122,7 @@ export function loosenedDelivery(base: unknown, current: unknown): string[] {
     for (const role of ["reviewer", "writer"] as const) {
       const added = c[role].filter(model => !b[role].includes(model));
       if (added.length && b[role].length) out.push(`delivery.stages.${stage}.${role} adds ${added.join(", ")}`);
-      // The lists are preference orders: moving a model earlier, by reordering or removing one before it, makes a
-      // fallback model usable without the fallback.
-      const promoted = c[role].filter(model => b[role].includes(model) && c[role].indexOf(model) < b[role].indexOf(model));
-      if (promoted.length) out.push(`delivery.stages.${stage}.${role} [${b[role].join(", ")}] → [${c[role].join(", ")}] promotes ${promoted.join(", ")}`);
+      // The lists are allowed sets: reordering one changes nothing; only an added model (or "*") is a loosening.
     }
     // A stage that had no writer list now accepting writers is a loosening too: before, nothing could be written there.
     if (!b.writer.length && c.writer.length) out.push(`delivery.stages.${stage}.writer adds ${c.writer.join(", ")}`);
@@ -235,18 +232,12 @@ export async function loadEvidence(root: string, policy: DeliveryPolicy): Promis
   return out;
 }
 
-/** A model must come from the stage's list; a model after the first needs a recorded fallback from each one before it,
- * except one the stage could not use: under different-model independence a reviewer skips every model that wrote the
- * work it reviews (delivery.reviewer-independent's fix hint). Ineligible is not unavailable, so no fallback records it. */
-function checkModel(file: string, at: string, stage: Stage, role: "writer" | "reviewer", actor: Actor, list: string[], fallbacks: Fallback[], defaultRole: "writer" | "reviewer", fallback: DeliveryPolicy["fallback"], ineligible: string[] = []): PlanFinding[] {
-  const out: PlanFinding[] = [];
-  const index = list.indexOf(actor.model);
-  if (index < 0) return [finding("delivery.model-allowed", file, `${at}: ${role} model '${actor.model}' is not in ${stage}.${role} [${list.join(", ")}]`)];
-  // `author` is the 0.9 spelling of `writer`.
-  const recorded = fallbacks.filter(entry => ((entry.role === "author" ? "writer" : entry.role) ?? defaultRole) === role).map(entry => entry.from);
-  for (const skipped of list.slice(0, index)) if (!recorded.includes(skipped) && !ineligible.includes(skipped)) out.push(finding("delivery.model-allowed", file, `${at}: ${role} '${actor.model}' is a fallback, but no fallback from '${skipped}' records why it was unavailable`));
-  for (const from of recorded) if (!list.slice(0, index).includes(from)) out.push(finding("delivery.model-allowed", file, `${at}: ${role} fallback from '${from}' does not precede '${actor.model}' in [${list.join(", ")}]`));
-  return out;
+/** A model must be one the stage allows: its list is the set of allowed models, in no order, and "*" allows any. Which
+ * listed model does the work is the operator's choice at any moment, so no fallback has to explain it; the record still
+ * names the model that ran, and reviewer independence is judged separately (delivery.reviewer-independent). */
+function checkModel(file: string, at: string, stage: Stage, role: "writer" | "reviewer", actor: Actor, list: string[], _fallbacks: Fallback[], _defaultRole: "writer" | "reviewer", _fallback: DeliveryPolicy["fallback"], _ineligible: string[] = []): PlanFinding[] {
+  if (list.includes("*") || list.includes(actor.model)) return [];
+  return [finding("delivery.model-allowed", file, `${at}: ${role} model '${actor.model}' is not allowed for ${stage}.${role} [${list.join(", ")}]`)];
 }
 
 /** The count and window of every recorded fallback are the driver's claims, but explicit ones: outside the policy is out of policy. */