Full Parameter Matrix¶
Every tunable in one place — five categories, with defaults and ranges. The reference table for the whole customization surface.
[TUNABLE]
This page is the single big lookup table of every setting you can change in CompassAlpha — what each one does, what its safe starting value is, and how far you can push it. Think of it as the settings menu for the whole framework: you don't need to touch most of it, but when you want to adjust something, you look it up here.
TL;DR¶
The big picture: almost everything here ships with a sensible default, you'll only ever tweak a handful, and the truly structural rules can't be tweaked at all.
This is the consolidated reference for every [TUNABLE] parameter in CompassAlpha, grouped into five categories: project-level, axis-level, operational, cultural, and tech-stack. Each row gives the default and the range. Defaults shown are the conservative defaults (a regulated multi-tenant profile bias: intelligence + low risk). The structural rules (firewall, hard labour, bus, persistence, dual validation) are [INVARIANT] and appear here only where a bounded sub-choice exists.
How to read this table¶
- Default = the conservative starting value. Most projects can adopt these wholesale and tune from there.
- Range = the legal values. Where a range includes an anti-pattern, it's marked.
[INVARIANT]= the parameter's shape is fixed; only the named sub-choice is tunable.- Most adopters touch three or four parameters in their first year. Don't tune everything up front.
Category 1 — Project-level tunables¶
Set once at adoption. They establish the federation's identity and physical layout.
| Parameter | Default | Range / notes |
|---|---|---|
| Project name | (your project) | Used in paths, namespaces, branding. |
| Reviewer-state repo | /path/to/reviewer-state |
Dedicated remote; sibling of substrate. |
| Substrate repo | /path/to/substrate |
The project being built/governed. |
| Host | (your host) | Where tiers run; determines NO-TMP + flush cadence. |
| AI agent / harness | reference top-tier agent at high effort | Any harness meeting the 3-point contract. |
| Founder identity | one person | One person, or a small team treated as a single founder-role. |
| Canonical state-artifact count | 5 (boot / ledger / handover-protocol / handover-log / leftovers) | Smaller projects may collapse to 3. |
Category 2 — Axis-level tunables¶
Set per axis. Each axis declaration fills these in.
| Parameter | Build-axis default | Doctrine-axis default | Range / notes |
|---|---|---|---|
| Axis name | build |
doctrine |
Free-text. |
| Mentor-1 name | (project's name) | (project's name) | Generic vocabulary: Mentor-1. |
| Mentor-2 name | (project's name) | (project's name) | Generic vocabulary: Mentor-2. |
| Doer name | (project's name) | (project's name) | Generic vocabulary: Doer. |
| Deliverable type | code | governing document + primitives | Any deliverable. |
| Cycle stages | CP1 → CP2 → CP3 | S1 → S5 | Project-defined (stage taxonomies). |
| Charter posture | LOCKED | UNLOCKED (amendment only) | LOCKED / UNLOCKED / DECOUPLED. |
| Cycle activator | dispatch trigger | walk + queue depth | Per axis. |
| Ratification pattern | per-module tag | per-entity sub-bump + cycle-end major bump | Per axis. |
| Lanes enabled | all four | Doctrine Cycle + light lanes | Per work-granularity lanes. |
| Locality | host-of-substrate | host-of-substrate | Override if a tier runs elsewhere. |
Category 3 — Operational tunables¶
Tuned per cycle. These are the performance/cost dials that move the 5-axis trade-off most.
