synthesis-before-tranche-craft

name	synthesis-before-tranche-craft
description	>- Use when this capability is needed.

Synthesis Before Tranche Craft

Why this skill exists

The cursor rule akos-synthesis-before-tranche.mdc and the canonical SYNTHESIS_BEFORE_TRANCHE_DISCIPLINE.md tell you when the discipline fires (in-scope tranche classes; trigger conditions; INFO→ FAIL ramp gates) and what the 10 dimensions + 6 tranche classes + 5-option disposition enum are. This skill tells you how to apply them well — how to classify a tranche's class under pressure, how to think about the 3 ERP-engagement-governance surfaces when SYN-06 fires, how to compose the disposition AskQuestion so the operator can ratify in seconds, and how to avoid the false-scope-creep anti-pattern the operator named verbatim at 2026-05-25 Q-A ratify:

"we need this kind of thinking to ensure we scale and don't find false scope creep that we knew was a logical tactical move and design from our part but we're not taking the full design in mind of these processes, why we're doing what we're doing."

Read this skill before authoring any tranche-charter, running scripts/synthesis_before_tranche_check.py --check-charter, or dispositioning a finding from the runbook output.

Principle 1 — Classify the tranche-class FIRST, before drafting any charter frontmatter

The single most load-bearing decision in this discipline is the tranche_class choice. The 6 classes each carry a different DIMENSION_FIRE_RULES always-fire + conditional-fire pair; picking wrong means firing the wrong subset of dimensions, which means missing the design-intent gaps the discipline was created to catch.

Decision tree:

Q1: Does the tranche ship value to a counterparty
    (customer / advisor / partner / ENISA / investor / recruiter /
    collaborator) via a deliverable, surface, or service?

  YES → Q1a: Is the deliverable an artifact you SEND to them
             (PDF / mail body / deck PDF) vs a surface they USE
             (operator dashboard / web page / portal)?

    SEND artifact → tranche_class = external_deliverable
                    (fires all 10 — the most exhaustive class)

    USE surface → Q1b: Is the surface engagement-scoped (a specific
                       counterparty uses it for a specific engagement)
                       vs general (all visitors / all users)?

      engagement-scoped → tranche_class = engagement
                          (fires all 10; SYN-06 ERP-surface citation
                          mandatory)

      general → tranche_class = brand_surface
                (fires 9; SYN-06 conditional only if engagement-scoped)

  NO → Q2: Does the tranche mint or modify a canonical CSV under
            docs/references/hlk/v3.0/Admin/O5-1/People/Compliance/
            canonicals/?

    YES → tranche_class = canonical_csv_mint
          (fires 6 baseline; SYN-02 + SYN-03 + SYN-10 conditional
          when the CSV surfaces in dashboards or external rollups)

    NO → Q3: Does the tranche mint a new Quality Fabric specialty
              (doctrine + chassis + validator + runbook + rule +
              skill + SOP) per akos-quality-fabric.mdc RULE 7?

      YES → tranche_class = specialty_mint
            (fires 7; SYN-03 conditional when end-user scenarios
            apply — usually no, because specialties are
            agent-and-operator-facing)

      NO → tranche_class = internal_governance
            (fires 3 baseline; SYN-01/02/04/09 conditional when
            non-J-OP audience OR ≥ 3 files touched)

Worked decision examples:

• This rule's mint (Commit 2c-a) → specialty_mint (recursive self-application). Fires 7 dimensions: SYN-01 (J-OP + J-AIC named), SYN-02 (cursor + repo channel), SYN-04 (internal CORPINT register; no translation needed), SYN-05 (D-IH-86-EA..EE), SYN-07 (atomic 4-commit-kit per akos-quality-fabric.mdc RULE 7), SYN-08 (reversibility medium; rule + skill + SOP are reversible by git revert; PRECEDENCE + DECISION_REGISTER + process_list edits at Commit 2c-b less so), SYN-09 (validator + runbook self-test in pre_commit + this skill's pre-flight checklist).
• SUEZ POC FULL KIT (Commit 4 target) → external_deliverable (mail + PDF dossier + Excel attachment ship to SUEZ rep). Fires all 10. SYN-06 conditional but recommended (the operator framing explicitly names the ERP-engagement-governance UX shape).
• I82 P1 capability registry mint (Commit 3 target) → canonical_csv_mint. Fires 6 baseline. SYN-10 conditional but recommended (capabilities surface in the operator dashboard eventually).
• Wave-close commit landing a single SOP edit + CHANGELOG entry + files-modified row → internal_governance (3-file scope; J-OP- only audience). Fires 3 baseline only.
• A new _candidates/i-nn-foo.md file → out of scope (under §9 exception 4: per-engagement WIP synthesis or candidate-stage scope sheets).

