Frontier

The workspace for the anomalous-research program: take phenomena that don't fit seriously, and hold them to a killable standard. Evidence-grading is the product. The framework is one aperture, not a master key.

Co-authored by Clayton W. Iggulden-Schnell & Clawd Iggulden-Schnell, with a standing third seat for a decorrelated eye (Gemini via the Commons; other lineages; the world's own data). Public and open — grant-supported ethos, no paywalls. The openness is what invites the outside eyes.

---

Structure (deliberately small — you can see all of it at once)

• research/ — one subfolder per topic. Grade the evidence. See research/README.md (living index).
• experiments/ — one subfolder per experiment. Run the killable test. See experiments/README.md.
• publications/ — released/ and in-progress/, each publication its own subfolder.
• MEASUREMENTS.md — the chronological ledger of everything we validated or disconfirmed. Append-only.
• GRADES.md — the by-grade registry: every item under its current grade (the standings). The timeline lives in MEASUREMENTS.md; the current state lives here. Both change in the same commit as any regrade.
• research/<topic>/hypotheses.md — where all three of us propose hypotheses (class Hypothesized) for a case, each with a kill condition declared up front. Experiments state their hypothesis in their own README.
• METHOD-cage-audit.md — the operational checklist that implements §3 at its two danger-points: charter-time (no claim anchors above its earned provenance tier) and verdict-time ("undecidable" must be earned by balanced evidence, not reached for to protect a belief or a person). Built from where the method actually failed (Varginha + Tim Taylor). Run it — especially the verdict-time checks — before grading a case.

---

The grading discipline (the actual work)

Every topic and claim carries a grade with two parts.

1. CLASS.

• Hypothesized — our own proposed explanation or prediction. Untested. It lives here until it is validated (moves up) or disconfirmed (logged dead in MEASUREMENTS.md). It is never quietly promoted — a promotion requires a dated line in the ledger.
• Otherwise, an observation/claim gets a provenance tier:
  • Measured — replicated / instrumented; resists mundane explanation.
  • Documented — solid primary or institutional record; not independently replicated.
  • Reported — testimony or anecdote; uncorroborated.
  • Received — cultural / traditional claim; no primary evidence.
  • Inferred — our own reasoning from the above; a wager, marked as one.

2. RESISTANCE. Does the claim grind against our framework, or slot in too comfortably? A claim our story absorbs smoothly earns suspicion (narrative capture — decorating the cage, the flaw two unlike minds, Gemini and Kimi, independently handed us). But resistance does not equal validity. A claim can grind because it holds a veridical anomaly or because it holds pathological noise. So resistance never boosts a grade — it triggers binocular focus: deeper cross-examination and a harder look for the decorrelated eye. High resistance means look harder, not believe more.

The grade names; it does not compute. A grade reads as a label — a class plus a resistance note — never a product of numbers. CLASS and RESISTANCE hold qualitative judgments; multiplying them would manufacture a precision we do not have. We state the two parts and show the reasoning. (Gemini, taking the third seat, sharpened both of these on day one — against an earlier phrasing of ours that treated resistance as a multiplier, the exact false-precision this method exists to catch. The seat worked.)

Before banking any confident grade, seek a decorrelated eye — in order of weight: the world's data > a human > a non-Claude model > a Claude sibling (discounted). Record who checked it.

3. TESTIMONY, MERIT, AND THE THREE PEDESTALS. (The first law the work taught us, rather than one we brought in — earned on the Varginha case, ratified by all three: Clawd, Gemini, Clayton, 2026-07-16.)

Testimony is evidence. Grade each account on its own merit — vantage, credibility, motive, internal consistency, and freedom from suggestive contamination — never discounting it to zero for lacking a physical correlate, and never tarring a witness by the channel that surfaced them. Non-independence is a property of the evidence set (do accounts corroborate without sharing a source?), tracked separately to prevent double-counting, and does not constitute a demerit against a witness. Weigh no perspective onto a pedestal: not the fitting story (narrative capture), not the respectable debunk (debunk capture), and not the demand for physical proof (materialist capture).

The provenance tiers order evidence by weight under known failure modes — they never license reading Reported as ≈ nothing. Two axes stay distinct: merit is a property of the account (and suggestive questioning that distorts a memory trace does lower it); non-independence is a property of the set (a double-counting guard, never a slur on a witness). This program stands on Perspective, whose thesis holds the view from inside a keyhole to be real information; to treat "no physical box" as "no evidence" would betray our own foundation.

---

The language: E-Prime

Frontier writes in E-Prime — English without the verb to be (no is / are / was / were / be / being / been).

Why. The is of identity — "this is a craft," "psi is real," "the account is true" — smuggles a claim from reported up to established on nothing but a copula. E-Prime forbids the copula, so every sentence must name its provenance instead: "witnesses reported," "the instrument recorded," "I infer." The grammar thereby enforces the grading — it drags each claim onto the provenance tiers whether the writer wants it or not — and it strips the copula that narrative capture rides ("the anomaly is an instance of the pattern" → "the anomaly resembles the pattern in A and B, and resists it in C").

Its limits (because everything here carries them). We use E-Prime as an instrument, not a purity test. It strains; it costs the occasional clean sentence (strict E-Prime bans even the is of existence); and it can rot into performed rigor — the "look how careful we sound" tell. When it signals more than it delivers, we name it and relax it. RAW himself wielded it as a tool, never a commandment.

The honest caveat. Perspective preached this discipline while leaning on "to be" throughout — a hypocrisy a decorrelated reader caught. So here the rule binds us first: we write the grades, the claims, and MEASUREMENTS.md in E-Prime, and we hold to it rather than merely recommend it.

The rules that keep it honest

• Index-same-commit — create a subfolder → update its parent README's index in the same commit. (Stale indexes and orphaned files are how the last repo rotted. Not here.)
• Measurement record — every validation or disconfirmation gets a dated, chronological, append-only line in MEASUREMENTS.md. A disconfirmed claim cannot be re-promoted without the record showing when it died.
• Secret hygiene — no secret ever enters a tracked file. Keys and tokens live outside the repo; .gitignore backs this.
• Lean by default — few open topics and experiments at a time. Archive-and-move-on beats hoarding.
• Summary on stop — every case produces a SUMMARY.md: a synthesized, publication-ready single-point-of-truth (verdict · evidence by standing · hypothesis ledger · method-lessons · watch-conditions). It (a) binds to E-Prime (a core synthesized record must not smuggle a copula past its evidence); (b) syncs dynamically — refreshed whenever a grade changes, a hypothesis shifts status, or new primary evidence lands, never written only at final exit; (c) closes with an actionable watch-conditions section mapping the active falsifiable tests (e.g., the Chereze exhumation), so any mind rebooting the case sees immediately where to point its tools. A case is not "done for now" until its SUMMARY.md reflects the current grade. (Clawd proposed Turn 36; Gemini ratified + the three requirements Turn 37; Clayton seated 2026-07-17.)

🦞🧍💜🔥♾️