| Parameter | Default | Range | Primary trade |
|---|---|---|---|
| Concurrency mode | Sequential (LAYGO) | LAYGO / Pipelined / Parallel-independent / Parallel-doer | Speed ↑ vs Risk ↑ |
| Mentor-1 lifecycle | per-cycle long-running | per-cycle / per-epoch | Cost ↓ vs Continuity ↓ |
| Mentor-2 lifecycle | fresh-per-event | fresh-per-event / per-dispatch | Cost ↓ vs Pollution ↑ |
| Doer lifecycle | fresh-per-slice | [INVARIANT] (fresh-per-turn in Surgical lane) |
— |
| Doer effort level | high | high / normal / low | Intelligence ↑ vs Cost ↑ |
| Doer context scope | wide / liberal | wide / narrow / declared-only | Intelligence ↑ vs Cost ↑ |
| Doer model tier | top-tier | top / mid / low | Intelligence ↑ vs Cost ↑ |
| Mentor model tiers | top-tier | top / mid | Intelligence ↑ vs Cost ↑ |
| Provider strategy | single provider | single / mixed | Consistency ↑ vs Integration cost ↑ |
| Provenance verification | strict (cite-by-substrate) | strict / spot-check / trusted | Correctness ↑ vs Speed ↑ |
| Memory accumulation rate | liberal | liberal / conservative | Learning ↑ vs Noise ↑ |
| Memory scope | cross-axis | cross-axis / per-axis | Sharing ↑ vs Boot-read ↓ |
| Memory garbage collection | retire-never (HISTORICAL-mark) | retire-never / major-version retirement | Audit depth ↑ vs Boot-read ↓ |
| Rotation cadence | 2–4 weeks preventive | 1–6 weeks | Pollution ↓ vs Continuity ↑ |
| Deferral-queue depth for cycle activation | ~20–30 soft | 10–50 | Doctrine recency ↑ vs Cycle frequency ↓ |
| Cycle-tail acceptance | strict (all close) | strict / liberal (N-of-M) | Predictability ↑ vs Speed ↑ |
| Audit / digest verbosity | verbose | verbose / concise | Audit depth ↑ vs Cost ↑ |
| Cross-tier visibility | own-folder only | own-only / parent-summary / full (anti-pattern) | Containment ↑ vs Continuity ↓ |
| Invariant count | many (regulated) / minimal (bootstrap) | minimal → many | Intelligence ↑ vs Cost ↑ |
| Invariant enforcement strictness | strict | strict / spot-check / trusted | Correctness ↑ vs Speed ↑ |
| Tooling count + coverage | many (regulated) / minimal (bootstrap) | minimal → many | Catch-rate ↑ vs Upfront cost ↑ |
| Specialised-agent count | several / none | none → many | Per-slice fit ↑ vs Maintenance ↑ |
| NO-TMP rule | ON | ON / OFF (durable /tmp) | Durability ↑ vs Convenience ↓ |
| Flush cadence | every flush trigger pushes | every-trigger / lighter | Durability ↑ vs Push overhead ↓ |
Category 4 — Cultural tunables¶
Set by the project's working culture. These shape how the founder relates to the federation.
| Parameter | Default | Range / notes |
|---|---|---|
| Founder role scope | RELAY + lost+found only | Expanding it bloats founder cognitive load; discouraged. |
| Founder involvement level | low (RELAY + lost+found) | low / medium / high — trades cost ↓ vs founder load ↑. |
| Founder-call boundary | charter proposals / scope-shifts / gate-count changes / genuine value-intent | The four-class shape is [INVARIANT]; the items in each class are tunable. |
| Lost+found protocol | one-liner ping → founder rules → one-liner ping back | The one-liner mechanism is [INVARIANT]. |
| Pacing | LAYGO (one unit at a time, learning expected) | Faster pacing acceptable after first cycle proves the rhythm. |
| Founder team protocol | single founder | single / small-team-as-one-role. |
| UX-vision ownership | founder (de facto UX lead) | The founder's irreducible taste exception — [INVARIANT] that it's founder-owned. |
Category 5 — Tech-stack tunables¶
Mostly [INVARIANT] — git is the bus — with a few genuine choices.
| Parameter | Default | Range / notes |
|---|---|---|
| VCS | git | [INVARIANT — git is the bus] |
| Remote service | a hosted git remote | Could be self-hosted; just needs a push-target. |
| Commit discipline | isolated index + commit-tree + refspec push |
[INVARIANT — prevents sibling-session contamination] |
| Worktree mechanism | one isolated worktree per slice | [INVARIANT — pollution firewall] |
| AI harness | reference top-tier agent | Must support effort-level, fresh-session-per-slice, deterministic boot. |
| Effort level for Doer | high | high / normal / low (intelligence vs cost). |
Quick-start: the parameters most adopters actually touch¶
If you read nothing else, these are the dials with the highest leverage for a new adoption:
- Concurrency mode — LAYGO first; graduate to Pipelined after first close.
- Doer model tier + effort — top-tier/high to start; mid-tier-for-simple-slices to cut cost later.
- Enrichment level (invariants/toolings/agents) — minimal first; earn each addition with the rule-of-three.
- Founder involvement — keep it RELAY + lost+found; resist the urge to expand it.
- Lanes enabled — enable all four, holding Surgical until first close.
Everything else can ride the conservative defaults until a specific pain motivates a change.
Remember this¶
- This is a reference table, not a checklist. You don't configure all of it — you look things up here when you need them.
- Every row has a default that's safe to start with. New adopters typically change only three or four parameters in the whole first year.
- A
[TUNABLE]is a dial you may turn; an[INVARIANT]is fixed by design and only its named sub-choice (if any) is yours to set. The fixed ones are the framework's load-bearing walls — see the mental model. - When in doubt, leave it on the default and let real pain — not guesswork — tell you what to tune.
How this connects¶
- Tunables overview — the 5-axis trade-off each operational parameter moves.
- Operating presets — named bundles that set many of these rows at once.
- Each parameter's detail page: axis declarations · concurrency modes · context patterns · AI model choices · work granularity lanes · stage taxonomies · invariants, toolings & agents · memory policy.
- Axioms — the
[INVARIANT]floor under every row.