@eir-labs/coltrane 0.24.13 → 0.24.15

package/standards/reconcile-work-order-v0.json

@@ -0,0 +1,137 @@
+{
+  "slug": "reconcile-work-order-v0",
+  "domain": "software-change",
+  "status": "active",
+  "agent_slugs": [
+    "receipt-mapper",
+    "reconciliation-auditor",
+    "proof-composer",
+    "default-adjudicator"
+  ],
+  "phases": [
+    {
+      "name": "cold-map",
+      "chairs": [
+        {
+          "role": "cold-map",
+          "agent_slug": "receipt-mapper",
+          "depends_on": [],
+          "input_contract": [
+            "change-request"
+          ],
+          "output_contract": [
+            "reconciliation-map"
+          ],
+          "optional_outputs": [],
+          "required_skills": [],
+          "turn_budget": 25
+        }
+      ],
+      "intent": "The deterministic clause-to-receipt mapping. Read the work order the dispatch names — its articles, terminals, receipts_due, and reconciliation contract (owner, budget, due date) — and the receipts landed against it: chancery_work_order_receipt rows joined to the sealed output chain, plus the completed_output_id FK. Match every clause to its receipts by stated equality; every clause with no receipt is listed absent, in so many words, never omitted. This is clerical work: no model judgment falls here and the chair spends nothing against the reconciliation budget — same order, same receipts, same map, on any machine."
+    },
+    {
+      "name": "hot-audit",
+      "chairs": [
+        {
+          "role": "hot-audit",
+          "agent_slug": "reconciliation-auditor",
+          "depends_on": [
+            "cold-map"
+          ],
+          "input_contract": [
+            "change-request",
+            "reconciliation-map"
+          ],
+          "output_contract": [
+            "change-verdict"
+          ],
+          "optional_outputs": [],
+          "required_skills": [],
+          "turn_budget": 30,
+          "turn_reserve": 6
+        }
+      ],
+      "intent": "The refutation pass — the half that SPENDS. Within the order's declared reconciliation_budget (carried on the change-request; the chair's turn budget is the structural cap it may only narrow, never widen), attempt to refute what the receipted clauses claim: re-run the checks a receipt says went green, probe the deployed thing a receipt points at, open the sealed output and test its claims against the world. Every finding is cited — the tool call, check name, path, or URL it rests on — and a finding without a citation does not enter the verdict. When the budget exhausts, stop and seal the partial audit with every unaudited clause named; absent receipts pass through as absent, never audited into existence."
+    },
+    {
+      "name": "proof-of-work",
+      "chairs": [
+        {
+          "role": "proof-of-work",
+          "agent_slug": "proof-composer",
+          "depends_on": [
+            "cold-map",
+            "hot-audit"
+          ],
+          "input_contract": [
+            "reconciliation-map",
+            "change-verdict"
+          ],
+          "output_contract": [
+            "proof-of-work-contract"
+          ],
+          "optional_outputs": [],
+          "required_skills": [],
+          "turn_budget": 15
+        }
+      ],
+      "intent": "The composed contract: the cold map, the audit's findings, and a verdict per clause — every clause of the order, each row joining the map's receipt status with the audit's finding verbatim — written as a sealed output, which is what makes it content-addressed: the engine stamps content_sha and the input_shas of exactly the two seals composed from, and no agent hashes anything. Copies only; the composer draws no new verdict. This document is what the sovereign settles against and what a default cites."
+    },
+    {
+      "name": "default-judgment",
+      "chairs": [
+        {
+          "role": "default-judgment",
+          "agent_slug": "default-adjudicator",
+          "depends_on": [
+            "cold-map",
+            "proof-of-work"
+          ],
+          "input_contract": [
+            "change-request",
+            "reconciliation-map",
+            "proof-of-work-contract"
