ECSS Communications — Time Correlation (space-systems/ecss/e50-time-correlation)
Use when the task is the single obligation of ECSS-E-ST-50C clause 5.6.14.6 —
that the space link provide the capability by which the ground can tie onboard
time to its own scale — and the question is how good the correlation that
capability yields is, against the accuracy the mission states, and how long it
lasts.
Domain quick reference
- A correlation is a fit, not a number. One pair gives an offset that is
already stale when it is written down; the pair set gives an offset, a
drift, and a statement about how well each is known.
- The stated accuracy is what makes the correlation checkable. Without a
bound there is nothing to compare the residuals against, and without a
drift uncertainty there is no interval over which to compare it.
- Propagation delay is a bias every pair shares, which is exactly the
error a fit cannot remove. A reference timestamp taken at ground
reception is late by the one-way light time, and the fit obligingly
absorbs that into the offset as though it were real.
- Two pairs are the trap. They determine a straight line exactly, the
residuals come out at zero, and every quality metric derived from
those residuals reports a perfect correlation that is simply
unmeasured.
- Drift is the onboard oscillator's frequency error seen end to end, and
its standard error is what sets the validity horizon. A well-fitted
drift decays slowly; a poorly fitted one makes yesterday's correlation
useless today.
- The horizon is the deliverable. Operators do not act on a residual
spread; they act on how long they may go before correlating again.
Workflow
- Collect the pairs as onboard reading against the matching instant on
the ground reference scale, each one taken through the correlation
service the space link itself provides rather than reconstructed
afterwards, and reject a set that does not strictly advance in
onboard time.
- Remove the one-way light time from the reference side before fitting
anything, and record the correction applied so the bias is visible.
- Fit the offset between the scales against onboard time by least
squares, anchored at the first reading, yielding offset and drift.
- Take the residuals and their root mean square. This is the dispersion
the pairs themselves show, and it is compared against the bound.
- Take the standard error of the drift. With fewer than three pairs it
does not exist, and the assessment says so rather than reporting zero.
- Turn the bound into a validity horizon: the margin left after the
residual spread, divided by the rate the drift uncertainty consumes
it. Report it as the re-correlation interval.
- Call out a light-time correction larger than the accuracy bound —
it is the one error that would have passed every residual check.
Obligations
| Item |
Step |
| ECSS-E-ST-50C Rev.2 5.6.14.6a |
1 |
Pitfalls
- Fitting uncorrected reference timestamps. Every residual looks fine,
the drift is right, and the offset is wrong by the one-way light time
for the whole mission.
- Correlating from two pairs and reporting the residuals. They are zero
because two points define a line, not because the clocks agree.
- Reporting an offset with no drift. The offset is only valid at the
epoch it was fitted at, and a consumer who applies it later is wrong
by the drift times the interval, silently.
- Treating an unestimated drift uncertainty as zero. That turns an
unmeasured fit into one that never expires, which is the most
dangerous of the three possible answers.
- Quoting accuracy without a horizon. A correlation inside the bound
today says nothing about the pass it will actually be used on.
- Extrapolating far past the fitted span. The drift standard error is
estimated over the span the pairs cover, and beyond it the stated
error is an assumption rather than a measurement.
Behavior contract (gate 3)
Pair-set validation with strictly advancing onboard time, the light-time
correction as a pure offset bias, the least-squares offset and drift, residuals
and their spread, the drift standard error with its undefined two-pair case,
extrapolation, and the validity horizon in its finite, never-decaying,
already-exceeded and undetermined forms are exercised by the gate 3 contract
test:
scripts/test_e50_time_correlation.py against
scripts/e50_time_correlation_logic.py (stdlib unittest, offline).
Run:
python3 scripts/test_e50_time_correlation.py
Compliance
- ECSS standards are freely downloadable (ESA); cite the source and
paraphrase per standards-map.yaml.
- compliance: STANDARDS-REF, gated: false.
