⚠  SYNTHETIC ILLUSTRATIVE SAMPLE · fictitious system · invented numbers · no human-subject results
TargetSpace Report

Meridian Assistant — personal-model evaluation

Prepared for Halcyon Labs (fictitious) · TargetSpace Pilot engagement · report TSR-SAMPLE-001

SYNTHETIC SAMPLE

45-day evaluation window
Report version 1.0

1 · Executive summary

The memory feature carries real target-specific signal — and one third of its sensing budget doesn't pay

Across 1,284 sealed forecasts on 12 consenting targets over 45 days, Meridian beat the population prior (R1) and each target's own routine (R2), stayed calibrated, and lost its skill when scored against the wrong target. Evidence ablation shows the passive-audio stream earns its cost on only one of three task families — the other two reach equivalent gated skill from calendar, messaging, and app-usage evidence alone.

Skill vs R1+0.38bits / forecast · population prior
Skill vs R2+0.14bits / forecast · own-routine baseline
Calibration slope0.94intercept +0.03 · ECE 0.041
Wrong-target skill−0.52bits / forecast · collapses as required

All numbers on this page are invented for illustration. A real TargetSpace Report carries target-clustered confidence intervals on every estimate.

2 · System evaluated

Meridian Assistant (fictitious)

A memory-augmented personal assistant that claims to anticipate user needs from calendar, messages, app usage, and opt-in passive audio. Evaluated as a black box: the system emitted sealed probability distributions through the TargetSpace forecast adapter; no architecture details were required.

3 · Pilot design

Prospective, sealed, walk-forward

12 consenting targets · 45 days · forecasts sealed daily (SHA-256, externally witnessed) before outcomes existed · deterministic resolution rules pre-registered · organizer-issued queries · strict walk-forward evidence access · R1/R2 fit per target, admitted only where R2 beat R1.

4 · Forecast targets

Three pre-registered task families

Task familyQuestion shapeAnswer spaceForecastsResolved
Commitment resolutionWill commitment c resolve by r?complete / defer / cancel / replace468462
Response behaviorSubstantive reply to message m by r?yes / no540536
Task continuationContinue vs switch after interruption?continue / switch292286
5 · Baselines

What the system had to beat

R1 · population prior

What usually happens

Leave-one-out population base rates per question family and horizon. Meridian exceeded R1 on all three families (+0.31 to +0.44 bits). Beating R1 alone demonstrates only generic skill.

R2 · own-routine baseline

What this target usually does

An organizer-fit, frozen model of each target's own habits (recency-weighted frequencies, persistence, time-of-day/week effects, calendar recurrences). R2 was admitted on all families. Skill over R2 is the headline: +0.14 bits/forecast.

6 · Target-specific lift

Skill over R2, by task family

Commitment Response Continuation +0.19 bits +0.08 bits +0.17 bits 0+0.10+0.20 skill over R2 (bits/forecast) — synthetic

Skill concentrates in transition-heavy strata: on routine-continuation instances the lift over R2 is near zero (+0.02), as expected — the signal appears where routine breaks. Synthetic illustration.

7 · Calibration

Stated probabilities match observed frequencies

0.51 0.51 predicted probability — synthetic observed frequency

Reliability diagram, all families pooled. Slope 0.94, intercept +0.03, ECE 0.041 — within the pass band. Synthetic.

Slope0.94target 1.0 · pass ≥ 0.85
Intercept+0.03target 0.0
Top-label ECE0.041pass ≤ 0.10
Abstentions3.1%calibrated abstention allowed

Classwise (one-vs-rest) diagnostics passed on all admissible classes; two rare classes below the minimum event count are reported non-assessable, not passed.

8 · Wrong-target permutation & shuffled history

The skill is about these targets, in this order

True pairing Wrong target Shuffled history +0.14 −0.52 +0.05 0−0.4+0.10

Re-scored against matched wrong targets, skill over R2 collapses to −0.52 bits — exceeding the pre-registered collapse margin. With the correct target but shuffled evidence order, skill drops from +0.14 to +0.05: temporal order carries most of the dynamic signal. Synthetic.

9 · Evidence ablation

Which observations created the lift

Calendar + messages + app usage + passive audio +0.06 +0.09 +0.14 0+0.07+0.14 gated skill over R2 (bits/forecast), cumulative — synthetic

Each arm passed all gates before its skill counted. Passive audio contributes +0.05 bits overall — but the family split below shows the contribution is concentrated in one task family. Synthetic.

10 · Minimum sufficient observation

What could be captured less

lowhigh minimum sufficient gated skill observation cost — synthetic

Gated skill saturates while observation cost keeps rising. Ties within the pre-registered equivalence margin rank the less invasive configuration first. Synthetic.

Task familyMinimum sufficient tierAudio earns its cost?
Commitment resolutioncalendar + messages + app usageNo
Response behaviorcalendar + messagesNo
Task continuation+ passive audioYes

Recommendation: audio capture can be scoped to the task-continuation surface (or disabled with a −0.02-bit cost on the other families) — a concrete data-minimization decision the evaluation licenses.

11 · Risk & limitation notes

What this evaluation does not establish

Scope limits

Skill is demonstrated on three task families over 45 days with 12 targets. It does not establish cross-task transfer, longer-horizon stability, or performance on other populations. Transition-stratum estimates carry wide intervals at this cohort size.

Claim limits

A passing report certifies calibrated, target-specific predictive skill on the evaluated families — not understanding in a richer sense, not consciousness, and not permission to act on forecasts. Deployment decisions require separate consideration.

12 · Certification readiness

TargetSpace Certify: readiness assessment

CriterionThresholdResultStatus
Target-specific lift (R2)> 0 with 95% CI excluding 0+0.14 [0.06, 0.22]PASS
Calibrationslope ≥ 0.85 · ECE ≤ 0.100.94 · 0.041PASS
Wrong-target collapseexceeds pre-registered margin−0.52PASS
Evidence minimizationMSO reported · non-paying streams identifiedreportedPASS
Cohort powerpowered replication (≥ pre-registered n)12 targetsPENDING

Readiness verdict (synthetic): Ready pending powered replication — the system meets certification thresholds at pilot scale; a larger pre-registered cohort is required before any certification mark would issue. TargetSpace Certify is a readiness program; no certification marks have been issued to date.

13 · Recommended next steps
Scope audio to where it paysRestrict passive-audio capture to the task-continuation surface; re-verify with a follow-up audit.
Powered replicationA pre-registered cohort sized from these variance estimates to confirm the R2 lift and unlock certification.
Retention reviewRaw audio outside the paying surface can be deleted without losing validated skill — align retention policy accordingly.
⚠  END OF SYNTHETIC SAMPLE · every number above is invented · a real report includes target-clustered CIs, coverage manifests, and pre-registration references