Skip to content

Hypothesis Computability Audit

Date: April 25, 2026 Scope: All 33 hypotheses articulated in this thread — the original 25-row H-battery (commits up through 52b385b) plus the 8 architectural-mode hypotheses (§10–§11.6.15, commits 942570638744a3).


Status legend

  • DONE — Computed and recorded; result is in results/phase1_hypotheses.csv or notebook.
  • CODE-READY — Algorithm defined, data available; can be computed today with existing code or a small extension (≤ 50 LOC).
  • SCOPED — Hypothesis is formal and falsifiable; computation requires moderate code (50–300 LOC) but no new data.
  • DATA-BLOCKED — Requires offline data we don't have (AMRP X-ray volumes, fragment photographs, paywalled papers).
  • ANALYTIC-RESOLVED — Worked through analytically; no computation needed.
  • ARCHITECTURAL — Open conjecture; not directly testable without committing to one specific implementation.

Phase 1 H-battery (25 hypotheses)

All landed and tested under the consolidated_tests.py battery. Final summary (DE422 + intermittent allowed): 17 PASS / 3 PARTIAL / 4 FAIL / 2 UNDETERMINED across 26 H-tags (E-H1c added in 52b385b).

A-series — Coprime addressing

ID Statement Result Computability
A-H1 Mechanism ratios are best rational approximations under tooth-count budget PARTIAL (top-3 CF: 15%; best-under-budget-500: 54%) DONE
A-H2 {7,17} planetary shared-prime choice is Pareto-optimal PARTIAL (on factor-reuse + proxy frontiers, NOT primary) DONE
A-H3 Prime spectrum non-random PARTIAL (small-prime overweight 1.15× null; sparse large primes 47/53/61/83/127/223 distinctive) DONE
A-H4 Rare large primes (47, 127, 223, 251) are forced by astronomy PASS (∞× inflation when removing each) DONE

B-series — Algebraic structure

ID Statement Result Computability
B-H1 Every cycle is element of ℂ[ℤ/D_LCMℤ] PASS (D_LCM = 27-digit composite with 16 distinct primes) DONE
B-H2 σ_day = roll(D, 1) is unit PASS (gcd(1, D) = 1) DONE
B-H3 HDC binding via coprime roll = gear composition PASS (13/13 dials at D=13440 round-trip exact) DONE

C-series — Bounds and aliasing

ID Statement Result Computability
C-H1 Zero intrinsic error correction PASS (40 gear pairs all bijective) DONE
C-H2 Spiral-dial wrap = chess §11.3.3 torus-clip PASS (Saros 223/4=55.75 m/turn; Metonic 235/5=47.0) DONE

D-series — T-breaking

ID Statement Result Computability
D-H1 Pin-and-slot is antisymmetric fiber PASS (
D-H2 Non-pin-and-slot dials are T-symmetric PASS (13/13 at residue 0 at epoch) DONE
D-H3 σ_day fails as unit on equant-encoded channel PASS (uniform std 0.0000° → equant std 0.0506° per day; ∞× ratio) DONE

E-series — Astronomical ground truth

ID Statement Result Computability
E-H1a Modern Saros syzygies (1999, 2017) within ±1 day PASS (3/3 anchors via DE421) DONE
E-H1b Almagest Hellenistic anchors (-382 to +125) within ±1 day AND ±2° FAIL (⅙; anchor JD data error in hellenistic_eclipses.py) DONE but FAILed due to data-curation, not encoder; DATA-BLOCKED for proper test (requires NASA Espenak catalog re-derivation of JDs)
E-H1c Sky-driven Saros chain: encoder anchored on DE422 syzygy reliably marks others within ±1 day PASS (backward_precision = 1.000) DONE
E-H2 Uniform Mars peak error ≥ 150° (uniform model insufficient) PASS (179.88° peak) DONE
E-H3 Hipparchus epicycle-only Mars peak ≤ 10° FAIL (51.48° peak; threshold was too optimistic — epicycle-only ≈ equant in peak error) DONE but threshold needs widening per §9.2 finding
E-H4 Ptolemy equant Mars peak in 30–50° band PASS (48.66° peak) DONE