When unclear, halt and re-read the doctrine §4 + §11 BEFORE posting any AskQuestion. Re-reading takes 5 minutes; re-running a sweep under the wrong class takes 30+ AND surfaces wrong findings the operator has to manually correct.

Principle 2 — Run synthesis BEFORE the tranche commits, not after

The defining temporal posture of this discipline (vs sister disciplines INDEX_INTEGRITY + INTER_WAVE_REGRESSION which run after-commit) is the pre-tranche cadence. Synthesis is design-layer pre-flight; UAT + index- sweep + regression-sweep are post-commit signals. Doing them in the wrong order is the most expensive failure mode.

Bad (synthesis-after):

1. Author tranche files
2. Commit
3. UAT verdict fills in
4. Realize SYN-06 ERP-surface was never named for the engagement
5. Amend the engagement folder + UAT report + commit-amend OR
   ship as-is and accept the design debt

Good (synthesis-before):

1. Classify tranche-class
2. Draft tranche-charter frontmatter
3. Run scripts/synthesis_before_tranche_check.py --check-charter <path>
4. Inspect findings; dispose via inline-ratify AskQuestion
5. Apply ratified scope adjustments to tranche files
6. Commit; UAT verdict fills in cleanly

The discipline's value is in the order of operations. Skipping step 3 defeats the discipline.

Principle 3 — When SYN-06 fires, walk the 3 ERP-engagement-governance surfaces explicitly

SYN-06 is the ERP-engagement-governance UX surface citation. It fires mandatorily for engagement class tranches and conditionally for brand_surface + canonical_csv_mint tranches when the artifact eventually surfaces in a dashboard. The 3 surfaces (per SYNTHESIS_BEFORE_TRANCHE_DISCIPLINE.md §11) are:

Surface 1 — Operator dashboard (Holistika-side)

The dashboard the operator uses to govern the engagement. Lives in HLK-ERP under /operator/<engagement-slug>/. Holds: counterparty brief; objections brief; commercial schedule; settlement statements; tranche progress; risk flags; AIC delegations. Built first because the operator is the always-on actor; collaborators may not use the dashboard but the operator always does.

Surface 2 — Customer dashboard (counterparty-side)

The dashboard the counterparty uses to consume the engagement. Lives in HLK-ERP under /customer/<engagement-slug>/ OR in the counterparty's own ERP via integration. Holds: read-only views of engagement progress; milestone deliverables; settlement summaries; status updates; AI-assisted summaries when the counterparty doesn't log in often.

Recipient-fallback (SYN-10): the customer dashboard MUST have a traditional-means counterpart (email summary; PDF report; SMS update) for counterparties who "don't log so much or see it that much". Citing the surface alone is not enough; the fallback channel is the load-bearing claim per the operator's 2026-05-25 framing.

Surface 3 — ERP workflow join

The workflow logic that joins operator-side governance to customer- side consumption. Lives in HLK-ERP app/workflows/<engagement-class>/

• the runbook scripts that fire on tranche close events. Holds: state transitions; webhook dispatches; data joins (operator dashboard ↔ customer dashboard ↔ external systems); audit trails.

Worked example (SUEZ POC SYN-06 inventory)

For the SUEZ POC engagement (tranche_class: engagement; FULL KIT Commit 4 target), SYN-06 fires mandatorily. The 3 surfaces:

Surface	SUEZ-POC-specific shape
Operator dashboard	/operator/suez-poc/ carrying counterparty brief + Aïsha continuity-role briefing + commercial-schedule (orchestration_broker_thin_margin) + tranche progress (libellé generator + parc engins + PO normalisation) + Aïsha delegation status.
Customer dashboard	/customer/suez-poc/ carrying Phase 1/2/3 milestone status + libellé generator output samples + parc engins lookup widget + Aïsha contact card. Recipient-fallback per SYN-10: email summaries to SUEZ rep on every milestone close + PDF architecture addendum on every Phase transition + Loom video for Phase 1 walkthrough.
ERP workflow join	app/workflows/engagement-suez/ firing on Aïsha continuity tasks + sync to SUEZ CTO office replicability checklist + commercial settlement events into finops.registered_fact.

