/* CalibrationLedger.jsx — the integrity instrument Every load-bearing coefficient in this document, classified into: CALIBRATED — sourced from a published reference or live data feed ASSERTED — reasoned assumption with explicit justification PROVISIONAL — placeholder named as a gap, awaiting calibration Naming the gap is the discipline. A model whose coefficients are all "CALIBRATED" is either lucky or lying; the proportion in each column is the honest measure of where the work still needs to happen. */ const LEDGER = [ /* ===== Inverse Design ===== */ { section: 'Inverse design', rows: [ { name: 'Coverage-line forward-map matrix (PER_LINE, 6×4)', status: 'ASSERTED', note: 'Direction (sign) is the load-bearing claim — primary retention is the most expensive and most cap-intensive line; cat reinsurance the cheapest with the strongest tail-transfer and regulatory credit. Magnitudes asserted from NZ commercial-lines treaty structure; full calibration pending a comparable reinsurance book.' }, { name: 'Cost-of-cover baseline (BASELINE_PREMIUM = 0.62)', status: 'ASSERTED', note: 'Cost units of an "all-primary plus quota-share" prior-year stack at a typical NZ commercial line. The optimiser maps premium reductions against this point. A real engagement re-anchors per client.' }, { name: 'Constitutional score formulas (5 × R(∂))', status: 'ASSERTED', note: 'Each score is a continuous monotone proxy for a regulatory floor. Coefficients chosen so that pathological allocations (all-primary, all-alt) breach floors, and well-spread allocations satisfy them. Per-coefficient calibration would require a regulator-supplied template; none exists publicly.' }, { name: 'Regulatory floors (rbnz=0.55, ifrs=0.50, xrb=0.45, cap=0.50, eth=0.55)', status: 'PROVISIONAL', note: 'Asserted thresholds for the demo. Real floors are firm-specific and live in private regulatory correspondence. The framework is the deliverable; the floors are the customer\'s.' }, { name: 'Loss weight λ = 1.5 on R(∂) penalty', status: 'ASSERTED', note: 'Sets how aggressively the optimiser pulls the configuration into the feasible region. Tuned so floors dominate at typical B*; lower values let target precision win at the cost of compliance margin.' }, { name: 'Optimiser learning rate / max iters / convergence threshold', status: 'CALIBRATED', note: 'Standard projected-gradient defaults with backtracking line search. Empirically: monotone non-increasing loss across all tested seeds; convergence in 30–60 iters at typical B*.' }, { name: 'Grade thresholds (A>0.85, B+>0.75, B>0.65, C>0.55, else D)', status: 'ASSERTED', note: 'Pedagogical bands for the visual grade. No statutory mapping; an actuary would replace with a firm-specific scoring rubric.' }, ], }, /* ===== Tempos ===== */ { section: 'Tempos', rows: [ { name: 'Event timeline (24 events × time t)', status: 'ASSERTED', note: 'Timing of each event roughly matches the public April 2026 sequencing of Hormuz-related supply-chain effects. Exact days are illustrative; trajectory shape is the load-bearing claim.' }, { name: 'Tempo gap = N(physical) / N(government) at scrub time', status: 'PROVISIONAL', note: 'A count-ratio stand-in for a real arrival-rate ratio. Patrick will notice the difference; the right replacement is λ_phys / λ_gov derived from the underlying timestamps. Future calibration target.' }, { name: 'Lane speed labels (days, hours, quarters, months)', status: 'ASSERTED', note: 'Order-of-magnitude descriptors of each tempo\'s characteristic time-constant. Defensible to a generalist; not fitted to a process model.' }, ], }, /* ===== PumpAbsorption ===== */ { section: 'Pump-price absorption', rows: [ { name: 'Crude-shock path: spike(amp=0.85, σ=0.18, μ=0.4) + tail(slope=0.25 from t=0.5)', status: 'ASSERTED', note: 'Stylised crude-spike profile consistent with major-strait disruption episodes. Not fit to any single historical event.' }, { name: 'Absorber magnitudes (hedge=0.55, cargo=0.18, credit=0.12, BI=0.08, buffer=0.07)', status: 'PROVISIONAL', note: 'Asserted to sum to 1.00 for illustrative balance under maximum activation. A real calibration would back out per-layer contribution from listed-entity hedge-disclosure regression and treaty pricing data — significant project on its own.' }, { name: 'EMA smoothing α = 0.18', status: 'ASSERTED', note: 'Effective lag ≈ 5 weeks, consistent with the rolling-tranche structure of NZ pump pricing. Documented in petrol-pricing-monitor reports.' }, { name: 'Additive absorption: total = Σ enabled mags', status: 'PROVISIONAL', note: 'Treats absorbers as independent; reality has cross-layer correlation (shocks that break a hedge counterparty also stress trade credit). The right model is a 5×5 layer correlation matrix; held back from this document deliberately to preserve scope.' }, ], }, /* ===== Pressures ===== */ { section: 'Pressures', rows: [ { name: 'Six structural pressures (XRB, IFRS 17, cyber, AVF, Marsden, cascade)', status: 'CALIBRATED', note: 'Each pressure traces to public regulatory or industry context (XRB CS 1/2/3; RBNZ IPSA; published cat-modelling literature; industry commentary on Marsden Pt). Selection reflects current Aotearoa market conditions.' }, { name: 'Initial (immediacy, magnitude) coordinates', status: 'ASSERTED', note: 'Asserted placement reflecting Axiom Intelligence\'s read of urgency and stakes. Patrick is invited to drag dots; the disagreement is itself useful data.' }, { name: 'Priority = immediacy × magnitude', status: 'ASSERTED', note: 'Simplest defensible aggregator. A more sophisticated treatment would weight by stakeholder perspective and time-decay; not required for this section\'s purpose.' }, ], }, /* ===== Cascade ===== */ { section: 'Cascade scenarios', rows: [ { name: 'Six scenarios × seven sectors weight matrix (42 cells)', status: 'ASSERTED', note: 'Each weight in [0,1] reflects qualitative consensus across NZ-domain readings. Hormuz, Wellington fault, AVF, MPI biosecurity, cloud-supply-chain cyber, Hawkes Bay — all current candidate-scenario peg points for Aotearoa cat modelling.' }, { name: 'Sector base-exposure values (banks=4.1, agri=2.6, infra=2.2, retail=1.8, fuel=1.4, transport=1.2, marae=0.9)', status: 'ASSERTED', note: 'Relative-magnitude order matches NZ insurance market structure (regional banks largest correlated exposure; iwi assets smallest in $-terms). Absolute scale chosen so totals sit in single-digit-billion-NZD range characteristic of large NZ events.' }, { name: 'Vendor coverage multiplier (0.55)', status: 'PROVISIONAL', note: 'Single coefficient stand-in for a 7-sector × 6-scenario vendor coverage matrix. The right replacement is per-cell calibration — each cell saying which sectors a vendor model handles well and which it misses for that scenario. Held back from this document deliberately.' }, { name: 'Correlation graph edge threshold (0.15)', status: 'ASSERTED', note: 'Visual threshold below which an edge is suppressed for legibility. Not a structural feature of the cascade.' }, ], }, /* ===== Offerings ===== */ { section: 'Offerings', rows: [ { name: 'Five offerings × five investability-axis scores (25 cells)', status: 'ASSERTED', note: 'Each score is Axiom Intelligence\'s own honest read of the offering against the axis. Patrick is invited to disagree; reasoning per cell is available on request and would migrate to per-cell explainers in a deeper revision.' }, { name: 'Default investor weights (warmth=0.30, capital=0.25, scale=0.20, regulatory=0.15, mission=0.10)', status: 'ASSERTED', note: 'A plausible weighting for a first-time NZ investor in this category. Patrick re-weights live; ranking re-sorts.' }, { name: 'Composite = Σ (score × weight)', status: 'CALIBRATED', note: 'Standard linear weighted sum. Honest, simple, defensible — no pretence of sophistication where none is needed.' }, ], }, ]; const STATUS_STYLE = { CALIBRATED: { bg: 'rgba(31, 138, 58, 0.14)', color: '#1f8a3a', border: '#1f8a3a' }, ASSERTED: { bg: 'rgba(196, 136, 31, 0.14)', color: '#a06a14', border: '#c4881f' }, PROVISIONAL: { bg: 'rgba(139, 26, 26, 0.12)', color: '#8b1a1a', border: '#8b1a1a' }, }; function CalibrationLedger() { // Tally — make the proportion visible const tally = { CALIBRATED: 0, ASSERTED: 0, PROVISIONAL: 0 }; let total = 0; for (const sec of LEDGER) for (const r of sec.rows) { tally[r.status]++; total++; } const pct = (n) => total === 0 ? 0 : Math.round(n / total * 100); return (
07 / 07 The integrity instrument