F-series — Open exploration (intentionally non-PASS/FAIL)

ID Statement Result Computability
F-E1 Mechanism prime spectrum matches modern Residue-HDC? UNDETERMINED ARCHITECTURAL (no clean operationalisation; intentionally open)
F-E2 D_LCM where every cycle is single integer PASS (D_LCM exists; LCMState symbolic) DONE
F-E3 Which cycles are "failed" (>0.1% residual)? UNDETERMINED (3 of 13 fail) DONE descriptively

G-series — Manufacturing tolerance

ID Statement Result Computability
G-H1 Saros pointer p95 ≤ 2°/19yr under default σ FAIL (continuous: 13.2°) → PASS (intermittent: 0.000°). G-H1's status is REGIME-DEPENDENT — flip captured in §11.6.10.8. DONE (both regimes)
G-H2 Pin-and-slot tolerance ≤ 1.2× straight baseline PASS (ratio 1.00) DONE
G-H3 Rare-prime trains within ±15% of cross-train median per-mesh σ FAIL (¼; selection effect with mean tooth count) DONE

H-series — Historical cross-references

ID Statement Result Computability
H-H1 Antikythera spectrum vs Almagest indistinguishable (chi² p > 0.05) PASS (p = 0.32, Cramér's V = 0.103, top-5 Jaccard 0.67) DONE
H-H2 MUL.APIN top-3 prime overlap with Antikythera ≥ ⅔ Jaccard PASS (Jaccard 1.00 — perfect {2,3,5}) DONE

Architectural-mode hypotheses (§10–§11.6.15)

These are the new hypotheses about how the mechanism is operationally configured, articulated in the recent thread. None are in the H-battery yet because they require either (a) extending the gear DAG model with mode-state, or (b) archaeological evidence beyond computational reach.

§10 — Missing gears as tolerance compensators

Aspect Status
Hypothesis The ~39 missing gears include compensator elements that absorb manufacturing-tolerance drift
Probability estimate LOW (≤15%) under §11.6.10.8's intermittent G-H1 PASS (eliminates the compensator necessity)
Computability WEAKENED. Original premise (compensators are needed) is computationally refuted by §11.6.10.8's empirical G-H1 flip. Specific compensator topologies could still be enumerated and scored, but §10's necessity is gone.
Next compute step Re-examine §10 conditional on continuous-operation assumption being relaxed; revisit only if §11.6.10 (clutch) is independently refuted.

§11.6.10 — Crank-as-clutch (global clutch hypothesis)

Aspect Status
Hypothesis Inserting the crank key axially depresses a release element; mechanism only ticks during active operation
Probability estimate 30–50%
Computability PARTIALLY DONE. (a) Numerical: Track C (f193781) flips G-H1 PASS under intermittent operation — done. (b) Categorical: Track A (f193781) catalogues 4 candidate release elements with verdicts — done. © Archaeological: Track B (8395311) compiles literature dossier; falsification path identified but DATA-BLOCKED (requires AMRP X-ray re-examination of Fragment A around a1/b1).
Next compute step Already maximally computed within available data. Further progress requires offline archaeological work.

§11.6.11 — Reverse-cranking for drift cancellation

Aspect Status
Hypothesis Operator cranks backwards to cancel accumulated drift
Probability estimate Mechanism is valid; superseded by anchor recalibration as the better strategy
Computability ANALYTIC-RESOLVED. Worked through in §11.6.11.⅔/4: systematic + eccentricity errors cancel; backlash + random tooth-pitch noise + pin-and-slot anharmonicity do not. Anchor recalibration is strictly cheaper AND strictly more accurate. No further computation needed.
Next compute step None — analytically resolved.