Naming the 3 surfaces explicitly in the engagement-tranche-charter frontmatter is the SYN-06 satisfaction; not naming them is the SYN-06 FAIL the discipline catches BEFORE commit.

Principle 4 — Compose the disposition AskQuestion so the operator can ratify in seconds, not minutes

Bad (raw finding-by-finding):

AskQuestion: SYN-04 brand-register-citation WARN finding. What to do?
  Options:
    1. fix
    2. defer

Good (per-finding rationale + recommended default per severity class):

AskQuestion: SUEZ FULL KIT synthesis sweep (engagement class; all 10 dims fired).
  3 FAIL + 2 WARN + 1 INFO surfaced; full report at
  reports/synthesis-suez-full-kit-2026-05-26.md.

  Finding #1: SYN-04 brand-register-citation FAIL — engagement-charter
    frontmatter missing audience-class-to-brand-register mapping.
    Recommended: scope-complete — add `brand_register_per_audience:`
    mapping `J-CU.suez-rep: translated-external` + `J-CO.aisha: internal-
    corpint-with-AL3-restriction` to the engagement-tranche-charter
    frontmatter (5-line edit; reversible). Ratify?
    [scope-complete (recommended) / scope-extend / defer-OPS]

  Finding #2: SYN-06 ERP-surface-citation FAIL — operator-dashboard +
    customer-dashboard surfaces named but ERP-workflow-join not cited.
    Recommended: scope-complete — add §"ERP workflow join" subsection
    naming app/workflows/engagement-suez/ + sync-to-SUEZ-CTO-office
    handoff per Aïsha-continuity-role (10-line edit; reversible). Ratify?
    [scope-complete (recommended) / scope-narrow / defer-OPS]

  ... (one option-set per finding; batch all in single AskQuestion call
       when total ≤ 10 findings; split into 2 batches when 11-20)

Each finding gets: severity class explicit + recommended default (rationale embedded inline) + a 3-option subset of the 5-option enum chosen per the finding's reversibility class.

When findings exceed 20, halt and propose splitting the tranche per SYNTHESIS_BEFORE_TRANCHE_DISCIPLINE.md §6 bandwidth-recovery pattern.

Principle 5 — When scope-extend is the disposition, file the forward-pointer in the same chat

scope-extend is the safety valve that keeps tranches atomic without losing the extended scope. But "extend later" with no follow-up pointer is the silent-debt-accumulation anti-pattern. When ratifying scope-extend, file ONE of:

1. OPS_REGISTER row — for operator-class follow-ups with a named owner + ETA. Append to docs/references/hlk/v3.0/Admin/O5-1/People/Compliance/canonicals/dimensions/OPS_REGISTER.csv.
2. Tracker file — for scope that needs more than a one-liner. Mint docs/wip/planning/_trackers/<slug>-tracker.md per the tracker template in akos-conflict-surfacing-and-blocker-trackers.mdc §"Scope-overlap-tracker file shape" (adapt to scope-extend shape).
3. _candidates/ file — for scope that's initiative-sized. Mint docs/wip/planning/_candidates/i-nn-<slug>.md with 3-paragraph scope sheet.

The pointer is filed in the SAME chat that dispositions the finding, not deferred to the next session. Deferred pointer-filing is the shape of the false-scope-creep the discipline catches.

Principle 6 — Update operator-scratchpad BEFORE the tranche commits

The operator-scratchpad continuity dimension (this rule's RULE 1 sister discipline INTER_WAVE_REGRESSION dimension 12) is implicitly enforced by SYN-05 (governance ratification lineage). Before any in-scope tranche commits, the scratchpad MUST carry:

• Tranche ID + tranche class + classification rationale (one line).
• Synthesis sweep verdict (PASS / PASS-WITH-FOLLOWUP / FAIL) with finding count + disposition summary.
• Decision IDs minted at the tranche (D-IH-NN-X for ratifying decisions; D-IH-NN-Y for disposition decisions when novel).
• Forward-pointers filed (OPS row IDs; tracker paths; candidate file paths).
• Next-tranche forward-pointer (the next tranche-charter's slug + expected tranche-class).

The scratchpad entry is the operator's audit trail of synthesis discipline application; it's also the next-session-continuity surface when chat rolls over.