Calibration ledger.

Every load-bearing coefficient in this document, classified. A model whose coefficients are all sourced is either lucky or lying. The proportion in each column is the honest measure of where the work still needs to happen.

{/* Tally row */}
{['CALIBRATED', 'ASSERTED', 'PROVISIONAL'].map((s) => { const sty = STATUS_STYLE[s]; return (
{s}
{tally[s]}/{total}
{pct(tally[s])}% of coefficients
{s === 'CALIBRATED' && 'Sourced from a published reference or live data feed.'} {s === 'ASSERTED' && 'Reasoned assumption with explicit justification.'} {s === 'PROVISIONAL'&& 'Placeholder named as a gap, awaiting calibration.'}
); })}
{/* Frame */}
How to read this: the page above is a methodology demonstration, not a production system. CALIBRATED rows are points where the math touches public reality; ASSERTED rows are where we make a defensible call; PROVISIONAL rows are where we have named the gap rather than papered over it. A first engagement converts ASSERTED into CALIBRATED for the client's specific book; a longer one converts PROVISIONAL.
{/* Per-section ledger */}
{LEDGER.map((sec) => (
{sec.section}
{sec.rows.map((r, i) => { const sty = STATUS_STYLE[r.status]; return ( ); })}
{r.status} {r.name} {r.note}
))}
The discipline is to update this ledger every time a coefficient changes status. A first calibration engagement on a client book converts most of the ASSERTED rows in the Inverse Design section into CALIBRATED. A longer engagement does the same for the PROVISIONAL rows in PumpAbsorption and Cascade. The ledger is therefore the roadmap, not just an audit.
); } window.CalibrationLedger = CalibrationLedger;