SSM-AI – Appendix N — Economics & Rollout Worksheets (N6–N9)

Procurement parity, dashboard blueprint, acceptance gates, and rollout plan.

N6) Procurement rules (bounded and auditable)

Purpose. Create a deterministic procurement framework where vendor choice follows reproducible symbolic lanes. Classical values stay pristine (phi((m,a)) = m); decisions depend on bounded alignment (a ∈ (-1,+1)) and chooser metrics (RSI_env).

Rule set (copy-pasteable):

Equal manifest. All vendors must use the same knobs_hash (manifest).
- Same KPIs, bands, weights, gains, eps_a, and eps_w.
- Same normalization, time window, and units.
Comparable scope. Compare only rows where both vendors share (svc, week_start, band >= A0).
Symbolic parity. Compute RSI_pool_env for each vendor via tanh( sum(U) / max(sum(W), eps_w) ).
- Vendor with higher RSI_pool_env wins the symbolic lane.
Audit trail. Retain CSV + JSON with (svc, vendor, RSI_pool_env, band, knobs_hash, stamp).
- No hidden parameters or subjective overrides.
Fallback. If parity fails (RSI_pool_env within hysteresis, e.g., ±0.02), default to classical m tie-breaker.
Ethics and neutrality. Bands (A++..A–) act only as bounded indicators. No demographic, textual, or PII terms allowed in lens design.

One-line summary. Procurement = alignment-based, order-invariant, fully auditable, never editorial.

N7) Dashboard blueprint (one-page weekly view)

Week	Service	Vendor	RSI_pool_env	Band	Tokens/Success	Retry %	Latency p95	Cost/1k	Net Savings USD	knobs_hash
2025-10-13	Search	A	0.5521	A+	320	7.0	310	0.90	182,400	sha256:…
2025-10-13	Search	B	0.4012	A0	355	10.0	345	1.00	0	sha256:…
2025-10-13	Tools	A	0.6125	A+	190	6.0	280	0.90	96,300	sha256:…
2025-10-13	Tools	B	0.4550	A0	215	9.0	312	1.00	0	sha256:…

Legend.

Band coloring. A++ green, A+ light green, A0 yellow, A– orange, A–– red.
Hover info. Manifest digest, last replay, U/W sum, and RSI stability delta.
Drill-down. Clicking row opens replay chart (RSI_pool_env vs. week).
Totals. Bottom line aggregates weekly savings and symbolic ROI (RSI_env_mean).

Implementation tip. This table can be rendered directly from the CSV emitted by weekly_rollup().

N8) Acceptance checklist (must pass before live rollout)

Check	Description	Method	Status
Parity	`phi((m,a)) = m` for every row	Spot check random sample	✓
Order invariance	Shuffle → identical `(U,W)` outcome	Batch & stream replay	✓
Determinism	Same `knobs_hash` ⇒ identical outputs	Dual-run diff	✓
Boundedness	`	a	<1`,`
Gate purity	Gate scales RSI only	Manifest verify	✓
Replayability	tanh(U/max(W,eps_w)) reproducible	Roll-up replay	✓
Audit logging	Stamps and hashes present	Log review	✓
Security	No PII in lens or telemetry	Manifest lint	✓

All must pass before production cutover.
Failures invalidate symbolic comparability for that week.

N9) Rollout plan (deterministic and reversible)

Phase 1 – Shadow Mode (2–3 weeks)

Run SSM-AI lanes in parallel; do not influence routing.
Validate parity (phi((m,a)) = m), clamps, and replay invariance.
Produce weekly dashboard via reducer; confirm 1:1 with baseline.

Phase 2 – Advisory Mode (4–6 weeks)

Begin reporting symbolic deltas to CFO and ops.
Tune lens gains (c, Unit) to reach |a| ≈ 0.3–0.8.
Validate RSI_pool_env correlation with observed savings.

Phase 3 – Active Mode (steady state)

Use RSI bands (A++, A+, A0, A–, A––) for vendor weighting and scheduling.
Gate retry rate and tool calls with g_t or RSI_env thresholds.
Maintain manifest digest and weekly stamps for full traceability.

Rollback (deterministic).
If anomalies occur, disable alignment lane; fall back to m directly (collapse parity ensures same baseline behavior).
To recover: recompute a := tanh(U / max(W, eps_w)) from last valid (U,W); confirm all values bounded.

Stamp example (one line, ASCII)

SSMCLOCK1|iso_utc|k=economics|U=1.236000|W=2.000000|a=0.617000|RSI=0.610000|g=0.90|RSI_env=0.549000|band=A+|manifest=knobs_hash

One-line takeaway.
Bounded procurement, auditable dashboards, deterministic replay, and no retraining — symbolic economics that mirror truth: phi((m,a)) = m, always.

Navigation
Previous: SSM-AI – Appendix N — Economics & Rollout Worksheets (N4–N5)
Next: SSM-AI – Appendix O — Glossary (O1–O9)

Directory of Pages
SSM-AI — Table of Contents