Skip to content

docs: emissions research harvest v0.1 (AIM activation criteria + candidate params) #1

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Gio2050 merged 1 commit into from
Jul 28, 2026
Merged

docs: emissions research harvest v0.1 (AIM activation criteria + candidate params) #1

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
67 changes: 67 additions & 0 deletions docs/AFI_EMISSIONS_RESEARCH_HARVEST.v0.1.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,67 @@
# AFI Emissions Research Harvest v0.1

**Register:** NON-CANONICAL research input (per `afi-governance/decisions/math-authority-v0.1` §3: afi-econ content is non-canonical unless a formula is formally promoted). This document **decides nothing**, self-labels nothing canonical, and authorizes no implementation. Every mechanism named here that touches mint, rewards, settlement, or economic activation is **reserved to the unfiled CHAIN-GOV governance domain**. Its purpose is to preserve the forward-facing design inputs from the pre-alpha emissions research corpus ("Emissions Article", Sept–Oct 2025) in current-architecture terms, so that the original artifacts could be deleted forward-only (2026-07-28). BT-86b (`mint-formula-bt-86b-alignment-v0.1`, accepted) cites this research's whitepaper doctrine (*"AFI: An Agentic Financial Intelligence Market"* §9.4/§9.5/§9.7); this harvest is the surviving in-org record of the parts of that research not already governed or implemented.

What is already covered elsewhere and deliberately NOT restated here: the three-phase 86B emissions schedule (canonical implementation: `afi-math` `src/emissions/emissionsSchedule.ts`, pinned as production B(t) by BT-86b D1); the epoch-budget → role-pools → pro-rata-verified-credits skeleton (BT-86b D3); the AFI Index and AIM simulation mechanics (implemented in this repo's `afi_econ_kit`: `index.py`, `emissions.py`).

---

## 1. Launch stance (as designed in the research; consistent with BT-86b)

Emissions follow the published baseline schedule B(t). The AFI Index is computed and published each epoch as a KPI but **does not modulate issuance at launch** (AIM_t ≡ 1 — matching BT-86b D6). The adaptive layer below is optional, bounded, and OFF until a governance activation decision.

## 2. Adaptive Issuance Multiplier (AIM) — designed form

Per-epoch issuance: `E_t = B(t) · AIM_t`, with AIM within published floor/ceiling.

- Smoothed impact signal (EMA): `Ī_t = (1−λ)·Ī_{t−1} + λ·I_t`, where I_t is the AFI Index.
- Coupling (clipped affine): `AIM_t = clip(1 + κ·(Ī_t − I*), minMult, maxMult)`.
- Per-epoch rate limit: `|AIM_t − AIM_{t−1}| ≤ δ_max`.
- Monthly budget guardrail bounding deviation from baseline: ≤ B_max.
- Circuit breakers pause adaptation / revert AIM_t to 1 on safety breaches (e.g., excessive reversions); parameters are public, versioned, and change only at epoch boundaries. Setting AIM_t ≡ 1 recovers the policy-constant path.

Stability layer generalized to allocation as well: `|E_t − E_{t−1}| ≤ Δ_E` and per-role gauge movement `|g_r(t) − g_r(t−1)| ≤ Δ_G`; role caps/guards ensure no role can absorb the entire flow in one epoch.

AFI Index component chain (implemented in `afi_econ_kit`): `AN_t` (active network), `EG_t` (economic growth), `Cov_t` (coverage) → `I_t` → `Ī_t` → AIM.

## 3. AIM activation criteria (readiness checklist — future CHAIN-GOV input)

All must hold before the closed loop is engaged:

1. **Data sufficiency & diversity** — ≥ K finalized evidence records spanning multiple venues/instruments (K set by governance; order tens of thousands); demonstrated multi-regime coverage (quiet / volatile / eventful, by volatility & volume quantiles).
2. **Measurement stability** — rolling 26–52-epoch drift of Index component weights < Δ_weights_max (e.g., 5–10%); out-of-sample error for attribution/efficiency components below E_oos_max on k-fold backtests; dispute/reversion rate < r_dispute_max (e.g., 1%) for N consecutive epochs.
3. **Integrity & observability** — attribution/watermarking coverage > p_tag_min; no unresolved governed-contract/schema migrations affecting Index inputs; open per-epoch CSVs publishing components, Ī_t, applied AIM, and parameter versions.
4. **Safety simulation** — re-run ≥30k Monte-Carlo with the proposed (κ, minMult, maxMult, λ) and confirm monthly minted-vs-baseline deviation ≤ B_max (e.g., 5%) and that rate-limit/circuit-breaker thresholds are never breached in stress paths.
5. **Governance & rollout** — pre-approve (κ, minMult, maxMult, λ, δ_max) and change windows; run a dry-run window publishing computed-but-unapplied AIM for M epochs; on activation publish test vectors and backtests; subsequent changes require notice and bounds.

**Timing recommendation:** enable coupling only after the 4-year bootstrap phase (~33⅓% minted), during the 33%→80% growth window — by then receipts span regimes, reducing over-reaction to sparse early data. Keep κ small, enforce floors/ceilings, cap per-epoch deltas.

## 4. Candidate parameters and Monte-Carlo safety evidence (research values — ungoverned)

Whitepaper simulation runs (cap 86e9, 52 epochs/yr, milestones y33=4 / y80=28 / y100=54):

- Open-loop multiplier: α=0.08, minMult=0.85, maxMult=1.5, afiScale=1.0.
- Closed-loop extras: ceBeta=0.15, ceKappa=0.2, ceMin=0.9, ceMax=1.3.
- Scenario families explored (Python model): conservative (α=0.08, k=0.90, mult∈[0.80,1.50]), nominal (α=0.10, k=0.90, mult∈[0.40,2.00]), aggressive (α=0.15, k=0.60, mult∈[0.50,2.50]).

30,000-scenario Monte-Carlo outcomes (milestone years, cumulative share):

| Milestone | p5 | p50 | p95 |
|---|---|---|---|
| 33⅓% | 4 | 4 | 5 |
| 80% | 27 | 28 | 29 |
| 100% | 49 | 54 | 54 |

Average multiplier across scenarios: mean 1.0132, p5 0.9994, p95 1.0451 — i.e., the bounded adaptive layer barely perturbs the milestone structure. This is the empirical safety basis a future AIM-parameter decision can cite or re-run.

## 5. Reputation → economics rule (eligibility vs realized share)

As stated in the whitepaper economic model, aligned with D-CONST-5 and the afi-core invariants:

> **PoI** (capability on pinned, deterministic suites) **governs eligibility**; **PoInsight** (live, receipt-verified contribution) **governs realized share**. Reputation decays without recent verified work. **Neither overrides UWR scoring or finality.**

All PoI/PoInsight formulas, weightings, and economic conversion remain reserved (CHAIN-GOV / future reputation decision); this records the designed *shape* of the linkage only.

## 6. Schedule-tail discrepancy requiring an eventual owner ruling

The research corpus and all whitepaper simulation runs target **100% of cap at year 54** (phases 4/24/26). The governed `afi-math` schedule pinned by BT-86b D1 reaches 100% at **~year 53** (phases 4/24/25). BT-86b records the pin as made "without amending whitepaper doctrine," and the one-year tail difference is documented nowhere else. Resolution (bless ~53 explicitly, or amend to 54) belongs to the eventual CHAIN-GOV schedule ratification; it has no effect on the cap, the milestones through year 28, or any current behavior.
Loading