Skip to content
85 changes: 85 additions & 0 deletions .codex/plans/2026-08-09-production-systems-program.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,85 @@
# Production-Systems Program — implementation plan

Date: 2026-08-09

Owner: Anthony

Implementation owner: cross-agent Limen conduct graph

Canonical graph: `institutio/positioning/program.yaml`

## Objective

Turn the owner’s demonstrated ability to build systems that build products into an evidence-backed
production-systems identity, a proof-led public front door, a bounded commercial offer, a qualified
inbound and delivery system, and ultimately a governed product-to-operator handoff loop. Preserve
valid live-profile evidence while correcting stale research, unsupported implications, and unclear
prominence decisions.

## Implementation shape

The program is represented once in the tracked manifest and projected idempotently into GitHub as:

- one root issue;
- fifteen phase issues, P00 through P14;
- 111 atomic work issues with explicit dependencies, repository/path scope, capabilities, dynamic
reasoning class, effect boundary, human gates, deliverables, acceptance, executable receipt
predicate, return evidence, and rollback path;
- one milestone and three dedicated labels;
- one deterministic issue-number map and generated human-readable index;
- thirteen validated execution chunks with assigned conductor models and copy/paste relay prompts.

The GitHub graph is a coordination projection, not a lease. Every mutation leaf is claimed by a
registered native lane through the conduct broker. Under the owner’s 2026-08-09 override, every
root/phase/leaf has an exact model and effort assignment validated against the live Codex catalog.
An unavailable assignment fails blocked and requires reviewed reassignment; there is no silent
fallback table.

## Sequence

1. P00 establishes the program control plane, validates it, projects GitHub issues, and proves one
non-Codex broker canary.
2. P01 repairs the wall-clock-sensitive CI test and lands the existing truth/custody foundations in
PRs 2136 and 2141.
3. P02 inventories the complete live estate, creates evidence packets, audits LAVREA percentile
inference, and adjudicates stale research claim by claim.
4. P03–P04 settle the production-systems identity and bounded offers.
5. P05–P06 produce flagship proof and design progressive disclosure while P11 builds the secure
delivery operating system on its own parallel branch.
6. P07–P09 build and verify public surfaces, create safe capture, and stage or owner-publish
proof-led material only through named gates.
7. P10-W01 through W07 build conversion before the pilot. P12-W01 and W02 then unlock P10-W08 while
the remaining P12 evidence work proceeds; P10 and P12 close together without a phase deadlock.
8. P13 inventories product candidates and pilots a legally and operationally bounded domain-operator
handoff or records an evidence-backed no-go.
9. P14 returns every result to the claims and decision systems, proves rollback, and requires two
unchanged Omega passes before the root closes.

## Completion contract

Each leaf first meets its human-readable acceptance condition. The executor then runs a
task-specific, non-circular command and posts a marked structured receipt to the leaf issue. The
leaf’s executable predicate, `python3 scripts/positioning-program.py --verify-work <WORK-ID>`, fails
unless that receipt is current, successful, authorized, exact-head-bound, evidence-linked, and
rollback-aware. Phase and root closure require their complete child set; Omega additionally
revalidates every leaf receipt and two unchanged whole-program state digests.

## Current publication transaction

1. Validate the local graph, generated payload sizes, receipt contract, and scoped repository gates.
2. Commit and push the isolated `codex/production-systems-program` branch.
3. Open a draft pull request before remote issue creation so the implementation has a durable code
owner.
4. Apply the idempotent GitHub projection and write all resulting issue identities back to the
tracked map and index.
5. Verify exact remote parity, commit the back-links, push the updated branch, and cross-link the
root issue and draft pull request.
6. Begin execution at the sole ready leaf; do not bypass phase or work dependencies.

## Recovery

The projection never closes or reopens issues, merges, sends, publishes, spends, changes DNS, signs,
or edits `tasks.yaml`. A failed projection reruns safely from stable markers. A manifest change
invalidates stale acceptance receipts by digest. A bad public claim is quarantined at the claims
ledger; a bad release returns to the last known-green release; a failed offer or handoff returns its
outcome to the appropriate decision ledger before any narrative revision.
Loading
Loading