1---2name: e50-time-correlation3description: "Derive the correlation between onboard time and the reference time scale under ECSS-E-ST-50C clause 5.6.14.6, which asks that the two be correlated to a stated accuracy. Remove the one-way light time first, because it biases every pair alike and no fit can recover a common bias. Then fit offset and drift by least squares, take the residual spread and the standard error of the drift, and turn the stated accuracy into a validity horizon — the interval after which the correlation must be taken again. Use when sizing a time-correlation cadence or reviewing onboard clock accuracy. Trigger: ecss, e-st-50-communications, onboard-time-correlation, onboard-clock-drift-fit, time-correlation-validity-horizon, correlation-light-time-bias, recorrelation-interval-sizing."4license: Apache-2.05compliance: STANDARDS-REF6standards:7 - id: ecss8 reference-only: true9gated: false10domain: space-systems11pack: space-systems12compatibility: "agentskills.io SKILL.md; any SKILL.md host (Claude Code, Hermes, OpenClaw)"13clauses:14 - standard: ECSS-E-ST-50C Rev.215 clause: 5.6.14.616 items: [a]17 relation: implements18metadata:19 source: https://skillmd.com/skills/ashfordeou/e50-time-correlation20 domain: space-systems21 subdomain: ecss22 tags: [ecss, e-st-50-communications, e50-time-correlation, onboard-time-correlation, onboard-clock-drift-fit, time-correlation-validity-horizon, correlation-light-time-bias, recorrelation-interval-sizing]23 version: 0.1.024 author: Aero Agent Skills25---2627# ECSS Communications — Time Correlation (space-systems/ecss/e50-time-correlation)2829Use when the task is the single obligation of ECSS-E-ST-50C clause 5.6.14.6 —30that the space link provide the capability by which the ground can tie onboard31time to its own scale — and the question is how good the correlation that32capability yields is, against the accuracy the mission states, and how long it33lasts.3435## Domain quick reference3637- A correlation is a fit, not a number. One pair gives an offset that is38 already stale when it is written down; the pair set gives an offset, a39 drift, and a statement about how well each is known.40- The stated accuracy is what makes the correlation checkable. Without a41 bound there is nothing to compare the residuals against, and without a42 drift uncertainty there is no interval over which to compare it.43- Propagation delay is a bias every pair shares, which is exactly the44 error a fit cannot remove. A reference timestamp taken at ground45 reception is late by the one-way light time, and the fit obligingly46 absorbs that into the offset as though it were real.47- Two pairs are the trap. They determine a straight line exactly, the48 residuals come out at zero, and every quality metric derived from49 those residuals reports a perfect correlation that is simply50 unmeasured.51- Drift is the onboard oscillator's frequency error seen end to end, and52 its standard error is what sets the validity horizon. A well-fitted53 drift decays slowly; a poorly fitted one makes yesterday's correlation54 useless today.55- The horizon is the deliverable. Operators do not act on a residual56 spread; they act on how long they may go before correlating again.5758## Workflow59601. Collect the pairs as onboard reading against the matching instant on61 the ground reference scale, each one taken through the correlation62 service the space link itself provides rather than reconstructed63 afterwards, and reject a set that does not strictly advance in64 onboard time.652. Remove the one-way light time from the reference side before fitting66 anything, and record the correction applied so the bias is visible.673. Fit the offset between the scales against onboard time by least68 squares, anchored at the first reading, yielding offset and drift.694. Take the residuals and their root mean square. This is the dispersion70 the pairs themselves show, and it is compared against the bound.715. Take the standard error of the drift. With fewer than three pairs it72 does not exist, and the assessment says so rather than reporting zero.736. Turn the bound into a validity horizon: the margin left after the74 residual spread, divided by the rate the drift uncertainty consumes75 it. Report it as the re-correlation interval.767. Call out a light-time correction larger than the accuracy bound —77 it is the one error that would have passed every residual check.7879## Obligations8081| Item | Step |82|---|---|83| ECSS-E-ST-50C Rev.2 5.6.14.6a | 1 |8485## Pitfalls8687- Fitting uncorrected reference timestamps. Every residual looks fine,88 the drift is right, and the offset is wrong by the one-way light time89 for the whole mission.90- Correlating from two pairs and reporting the residuals. They are zero91 because two points define a line, not because the clocks agree.92- Reporting an offset with no drift. The offset is only valid at the93 epoch it was fitted at, and a consumer who applies it later is wrong94 by the drift times the interval, silently.95- Treating an unestimated drift uncertainty as zero. That turns an96 unmeasured fit into one that never expires, which is the most97 dangerous of the three possible answers.98- Quoting accuracy without a horizon. A correlation inside the bound99 today says nothing about the pass it will actually be used on.100- Extrapolating far past the fitted span. The drift standard error is101 estimated over the span the pairs cover, and beyond it the stated102 error is an assumption rather than a measurement.103104## Behavior contract (gate 3)105106Pair-set validation with strictly advancing onboard time, the light-time107correction as a pure offset bias, the least-squares offset and drift, residuals108and their spread, the drift standard error with its undefined two-pair case,109extrapolation, and the validity horizon in its finite, never-decaying,110already-exceeded and undetermined forms are exercised by the gate 3 contract111test:112scripts/test_e50_time_correlation.py against113scripts/e50_time_correlation_logic.py (stdlib unittest, offline).114Run:115python3 scripts/test_e50_time_correlation.py116117## Compliance118119- ECSS standards are freely downloadable (ESA); cite the source and120 paraphrase per standards-map.yaml.121- compliance: STANDARDS-REF, gated: false.