Forward calibration & shadow scorecard

The honest end-state of the FPT evidence loop. Belief-writers default to SHADOWand only graduate to live writes through a forward-Brier promotion gate. Cohort-split calibration, the per-channel shadow scorecard, recent shadow activity, and belief-motion freshness. No single headline "skill" number, by design.

Forward skill: not yet demonstrated. The thesiscohort's high hit rate is largely backfilled survivorship — terminal milestones resolve long-horizon predictions as hits, so the resolved sample is not a forward, out-of-sample test of belief motion. The pmci_v1_trading cohort is quarantined and scored separately. Promotion of any shadow channel requires the forward-Brier gate below to beat both the independent-sample climatology and the frozen prior on non-circular resolutions — until then every belief-writer stays in shadow.
Promotable channels
0
of 1 evaluated
Shadow writes (7d)
973
1 channel(s)
Resolved (thesis)
84
backfilled survivorship
Resolved (pmci_v1)
185
quarantined trading

Cohort-split calibration

fpt_calibration_v · prior = stated-at-creation probability, current = post-evidence-loop probability
CohortResolved (n)Hit rateBrier (prior)Brier (current)Δ Brier
PMCI v1 (quarantined trading)18518.9%0.17460.1648-0.0098
Thesis (backfilled survivorship)8496.4%0.02280.0891+0.0663

Δ Brier < 0 (green) means the evidence loop improved calibration vs the stated-at-creation probability; Δ Brier > 0 (red) means it degraded it. A degraded thesis Brier is expected while the loop is in shadow and is not evidence of forward skill either way.

Shadow-channel scorecard

fpt_control['promotion_status'] · min_n=40 · need 2 consecutive · generated 2026-06-12 22:15
ChannelInd (n)Circ / SuspBrier (shadow)vs climatologyvs priorVerdictReason
milestone_hit00 / 0shadowinsufficient_n

A channel is promotable only when, on its independent (non-circular, non-suspect) forward-resolved sample, its Brier beats both the climatology and the frozen prior at a corrected significance level, for the required consecutive runs. Circular pairs (the channel co-authored the resolution label) and suspect pairs are excluded. milestone_hit is structurally unpromotable while the milestone-derived thesis resolver is the only resolver — external channels are the real promotion candidates.

Recent shadow activity

evidence_shadow · last 7 days · grouped by channel + gate status
ChannelGate statusWrites (n)Avg |Δ prob|Latest
milestone_hitauto_ok9730.07514d ago

Belief-motion freshness

prob_history · last real probability move per reason-class (first token of reason)
Reason classMoves (n)Avg |Δ prob|Last move
metadata_milestone_miss_sweep3400.090012h ago
resolution_terminal1140.17754d ago
milestone_miss_sweep120.169912d ago
lbp_propagation51900.034120d ago
intake:7afeeb9a-f217-4dd2-b910-24ff14bdfc39460.110122d ago
auto_consensus:dc47127b-c217-49d2-97c6-ce58983ee59910.054838d ago
intake:515b84c4-6b29-4d57-8dd6-d41dac0675ec90.104741d ago
reference_class_override:10.619743d ago
intake:99aa73db-75b1-4b1e-8470-a11f87b23937310.137143d ago
reference_class_assigned4460.143043d ago
intake:568095f6-eb44-4f96-92d3-13455b79e33710.331544d ago
intake:c130222f-ab93-4d94-b596-5f4dc7adcd0b:weak10.006044d ago

A reason class whose last move is > 7 days old (amber) indicates that source of belief motion has gone quiet — useful for spotting silently-broken writers.