§11.6.12 — Selective lock per-subsystem (G-H6 sketch)

Aspect Status
Hypothesis Per-cluster selective engagement: each subsystem has its own clutch, allowing operator to set one subsystem at a time
Probability estimate Composes naturally with G-H8; treat as part of the same architecture rather than a separate hypothesis
Computability SCOPED. Requires (a) defining a per-subsystem clutch model in gear_topology.py (~50 LOC); (b) for each subsystem, identifying the lock-attachment point; © scoring the periphery-rule consistency.
Next compute step Implement SubsystemClutch dataclass + clutch_attachment_evaluator() in gear_topology.py. Score each existing subsystem (lunar, Saros, Metonic, planetary) for "where would the lock attach?" + verdict. ~75 LOC. Then add G-H6 row to consolidated_tests.py.

§11.6.14 — Carrier-gear hypothesis (G-H7)

Aspect Status
Hypothesis Some "missing gears" are PORTABLE bronze carriers inserted by the operator through side ports to bridge gaps in the gear train
Probability estimate 10–25%
Computability SCOPED but speculative. (a) Side-port observation: through-shafts documented (consistent). (b) Carrier insertion consistency: doable — for each candidate carrier, check whether it can geometrically connect two existing surviving sub-axles. ~100 LOC. © Archaeological: requires unaccounted-for bronze in wreck site; DATA-BLOCKED.
Next compute step Define a CarrierGearElement class + carrier_insertion_geometry_check() that takes (sub-axle A, sub-axle B, carrier tooth count) and computes whether axial insertion is mechanically possible given the case dimensions. Then enumerate plausible carrier topologies for each of the 5 planets under Freeth 2021. ~150 LOC. Result: a Pareto frontier of carrier topologies vs DE422 fit. Doable today.

§11.6.15 — Setting-mode gears (G-H8) — the most testable

Aspect Status
Hypothesis Some gears are permanently mounted but kinematically active / functionally inert during normal operation, becoming load-bearing only during setting (clock-style). Specifically: synchronised-input differentials in the missing planetary plate.
Probability estimate 30–50% — highest of the architectural readings
Computability WELL-SCOPED. The headline prediction is concrete: "paired chains computing the same quantity for at least some planets, converging on a differential whose output reads zero under normal operation." This is directly testable.
Next compute step (1) For each planet, enumerate alternative rational approximations to the same target ratio using pareto_analysis.best_pq_constrained(). Each alternative chain is a candidate "second chain" in the synchronised-input pair. (2) Score the (chain_A, chain_B) pair: do the rational approximations agree (synchronised) under normal operation? Is the differential output identically zero or near-zero? (3) Result: per-planet table of "what would a synchronised-input differential pair look like for this planet?" Compare against Freeth 2021's single-chain reconstruction. ~120 LOC. Doable today.

Computability summary

Status Count Hypotheses
DONE 25 of 33 All 25 H-battery rows (A-H1..A-H4, B-H1..B-H3, C-H1..C-H2, D-H1..D-H3, E-H1a/b/c, E-H2..E-H4, F-E1..F-E3, G-H1..G-H3, H-H1..H-H2) — fully computed in results/phase1_hypotheses.csv
PARTIALLY DONE 1 §11.6.10 crank-as-clutch (numerical part done; archaeological part DATA-BLOCKED)
ANALYTIC-RESOLVED 1 §11.6.11 reverse-cranking (resolved without computation)
CODE-READY / SCOPED 3 §11.6.12 G-H6, §11.6.14 G-H7, §11.6.15 G-H8
WEAKENED 1 §10 (compensator necessity refuted by §11.6.10.8's intermittent G-H1 PASS)
DATA-BLOCKED (offline archaeology) aspects of §11.6.10, §11.6.14 Requires AMRP X-ray re-exam, fragment surveys

Priority 1 — G-H8 paired-chain enumeration (most testable, highest probability)