+          ],
+          "output_contract": [
+            "reconciliation-record"
+          ],
+          "optional_outputs": [],
+          "required_skills": [],
+          "turn_budget": 15
+        }
+      ],
+      "intent": "The default ruling — deterministic, sealed BEFORE the human seat so a parked approval cannot mask it. An order past reconciliation_due_at with clauses the cold map lists absent (unreceipted) reconciles as DEFAULTED: the record seals reconciliation_state 'defaulted' naming every absent clause_ref, and that state lands in chancery_work_order.reconciliation_state. The conductor's sweep (chancery_work_order_reconciliation_due) fires this standard at the committed date regardless of any completion signal, so abandonment cannot hide: the cold map runs, shows the absence, and the default is sealed with no human in the loop to not-show-up. Not past due, or every clause receipted: the record seals 'pending' and waits for the sovereign. This seat never seals 'satisfied'."
+    },
+    {
+      "name": "sovereign-settle",
+      "chairs": [
+        {
+          "role": "sovereign-settle",
+          "human": true,
+          "depends_on": [
+            "proof-of-work",
+            "default-judgment"
+          ],
+          "input_contract": [
+            "proof-of-work-contract",
+            "reconciliation-record"
+          ],
+          "output_contract": [
+            "reconciliation-record"
+          ],
+          "optional_outputs": [],
+          "required_skills": []
+        }
+      ],
+      "intent": "The parked human step — an approval office, not an agent. The gig PARKS here (awaiting_approval) until the reconciliation_owner — or the governor, where the order seats that terminal — reads the proof-of-work contract and settles. Settling seals the reconciliation-record with reconciliation_state 'satisfied' and settled_by naming the seat that settled — and that seal is the ONLY door to that state: no model seat can rule its own work satisfied, and a completion sealed without a proof-of-work the sovereign settled is forbidden by the order that commissioned this standard. A defaulted record needs no settling to stand — it was sealed upstream — but the owner may still settle over it once the absences are cured."
+    }
+  ],
+  "input_types": [
+    "change-request"
+  ],
+  "output_types": [
+    "reconciliation-record"
+  ],
+  "description": "Reconciliation as a standing office over WORK ORDERS — the standard WO-F02 ('The Reconciliation', Documenso #1917537, signed 2026 and never built) owes, generalizing reconcile-gig-v0's shape from gigs to signed orders. It runs against the chancery schema's reconciliation contract: chancery_work_order.reconciliation_owner / reconciliation_budget / reconciliation_due_at / reconciliation_state ('pending'|'satisfied'|'defaulted'), the chancery_work_order_receipt table joining clauses to the sealed output chain, and the chancery_work_order_reconciliation_due sweep view the conductor fires — the arming door (signing requires owner + budget + due) guarantees every sealed order carries the contract this standard executes. Five offices: COLD-MAP (deterministic clause-to-receipt mapping, absences listed, no judgment, no spend), HOT-AUDIT (refutation that spends within the order's declared reconciliation_budget, every finding cited), PROOF-OF-WORK (the composed per-clause contract, content-addressed by the engine's own seal), DEFAULT-JUDGMENT (past reconciliation_due_at with clauses unreceipted reconciles as defaulted — sealed before the human seat so abandonment cannot hide behind an approval nobody gives), and SOVEREIGN-SETTLE (the parked human office: only the owner's settling records reconciliation_state 'satisfied'). ADICO, roughly: the reconciliation_owner [A] must reconcile every clause of the order [I] against sealed receipts [C: at reconciliation_due_at, swept regardless of completion signals] or the order defaults [O] — and the enforcement is structural, not hortatory: the sweep view fires on the date, the default seat cannot be parked, and 'satisfied' is unreachable except through the human seal.",
+  "reserve_pool": 12
+}