Pre-flight checklist (walk mentally before posting any tranche-charter)

1. ✓ tranche_class resolved via Principle 1 decision tree (NOT defaulted silently).
2. ✓ Tranche-charter frontmatter drafted with: tranche_id, tranche_class, tranche_title, audiences_named[], channels_named[] (if applicable), scenarios_named[] (if applicable), brand_register_per_audience (if non-J-OP-only), ratifying_decisions[], erp_surfaces_named[] (if engagement class), reversibility_class + rationale, closing_loop_test, recipient_fallback_channel (if external surface).
3. ✓ scripts/synthesis_before_tranche_check.py --check-charter <path> invoked; findings table inspected.
4. ✓ Per-finding disposition mapped to one of 5 enum options with per-severity-class recommended default.
5. ✓ AskQuestion batch posted with one option-set per finding; total findings ≤ 10 per batch.
6. ✓ For every scope-extend ratify, the forward-pointer is filed in the same chat.
7. ✓ For every scope-narrow ratify, the removed scope is named in the synthesis report's SYN-07 row.
8. ✓ For every escalate-to-blocker-tracker ratify, the tracker file is minted in the same chat.
9. ✓ operator-scratchpad entry appended with tranche_id, class, verdict, decisions, forward-pointers.
10. ✓ Tranche commit assembled; UAT verdict line drafted ONLY AFTER synthesis sweep verdict is PASS or PASS-WITH-FOLLOWUP.

Anti-patterns (recovery patterns)

• Anti-pattern: classify silently — most common failure when agent under pressure defaults to internal_governance for a tranche that's actually engagement or external_deliverable. Recovery: re-run synthesis under correct class; expect more findings; re- dispose.
• Anti-pattern: synthesis-after-commit — same shape as the operator's "false scope creep" framing. Recovery: amend the tranche if reversibility allows; OR file the post-hoc synthesis report as evidence-only and forward-charter the next tranche to apply synthesis-before.
• Anti-pattern: SYN-06 cite operator-dashboard only, forget customer-dashboard + ERP-workflow-join — partial citation passes the validator at INFO ramp but misses the operator's full design intent. Recovery: walk the 3 surfaces per Principle 3 worked example; amend the tranche-charter.
• Anti-pattern: ratify scope-extend without filing forward-pointer — the silent-debt-accumulation shape. Recovery: file the pointer retroactively in the same chat; never let the disposition close without the pointer.
• Anti-pattern: post the AskQuestion with one finding per call — exhausts the operator's bandwidth + breaks the batching guidance. Recovery: re-compose into ≤ 10 findings per call; split into 2 calls when 11-20.
• Anti-pattern: defer the operator-scratchpad entry to next session — breaks the continuity audit trail. Recovery: write the entry before the commit; never let "I'll catch up next session" rationale win.

Recovery pattern: tranche-charter retroactive backfill

When a pre-2026-05-25 tranche needs to be re-validated under this discipline (e.g., I86 cluster wave-close UATs), the retroactive backfill pattern:

1. Mint a tranche-charter retroactive file at docs/wip/planning/<initiative>/charters/<tranche-slug>.tranche-charter.md capturing the tranche's as-shipped scope using current canonical tag values.
2. Run --check-charter against the retroactive file.
3. Document findings as RETRO-class (no commit changes).
4. File any high-severity findings as forward-charters for the next in-scope tranche to inherit + remediate.

Per SYNTHESIS_BEFORE_TRANCHE_DISCIPLINE.md §9 migration posture, the discipline is forward-only from this rule's mint commit; retroactive backfill is OPTIONAL and used only when the operator explicitly requests it.

Cross-references

• Doctrine: SYNTHESIS_BEFORE_TRANCHE_DISCIPLINE.md.
• Cursor rule: akos-synthesis-before-tranche.mdc — the when layer.
• Paired SOP: SOP-PEOPLE_SYNTHESIS_BEFORE_TRANCHE_001.md.
• Runbook: scripts/synthesis_before_tranche_check.py.
• Validator: scripts/validate_synthesis_before_tranche.py.
• Pydantic chassis: akos/hlk_synthesis_before_tranche.py.
• Sister skill: inline-ratify-craft — disposition AskQuestion craft per Principle 4.
• Sister skill: index-integrity-craft — Wave N specialty mint precedent for pre-flight checklist shape.
• Sister skill: collaborator-share-craft — Wave R+1 13th specialty mint precedent for per-pattern decision-tree shape.
• Sister discipline doctrines: INDEX_INTEGRITY_DISCIPLINE.md, INTER_WAVE_REGRESSION_DISCIPLINE.md, UAT_DISCIPLINE.md, COLLABORATOR_SHARE_DOCTRINE.md.

---

Source: FraysaXII/openclaw-akos — distributed by TomeVault.