For each of the 5 planets in astronomical_cycles.CYCLES: 1. Take the target ratio (numerator / denominator). 2. Use pareto_analysis.best_pq_constrained() (Track 4) to enumerate alternative rational approximations within the same prime-constraint family. 3. For each alternative, score the synchronisation residual: |target_ratio − alt_ratio| over the design epoch. 4. Output: per-planet table of "candidate synchronised-input differential pairs" with synchronisation residual, total bronze cost (sum of teeth), and prime-spectrum overlap with Freeth 2021's single chain. 5. Falsification target: if no plausible alternative chain exists for any planet within Greek bronze-cutting tolerance, G-H8's specific prediction (paired chains in missing plate) is weakened. If multiple plausible alternatives exist, G-H8 is supported.

Estimated work: ~120 LOC, ~30 min compute. Doable today. Result: a clear empirical signature for whether the missing-gear inventory could plausibly include paired-chain differentials.

Priority 2 — G-H7 carrier insertion geometry

For each side-port through-shaft termination + each candidate carrier tooth count + each plausible target sub-axle: 1. Compute whether axial insertion is mechanically possible (case-depth constraint, sub-axle alignment). 2. Score against Freeth 2021's planetary-plate axle map. 3. Output: feasibility table.

Estimated work: ~150 LOC. Doable today. Result: shows whether G-H7's carrier reading is geometrically possible at all (could rule it out outright).

Priority 3 — G-H6 selective-lock attachment

For each surviving subsystem (lunar, Saros, Metonic, plus hypothetical planetary): 1. Define lock-attachment-point evaluator. 2. Score each subsystem's natural lock attachment under the periphery rule. 3. Output: per-subsystem lock-design table.

Estimated work: ~75 LOC. Quickest of the three. Result: completes the G-H6/G-H8 composition and adds a row to the consolidated battery.

Priority 4 — Update consolidated_tests.py with the three new G-tags

Once Priorities 1–3 land, add G-H6 / G-H7 / G-H8 evaluator functions to consolidated_tests.py so they emit rows to the canonical CSV. The H-battery would grow from 26 → 29 rows.


What's NOT computable from where we sit

These require offline archaeology / data acquisition:

  • §11.6.10 archaeological confirmation: AMRP X-ray volume re-examination of Fragment A around a1/b1 for non-gear bronze release elements. Held by National Archaeological Museum Athens.
  • E-H1b proper anchor JDs: NASA Espenak catalog re-derivation of the 6 Almagest-attributed eclipse JDs. Open data exists but requires manual cross-referencing with Toomer 1984.
  • Voulgaris 2024 paywalled paper: would clarify what the "two missing mechanical structures on b1" actually look like.
  • Keyway depth measurement: never published. Requires either a high-resolution Fragment A photograph or museum measurement.
  • Wreck-site bronze inventory: complete catalogue of all bronze artefacts recovered, with classification beyond "gear" / "case-piece" — would directly test G-H7 (carrier gears).

These are real research blockers that no amount of computation can resolve. They are listed in clutch_evidence_dossier.md as "Actionable Next Steps" with URLs and museum contacts.


Bottom line

~76% of the hypotheses are fully computed (25 of 33). Of the remaining 8: - 1 is partially computed with the rest archaeology-dependent (§11.6.10) - 1 is analytically resolved (§11.6.11) - 3 are code-ready and could be computed today (G-H6, G-H7, G-H8) - 1 is weakened to subsumed status (§10) - The remaining 2 (§11.6.13 SVG caveat, F-series) are open or descriptive

The single most valuable next compute step is G-H8 paired-chain enumeration (Priority 1). It's well-scoped, has clear pass/fail criteria, and would directly test the leading architectural hypothesis from this thread. Estimated ~120 LOC, ~30 minutes compute. Result either supports or weakens G-H8 with concrete numbers.

If pursued, the H-battery grows from 26 to 29 rows (G-H6, G-H7, G-H8 added) and the architectural-mode hypotheses move from "scoped" to "tested."