package/standards/software-change-v1.json

@@ -2,7 +2,7 @@
   "slug": "software-change-v1",
   "domain": "software-change",
   "status": "active",
-  "description": "The default software-change standard. Read the material exhaustively, fix the direction, settle the structure, make the change, then say whether what was built is what was decided. The seams are deliberate: the seat that reads does not decide, the seat that decides does not build, and the seat that builds does not judge its own work. Only the reading seat holds a tool grant, and it is read-only \u2014 the change is emitted as reviewable content, never applied.",
+  "description": "The default software-change standard. Read the material exhaustively, fix the direction, settle the structure, make the change, then say whether what was built is what was decided. The seams are deliberate: the seat that reads does not decide, the seat that decides does not build, and the seat that builds does not judge its own work. Only the reading seat holds a tool grant, and it is read-only — the change is emitted as reviewable content, never applied.",
   "agent_slugs": [
     "john",
     "bill",
@@ -25,7 +25,7 @@
           "required_skills": []
         }
       ],
-      "intent": "Read the material the change lands in, to the declared boundary and no further: entry points, the files the change plausibly touches, the checks that already cover the region, the conventions it must hold to \u2014 and the unknowns, which are part of the reading, not a failure of it. No decision is taken here."
+      "intent": "Read the material the change lands in, to the declared boundary and no further: entry points, the files the change plausibly touches, the checks that already cover the region, the conventions it must hold to — and the unknowns, which are part of the reading, not a failure of it. No decision is taken here."
     },
     {
       "name": "frame-change",
@@ -45,7 +45,7 @@
           "required_skills": []
         }
       ],
-      "intent": "Fix the direction from the reading: scope, non-goals, stop condition, and the alternatives set aside with their reasons. Stated before any structure exists, so the structure is answerable to it. Settled for the run \u2014 later phases append observations, never revisions."
+      "intent": "Fix the direction from the reading: scope, non-goals, stop condition, and the alternatives set aside with their reasons. Stated before any structure exists, so the structure is answerable to it. Settled for the run — later phases append observations, never revisions."
     },
     {
       "name": "plan-change",
@@ -93,7 +93,7 @@
           ]
         }
       ],
-      "intent": "Make the change the structure implies, as patch content plus the checks it introduces. Where the artefact left the plan, the departure is recorded with its reason \u2014 the same seat that settled the structure is the one accountable for saying where it did not hold."
+      "intent": "Make the change the structure implies, as patch content plus the checks it introduces. Where the artefact left the plan, the departure is recorded with its reason — the same seat that settled the structure is the one accountable for saying where it did not hold."
     },
     {
       "name": "verify-change",
@@ -115,7 +115,7 @@
           "required_skills": []
         }
       ],
-      "intent": "Measure the change against the commitments as fixed, criterion by criterion: which acceptance criteria are unmet, where the scope drifted, and \u2014 separately and always \u2014 which checks could not be executed and were reasoned about instead. When the evidence does not reach the question, the verdict is insufficient evidence, not a pass."
+      "intent": "Measure the change against the commitments as fixed, criterion by criterion: which acceptance criteria are unmet, where the scope drifted, and — separately and always — which checks could not be executed and were reasoned about instead. When the evidence does not reach the question, the verdict is insufficient evidence, not a pass."
     },
     {
       "name": "approve",

package/standards/user_flow_correctness.json

@@ -46,9 +46,9 @@
   "scoring": {
     "scale": "0-1 per criterion (float)",
     "pass_threshold_per_criterion": 0.6,
-    "overall_pass_threshold_sum": 3.0,
-    "overall_pass_threshold_max": 5.0,
-    "noise_band_repro": 1.0
+    "overall_pass_threshold_sum": 3,
+    "overall_pass_threshold_max": 5,
+    "noise_band_repro": 1
   },
   "prompt_templates": {
     "sense": "You are the SENSE phase of the user_flow_correctness standard.\n\nINPUT: gig_input.transcript_json (a JSON-encoded array of turn objects). Each turn has: { user_intent: string, claude_response: string, tool_calls: array, post_turn_genome_state?: object|null, malformed_input?: boolean }.\n\nTASK: parse the transcript into a normalized `user-flow-transcript`. For each turn capture turn_idx, user_intent (verbatim), claude_response (verbatim), tool_calls (array of { tool_name: string, args?: object }), post_turn_genome_state (object or null), and malformed_input (boolean, default false). If the input cannot be parsed, set turns=[] and put the diagnosis in notes — do NOT fabricate turns.\n\nOUTPUT (single JSON object, no fence, no prose):\n{\n  \"transcript_present\": <bool>,\n  \"turn_count\": <int>,\n  \"turns\": [\n    { \"turn_idx\": 0, \"user_intent\": \"<text>\", \"claude_response\": \"<text>\", \"tool_calls\": [{\"tool_name\":\"<slug>\",\"args\":{}}], \"post_turn_genome_state\": null, \"malformed_input\": false }\n  ],\n  \"notes\": \"<honesty-note about parse failures or empty input>\"\n}",