7 · Trustworthiness: Why the Index Can Be Scrutinized
7.1 · Scientific
The four elements of a measurement scale (Section 5.5) are met: a common scale (the unified contract (3)–(6) over a shared reference set); a stable center (the conservation law); domain-universal operations (every unit passes through the same composition tree); and comparability across time (settlement windows and a frozen anchor). The claims that can be checked against public data are stated as propositions (Section 6.9) and are partially reproducible: the conservation and boundedness invariants can in principle be recomputed from the documented FA methodology, the reference parameters, and the published snapshot (Appendix G). The system is designed to remain scrutable even where it is not end-to-end reproducible: the algebraic structure of the withheld components is public, and only their numerical calibration is reserved. The honest boundaries of the scientific claim are: the board is small (twelve Originators in the first assessment); confidence intervals from overlapping capability sources overlap substantially for mid-board units, and ties are reported transparently rather than broken arbitrarily; and the conservation claim on the base anchor is approximate (Proposition 6.11). None of these boundaries is hidden; Section 8.9 states them in full. The scientific claim also has a negative form that is as important as the positive one: the system is designed so that its invariants are falsifiable. If an independent recomputation of the conservation finds it violated, or of the audit inequality finds it broken, the system has failed a test it itself declared. This is the difference between a measurement system and a dashboard: a dashboard can be wrong in ways that are invisible by construction, while a measurement system with stated invariants can be wrong in ways that are publicly detectable. That is the sense in which “scrutable” is a load-bearing property.
7.2 · Fair
Fairness is designed as an institutional property, not a numerical one. Three mechanisms carry it. Asymmetric disclosure. The system opens what is hard to manipulate and closes what is easy to manipulate. FA inputs are slow stocks from third-party sources; its methodology and reference parameters are documented here. RI is a fast, short-horizon sentiment signal; its per-source weights, half-life tiers, baselines, and implementation are retained. The institutional reason is manipulation resistance: a rating or index that publishes exact short-horizon parameters invites arbitrage against the measure, a well-known failure mode in evaluation practice (Manheim and Garrabrant, 2018). This is the same posture as index and rating providers whose methods are public while exact parameters are reserved. Don't measure yourself with your own ruler. The FA backbone is third-party open data; self-collected public data is a supplement and is reconcilable against the third-party sources. Sources are replaceable: each source has an independent circuit breaker, multiple sources are weighted for the same quantity, and a source-swap migration procedure exists (Appendix D). The system is therefore structurally not in competition with its sources — it does not produce, own, or sell the underlying data (Section 13) — and this is the institutional guarantee that it has no incentive to distort a source's signal. The argument is structural, not hortatory: the decoupling of the method layer from the data layer (Section 13.1) means the system's own incentives point toward source fidelity, not source manipulation. Bounded negative correction. The real-harm correction applies only to officially confirmed substantive incidents (data breaches, critical security vulnerabilities, regulatory penalties), with sources attached and a correction channel open; negative attention never enters FA through any other path. The system declares itself an independent third-party evaluation, not an endorsement, and does not use the logos of the measured organizations, in respect of corporate-reputation rights. The boundedness of (a small fraction of FA, Appendix B) is a fairness design choice: the anchor is a measure of cumulative influence, and a single incident — however serious — is not allowed to erase a cumulative record. The RI negative tracks and the risk family carry the incident's current-period impact; the anchor carries a bounded, verified correction.
7.3 · Real-time
“Real-time” is an engineering caliber, and we separate the three layers in which it is realized: collection (sources polled at L1 cadence, on the order of 1–5 minutes), pricing (the composition engine at L2, bucketed), and display (the flagship line published at minute level, with second-level display capability). We distinguish the weak claim (updates quickly) from the strong claim we make (a market-semantics quote machine): a real-time sequence, periodic review-and-release of the slow anchor, and a suspend–resume release mechanism when measurement integrity is at risk. During closed-market windows the anchor is frozen and the live signal only decays; reopenings are announced and settled in batch. This is the research wording of what a market practitioner would call “quote / close / halt / resume”. The real-time claim is deliberately scoped. It is a claim about the pipeline (collection at L1 cadence, pricing at L2, display at minute level), not a claim that the underlying construct moves at that speed. The anchor is a slow variable and moves at settlement; the live signal moves at the pipeline cadence; the display shows both. Confusing “the display updates every minute” with “influence changes every minute” would be a category error, and the dual-value process (Section 6.8) is designed to make the distinction legible.
7.4 · Accessible: a property of the index layer
The paper's premise about accessibility is narrow and precise: accessibility is a property of the index output, not of the research system, and not of this document. The presentation-layer mapping (Definition 6.2) is the formal object that makes “accessible” checkable: it is a deterministic projection of the full indicator state onto a single number, a single line, and a consistent standard, with a declared fidelity boundary (the communication anchor of Section 6.8). The mapping exists so that a layperson can read the pattern without mastering the technology — see the pattern, not master the technology. Every simplification in the output is subject to the same “does not violate scientific fairness” check as the presentation layer itself. The V1 snapshot, an “understand at a glance” demonstration (Appendix G), exercises the output layer only.
7.5 · Reflexivity and manipulation resistance
We treat manipulation as a formal threat model rather than a public-relations concern. An attacker who controls a manipulable input can push the index toward a target value; the design's response is layered. (i) The conservation law (Propositions 6.9–6.11) bounds the aggregate effect: a burst that inflates one Originator's live signal must, by the centering, deflate the others, and the total pool is anchored by FA. (ii) The volatility budget bounds the single-burst magnitude; a per-Originator movement bound raises an audit alert without truncating the value. (iii) The anchor is hard to game because it is a weighted mean of independent third-party stocks whose replaceability is institutional (Section 7.2). (iv) The asymmetry of disclosure denies an attacker the exact short-horizon parameters that would make the live signal predictable. This is the engineering reading of Goodhart-style failure (Manheim and Garrabrant, 2018; Goodhart, 1984): the measure is designed so that its target is costly to hit. Reflexivity deserves a separate paragraph because it is the subtle case. A public influence index does not merely measure; it changes the measured, because the measured actors can observe it and adjust. The reflexive loop “discussion of the index feeds the index” is real — self-fulfilling dynamics of public ratings are documented in experimental cultural markets (Salganik and Watts, 2008) — and the system's response is deliberately conservative: the reflexive amplifier (a mechanism that would feed index-related attention back into the index) is disabled in V1, and index-related discussion enters the live feed only through the ordinary channels that any entity-word discussion uses — with the same source verification, the same polarity model, and the same budget. Whether this is sufficient is an empirical question, and the reflexivity of the system is itself listed as a study object (Section 12.3). The design's position is that reflexivity cannot be eliminated, only made legible and bounded.
Back to contents | ← 6 · Formal Framework: The Mathematical Shell | 8 · Verification: Segment-by-Segment Acceptance and Evidence →